These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
The presence of HDFS short-circuit read configuration properties in hbase-default.xml inadvertently causes short-circuit reads to not happen inside of RegionServers, despite short-circuit reads being enabled in hdfs-site.xml.
Ban the imports of guava VisiableForTesting, which means you should not use this annotation in HBase any more. For IA.Public and IA.LimitedPrivate classes, typically you should not expose any test related fields/methods there, and if you want to hide something, use IA.Private on the specific fields/methods. For IA.Private classes, if you want to expose something only for tests, use the RestrictedApi annotation from error prone, where it could cause a compilation error if someone break the rule in the future.
RsRpcServices APIs that can be accessed only through Admin rights:
setTableStateInMeta and fixMeta can be accessed only through Admin rights
Added IntegrationTestImportTsv.generatedHFileFolder configuration property to override the default location in IntegrationTestImportTsv. Useful for running the integration test when HDFS Transparent Encryption is enabled.
setRegionStateInMeta can be accessed only through Admin rights
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Three new commands are added to PE:
metaWrite, metaRandomRead and cleanMeta.
Usage example: hbase pe --rows=100000 metaWrite 1 hbase pe --nomapreduce --rows=100000 metaRandomRead 32 hbase pe --rows=100000 cleanMeta 1
metaWrite and cleanMeta should be run with only 1 thread and the same number of rows so all the rows inserted will be cleaned up properly.
metaRandomRead can be run with multiple threads. The rows option should set to within the range of rows inserted by metaWrite
`hbase master stop` should shutdown only master by default.
Help added to `hbase master stop`: To stop cluster, use `stop-hbase.sh` or `hbase master stop --shutDownCluster`
Help added to `stop-hbase.sh`: stop-hbase.sh can only be used for shutting down entire cluster. To shut down (HMaster|HRegionServer) use hbase-daemon.sh stop (master|regionserver)
After HBASE-25242, we can add Increment/Append operations to RowMutations and perform those operations atomically in a single row. HBASE-25242 includes an API change where the mutateRow() API returns a Result object to get the result of the Increment/Append operations.
Since the backward-compatible change we introduced in HBASE-25263, we use the more secure PBKDF2WithHmacSHA384 key generation algorithm (instead of PBKDF2WithHmacSHA1) to generate a secret key for HFile / WalFile encryption, when the user is defining a string encryption key in the hbase shell.
This change allows the REST and Thrift servers to handle the “doAs” parameter case-insensitively, which is deemed as correct per the “specification” provided by the Hadoop community.
A new option, CACHE_BLOCKS, was added to the `count` shell command which will force the data for a table to be loaded into the block cache. By default, the `count` command will not cache any blocks. This option can serve as a means to for a table's data to be loaded into block cache on demand. See the help message on the count shell command for usage details.
“Async WAL Replication” [1] was added by HBASE-11183 “Timeline Consistent region replicas - Phase 2 design” but only for user-space tables. This feature adds “Async WAL Replication” for the hbase:meta table. It also adds a client ‘LoadBalance’ mode that has reads go to replicas first and to the primary only on fail so as to shed read load from the primary to alleviate *hotspotting* on the hbase:meta Region.
Configuration is as it was for the user-space ‘Async WAL Replication’. See [2] and [3] for details on how to enable.
See parent issue, HBASE-18070, release notes for how to enable.
Adds a new metric where we collect the number of full region scan requests at the RPC layer. This will be collected under “name” : “Hadoop:service=HBase,name=RegionServer,sub=Server”
Since 2.4.0, deprecated all master carrys regions related methods(LoadBalancer,BaseLoadBalancer,ZNodeClearer) and configs(hbase.balancer.tablesOnMaster, hbase.balancer.tablesOnMaster.systemTablesOnly), they will be removed in 3.0.0.
The HBase shell now relies on JRuby 9.2. This is a new major version change for JRuby. The most significant change is Ruby compatibility changed from Ruby 2.3 to Ruby 2.5. For more detailed changes please see the JRuby release announcement for the start of the 9.2 series as well as the general release announcement page for updates since that version.
The runtime dependency versions present on the server side classpath for the Joni (now 2.1.31) and JCodings (now 1.0.55) libraries have also been updated to match those found in the JRuby version shipped with HBase. These version changes are maintenance releases and should be backwards compatible when updated in tandem.
This change adds options for disabling column family encryption and choosing hash algorithm for wrapped encryption keys. Changes are done such that defaults will keep the same behavior prior to this issue.
Prior to this change HBase always used the MD5 hash algorithm to store a hash for encryption keys. This hash is needed to verify the secret key of the subject. (e.g. making sure that the same secrey key is used during encrypted HFile read and write). The MD5 algorithm is considered weak, and can not be used in some (e.g. FIPS compliant) clusters. Having a configurable hash enables us to use newer and more secure hash algorithms like SHA-384 or SHA-512 (which are FIPS compliant).
The hash is set via the configuration option hbase.crypto.key.hash.algorithm. It should be set to a JDK MessageDigest algorithm like “MD5”, “SHA-256” or “SHA-384”. The default is “MD5” for backward compatibility.
Alternatively, clusters which rely on an encryption at rest mechanism outside of HBase (e.g. those offered by HDFS) and wish to ensure HBase's encryption at rest system is inactive can set hbase.crypto.enabled to false.
Fixes master procedure store migration issues going from 2.0.x to 2.2.x and/or 2.3.x. Also fixes failed heartbeat parse during rolling upgrade from 2.0.x. to 2.3.x.
Fixes so auto-migration of master procedure store works again going from 2.0.x => 2.2+. Also make it so heartbeats work when rolling upgrading from 2.0.x => 2.3+.
If hbase.regionserver.close.wait.abort is set to true, interrupt RPC handler threads holding the region close lock.
Until requests in progress can be aborted, wait on the region close lock for a configurable interval (specified by hbase.regionserver.close.wait.time.ms, default 60000 (1 minute)). If we have failed to acquire the close lock after this interval elapses, if allowed (also specified by hbase.regionserver.close.wait.abort), abort the regionserver.
We will attempt to interrupt any running handlers every hbase.regionserver.close.wait.interval.ms (default 10000 (10 seconds)) until either the close lock is acquired or we reach the maximum wait time.
This patch adds dynamic configuration support for the following configuration keys related to the normalizer:
Changed the max sleep time during meta and namespace regions availability check to be 60 sec. Previously there was no such cap
Introduces a new configuration, hbase.normalizer.throughput.max_bytes_per_sec, for specifying a limit on the throughput of actions executed by the normalizer. Note that while this configuration value is in bytes, the minimum honored valued is 1,000,000, or 1m. Supports values configured using the human-readable suffixes honored by Configuration.getLongBytes
The hbase-shell artifact now contains the ruby files that implement the hbase shell. There should be no downstream impact for users of the shell that rely on the hbase shell command.
Folks that wish to include the HBase ruby classes defined for the shell in their own JRuby scripts should add the hbase-shell.jar file to their classpath rather than add ${HBASE_HOME}/lib/ruby to their load paths.
The “force” flag to various unassign commands (java api, shell, etc) has been ignored since HBase 2. As of this change the methods that take it are now deprecated. Downstream users should stop passing/using this flag.
The Admin and AsyncAdmin Java APIs will have the deprecated version of the unassign method with a force flag removed in HBase 4. Callers can safely continue to use the deprecated API until then; the internal implementation just calls the new method.
The MasterObserver coprocessor API deprecates the preUnassign and postUnassign methods that include the force parameter and replaces them with versions that omit this parameter. The deprecated methods will be removed from the API in HBase 3. Until then downstream coprocessor implementations can safely continue to just implement the deprecated method if they wish; the replacement methods provide a default implementation that calls the deprecated method with force set to false.
Now you can change the region replication config for meta table by altering meta table. The old “hbase.meta.replica.count” is deprecated and will be removed in 4.0.0. But if it is set, we will still honor it, which means, when master restart, if we find out that the value of ‘hbase.meta.replica.count’ is different with the region replication config of meta table, we will schedule an alter table operation to change the region replication config to the value you configured for ‘hbase.meta.replica.count’.
Use shaded json and jersey in HBase. Ban the imports of unshaded json and jersey in code.
Increase timeout value for nightly jobs to 16 hours since the new build machines are dedicated to hbase project, so we are allowed to use it all the time.
WALPlayer can replay the content of recovered.edits directories.
Side-effect is that WAL filename timestamp is now factored when setting start/end times for WALInputFormat; i.e. wal.start.time and wal.end.time values on a job context. Previous we looked at wal.end.time only. Now we consider wal.start.time too. If a file has a name outside of wal.start.time<->wal.end.time, it'll be by-passed. This change-in-behavior will make it easier on operator crafting timestamp filters processing WALs.
Start time on the Master UI is now displayed using ISO8601 format instead of java Date#toString().
Now you do not need to disable a table before changing its ‘region replication’ property. If you are decreasing the replica count, the excess region replicas will be closed before reopening other replicas. If you are increasing the replica count, the new region replicas will be opened after reopening the existing replicas.
Change the java.io.tmpdir to project.build.directory in surefire-maven-plugin, to avoid writing std*deferred files to /tmp which may blow up the /tmp disk on our jenkins build node.
Set hbase.region.replica.replication.catalog.enabled to enable async WAL Replication for hbase:meta region replicas. Its off by default.
Defaults to the RegionReadReplicaEndpoint.class shipping edits -- set hbase.region.replica.catalog.replication to target a different endpoint implementation.
Adds a WALPlayer to MR Counter output:
org.apache.hadoop.hbase.mapreduce.WALPlayer$Counter CELLS\_READ=89574 CELLS\_WRITTEN=89572 DELETES=64 PUTS=5305 WALEDITS=4375
Without this fix there are situations in which locateRegionInMeta() on a client is not bound by a timeout. This happens because of a global lock whose acquisition was not under any lock scope. This affects client facing API calls that rely on this method to locate a table region in meta. This fix brings the lock acquisition under the scope of “hbase.client.meta.operation.timeout” and that guarantees a bounded wait time.
Adds a new configuration parameter “hbase.replication.peer.base.config” which accepts a semi-colon separated key=CSV pairs (example: k1=v1;k2=v2_1,v3...). When this configuration is set on the server side, these kv pairs are added to every peer configuration if not already set. Peer specific configuration overrides have precedence over the above default configuration. This is useful in cases when some configuration has to be set for all the peers by default and one does not want to add to every peer definition.
Expose Hadoop hedgedReadOpsInCurThread metric to HBase. This metric counts the number of times the hedged reads service executor rejected a read task, falling back to the current thread. This will help determine the proper size of the thread pool (dfs.client.hedged.read.threadpool.size).
HBASE-24776 added the following command line parameters to hbtop: | Argument | Description | |---|---| | -n,--numberOfIterations <arg> | The number of iterations | | -O,--outputFieldNames | Print each of the available field names on a separate line, then quit | | -f,--fields <arg> | Show only the given fields. Specify comma separated fields to show multiple fields | | -s,--sortField <arg> | The initial sort field. You can prepend a `+' or `-' to the field name to also override the sort direction. A leading `+' will force sorting high to low, whereas a `-' will ensure a low to high ordering | | -i,--filters <arg> | The initial filters. Specify comma separated filters to set multiple filters | | -b,--batchMode | Starts hbtop in Batch mode, which could be useful for sending output from hbtop to other programs or to a file. In this mode, hbtop will not accept input and runs until the iterations limit you‘ve set with the `-n’ command-line option or until killed |
Summary of the change of HBASE-24602:
// Perform a CheckAndIncrement operation CheckAndMutateResult checkAndMutateResult = table.checkAndMutate(checkAndMutate);
// Get whether or not the CheckAndIncrement operation is successful boolean success = checkAndMutateResult.isSuccess();
// Get the result of the increment operation Result result = checkAndMutateResult.getResult(); ```
Adds option for the flush command to flush all stores from the specified column family only, among all regions of the given table (stores from other column families on this table would not get flushed).
We add a method getSyncedLength in WALProvider.WriterBase interface for WALFileLengthProvider used for replication, considering the case if we use AsyncFSWAL,we write to 3 DNs concurrently,according to the visibility guarantee of HDFS, the data will be available immediately when arriving at DN since all the DNs will be considered as the last one in pipeline.This means replication may read uncommitted data and replicate it to the remote cluster and cause data inconsistency.The method WriterBase#getLength may return length which just in hdfs client buffer and not successfully synced to HDFS, so we use this method WriterBase#getSyncedLength to return the length successfully synced to HDFS and replication thread could only read writing WAL file limited by this length. see also HBASE-14004 and this document for more details: https://docs.google.com/document/d/11AyWtGhItQs6vsLRIx32PwTxmBY3libXwGXI25obVEY/edit#
Before this patch, replication may read uncommitted data and replicate it to the slave cluster and cause data inconsistency between master and slave cluster, we could use FSHLog instead of AsyncFSWAL to reduce probability of inconsistency without this patch applied.
New metrics are exposed, on the global source, for replication which indicate the “WAL entry buffer” that was introduced in HBASE-15995. When this usage reaches the limit, that RegionServer will cease to read more data for the sake of trying to replicate it. This usage (and limit) is local to each RegionServer is shared across all peers being handled by that RegionServer.
This adds an extra “flush” command option that allows for specifying an individual family to have its store flushed.
Usage: flush ‘REGIONNAME’,‘FAMILYNAME’ flush ‘ENCODED_REGIONNAME’,‘FAMILYNAME’
Users of HBaseTestingUtility can now safely call the getConnection method from multiple threads.
As a consequence of refactoring to improve the thread safety of the HBase testing classes, the protected conf member of the HBaseCommonTestingUtility class has been marked final. Downstream users who extend from the class hierarchy rooted at this class will need to pass the Configuration instance they want used to their super constructor rather than overwriting the instance variable.
Disables per-user metrics. They were enabled by default for the first time in hbase-2.3.0 but they need some work before they can be on all the time (See HBASE-15519)
Improve the layout of column family from vertical to horizontal in table UI.
In shell, all HBase constants and commands have been moved out of the top-level and into an IRB Workspace. Piped stdin and scripts passed by name to the shell will be evaluated within this workspace. If you absolutely need the top-level definitions, use the new compatibility flag, ie. hbase shell --top-level-defs or hbase shell --top-level-defs script2run.rb.
Enables procedure-based distributed WAL splitting as default (HBASE-20610). To use ‘classic’ zk-coordinated splitting instead, set ‘hbase.split.wal.zk.coordinated’ to ‘true’.
Flips default for ‘HBASE-23994 Add WebUI to Canary’ The UI defaulted to on at port 16050. This JIRA changes it so new UI is off by default.
To enable the UI, set property ‘hbase.canary.info.port’ to the port you want the UI to use.
HBASE-24650 introduced CheckAndMutateResult class and changed the return type of checkAndMutate methods to this class in order to support CheckAndMutate with Increment/Append. CheckAndMutateResult class has two fields, one is *success* that indicates whether the operation is successful or not, and the other one is *result* that's the result of the operation and is used for CheckAndMutate with Increment/Append.
The new APIs for the Table interface: ``` /** * checkAndMutate that atomically checks if a row matches the specified condition. If it does, * it performs the specified action. * * @param checkAndMutate The CheckAndMutate object. * @return A CheckAndMutateResult object that represents the result for the CheckAndMutate. * @throws IOException if a remote or network exception occurs. */ default CheckAndMutateResult checkAndMutate(CheckAndMutate checkAndMutate) throws IOException { return checkAndMutate(Collections.singletonList(checkAndMutate)).get(0); }
/** * Batch version of checkAndMutate. The specified CheckAndMutates are batched only in the sense * that they are sent to a RS in one RPC, but each CheckAndMutate operation is still executed * atomically (and thus, each may fail independently of others). * * @param checkAndMutates The list of CheckAndMutate. * @return A list of CheckAndMutateResult objects that represents the result for each * CheckAndMutate. * @throws IOException if a remote or network exception occurs. */ default List<CheckAndMutateResult> checkAndMutate(List<CheckAndMutate> checkAndMutates) throws IOException { throw new NotImplementedException(“Add an implementation!”); } {code}
The new APIs for the AsyncTable interface: {code} /** * checkAndMutate that atomically checks if a row matches the specified condition. If it does, * it performs the specified action. * * @param checkAndMutate The CheckAndMutate object. * @return A {@link CompletableFuture}s that represent the result for the CheckAndMutate. */ CompletableFuture<CheckAndMutateResult> checkAndMutate(CheckAndMutate checkAndMutate);
/** * Batch version of checkAndMutate. The specified CheckAndMutates are batched only in the sense * that they are sent to a RS in one RPC, but each CheckAndMutate operation is still executed * atomically (and thus, each may fail independently of others). * * @param checkAndMutates The list of CheckAndMutate. * @return A list of {@link CompletableFuture}s that represent the result for each * CheckAndMutate. */ List<CompletableFuture<CheckAndMutateResult>> checkAndMutate( List<CheckAndMutate> checkAndMutates);
/** * A simple version of batch checkAndMutate. It will fail if there are any failures. * * @param checkAndMutates The list of rows to apply. * @return A {@link CompletableFuture} that wrapper the result list. */ default CompletableFuture<List<CheckAndMutateResult>> checkAndMutateAll( List<CheckAndMutate> checkAndMutates) { return allOf(checkAndMutate(checkAndMutates)); } ```
Add excludefile and designatedfile options to graceful_stop.sh.
Designated file with <hostname:port> per line as unload targets.
Exclude file should have <hostname:port> per line. We do not unload regions to hostnames given in exclude file.
Here is a simple example using graceful_stop.sh with designatedfile option: ./bin/graceful_stop.sh --maxthreads 4 --designatedfile /path/designatedfile hostname The usage of the excludefile option is the same as the above.
Add a new option “designatedfile” in RegionMover.
If designated file is present with some contents, we will unload regions to hostnames provided in designated file.
Designated file should have ‘host:port’ per line.
Enhance DateTieredCompaction to support HDFS storage policy within one class family.
To turn on Date Tiered Compaction (It is not recommended to turn on for the whole cluster because that will put meta table on it too and random get on meta table will be impacted): hbase.hstore.compaction.compaction.policy=org.apache.hadoop.hbase.regionserver.compactions.DateTieredCompactionPolicy
hbase.hstore.compaction.date.tiered.max.storefile.age.millis: Files with max-timestamp smaller than this will no longer be compacted.Default at Long.MAX_VALUE. hbase.hstore.compaction.date.tiered.base.window.millis: base window size in milliseconds. Default at 6 hours. hbase.hstore.compaction.date.tiered.windows.per.tier: number of windows per tier. Default at 4. hbase.hstore.compaction.date.tiered.incoming.window.min: minimal number of files to compact in the incoming window. Set it to expected number of files in the window to avoid wasteful compaction. Default at 6.
hbase.hstore.compaction.date.tiered.storage.policy.enable=true hbase.hstore.compaction.date.tiered.hot.window.age.millis=3600000 hbase.hstore.compaction.date.tiered.hot.window.storage.policy=ALL_SSD hbase.hstore.compaction.date.tiered.warm.window.age.millis=20600000 hbase.hstore.compaction.date.tiered.warm.window.storage.policy=ONE_SSD hbase.hstore.compaction.date.tiered.cold.window.storage.policy=HOT
hbase.wal.storage.policy=ALL_SSD create ‘table’,{NAME=>‘f1’,CONFIGURATION=>{‘hbase.hstore.block.storage.policy’=>‘ALL_SSD’}}
hbase.hstore.compaction.date.tiered.storage.policy.enable=false
Add a canSplit method to RegionSplitPolicy to determine whether we can split a region. Usually it is not related to RegionSplitPolicy so in the default implementation, it will test whether region is available and does not have reference file, but in DisabledRegionSplitPolicy, we will always return false.
Change the flush level from region to store when there are too many wals, benefit from this we can reduce unnessary flush tasks and small hfiles.
Add a metric to show the locality of ssd in table.jsp, and move the locality related metrics to a new tab named localities.
HBASE-8458 introduced CheckAndMutate class that's used to perform CheckAndMutate operations. Use the builder class to instantiate a CheckAndMutate object. This builder class is fluent style APIs, the code are like: ``` // A CheckAndMutate operation where do the specified action if the column (specified by the family and the qualifier) of the row equals to the specified value CheckAndMutate checkAndMutate = CheckAndMutate.newBuilder(row) .ifEquals(family, qualifier, value) .build(put);
// A CheckAndMutate operation where do the specified action if the column (specified by the // family and the qualifier) of the row doesn't exist CheckAndMutate checkAndMutate = CheckAndMutate.newBuilder(row) .ifNotExists(family, qualifier) .build(put);
// A CheckAndMutate operation where do the specified action if the row matches the filter CheckAndMutate checkAndMutate = CheckAndMutate.newBuilder(row) .ifMatches(filter) .build(delete); ```
And This added new checkAndMutate APIs to the Table and AsyncTable interfaces, and deprecated the old checkAndMutate APIs. The example code for the new APIs are as follows: ``` Table table = ...;
CheckAndMutate checkAndMutate = ...;
// Perform the checkAndMutate operation boolean success = table.checkAndMutate(checkAndMutate);
CheckAndMutate checkAndMutate1 = ...; CheckAndMutate checkAndMutate2 = ...;
// Batch version List<Boolean> successList = table.checkAndMutate(Arrays.asList(checkAndMutate1, checkAndMutate2)); ```
This also has Protocol Buffers level changes. Old clients without this patch will work against new servers with this patch. However, new clients will break against old servers without this patch for checkAndMutate with RM and mutateRow. So, for rolling upgrade, we will need to upgrade servers first, and then roll out the new clients.
Move all the meta initialization code in MasterFileSystem and HRegionServer to InitMetaProcedure. Add a new step for InitMetaProcedure called INIT_META_WRITE_FS_LAYOUT to place the moved code.
This is an incompatible change, but should not have much impact. InitMetaProcedure will only be executed once when bootstraping a fresh new cluster, so typically this will not effect rolling upgrading. And even if you hit this problem, as long as InitMetaProcedure has not been finished, we can make sure that there is no user data in the cluster, you can just clean up the cluster and try again. There will be no data loss.
Changed master, branch-2, and branch-2.1 to twice a day. Left branch-2.3, branch-2.2, and branch-1 at every 4 hours. Changed branch-1.4 and branch-1.3 to @daily (1.3 was running every hour).
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Fixes a couple of bugs in ZooKeeper interaction. Firstly, zk sync() call that is used to sync the lagging followers with leader so that the client sees a consistent snapshot state was actually asynchronous under the hood. We make it synchronous for correctness. Second, zookeeper events are now processed in a separate thread rather than doing it in the thread context of zookeeper client connection. This decoupling frees up client connection quickly and avoids deadlocks.
Update our package version numbers throughout the Dockerfiles to be pinned to their epic:upstream-version components only. Previously we'd specify the full debian package version number, including the debian-revision. This lead to instability as debian packaging details changed. See also man deb-version
Adds a new metric where we collect the number of read requests (tracked per row) whether the row was fetched completely from memstore or it was pulled from files and memstore. The metric is now collected under the mbean for Tables and under the mbean for regions. Under table mbean ie.- ‘name": "Hadoop:service=HBase,name=RegionServer,sub=Tables’ The new metrics will be listed as {code} “Namespace_default_table_t3_columnfamily_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_columnfamily_f1_metric_mixedRowReadsCount”: 1, {code} Where the format is Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_mixedRowReadsCount {code}
The same one under the region ie. “name”: “Hadoop:service=HBase,name=RegionServer,sub=Regions”, comes as {code} “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_mixedRowReadsCount”: 1, {code} where Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_mixedRowReadsCount This is also an aggregate against every store the number of reads that happened purely from the memstore or it was a mixed read that happened from memstore and file.
This adds [-h|-help] options to rowcounter. Passing either -h or -help will print rowcounter guide as below:
$hbase rowcounter -h
usage: hbase rowcounter <tablename> [options] [<column1> <column2>...] Options: --starttime=<arg> starting time filter to start counting rows from. --endtime=<arg> end time filter limit, to only count rows up to this timestamp. --range=<arg> [startKey],[endKey][;[startKey],[endKey]...]] --expectedCount=<arg> expected number of rows to be count. For performance, consider the following configuration properties: -Dhbase.client.scanner.caching=100 -Dmapreduce.map.speculative=false
CI coverage has been extended to include Hadoop 3.2.x for HBase 2.2+.
Adds being able to edit hbase:meta table schema. For example,
hbase(main):006:0> alter ‘hbase:meta’, {NAME => ‘info’, DATA_BLOCK_ENCODING => ‘ROW_INDEX_V1’} Updating all regions with the new schema... All regions updated. Done. Took 1.2138 seconds
You can even add columnfamilies. Howevert, you cannot delete any of the core hbase:meta column families such as ‘info’ and ‘table’.
This ticket summarizes significant improvements and expansion to the metrics surface area. Interested users should review the individual sub-tasks.
Adds backoff in ServerCrashProcedure wait on WAL split to complete if large backlog of files to split (Its possible to avoid SCP blocking, waiting on WALs to split if you use procedure-based splitting -- set ‘hbase.split.wal.zk.coordinated’ to false to enable procedure based wal splitting.)
Notice this has changed log level for mismatching row keys, originally those were being logged at INFO level, now it's logged at DEBUG level. This is consistent with the logging of mismatching cells. Also, for missing row keys, it now logs row key values in human readable format, making it more meaningful for operators troubleshooting mismatches.
Introduce a new config hbase.replication.drop.on.deleted.columnfamily, default is false. When config to true, the replication will drop the edits for columnfamily that has been deleted from the replication source and target.
This change extends the Normalizer with a handful of new configurations. The configuration points supported are:
hbase.normalizer.split.enabled Whether to split a region as part of normalization. Default: true.hbase.normalizer.merge.enabled Whether to merge a region as part of normalization. Default true.hbase.normalizer.min.region.count The minimum number of regions in a table to consider it for merge normalization. Default: 3.hbase.normalizer.merge.min_region_age.days The minimum age for a region to be considered for a merge, in days. Default: 3.hbase.normalizer.merge.min_region_size.mb The minimum size for a region to be considered for a merge, in whole MBs. Default: 1.Add a hbase-logging module, put the log4j related code in this module only so other modules do not need to depend on log4j at compile scope. See the comments of Log4jUtils and InternalLog4jUtils for more details.
Add a log4j.properties to the test jar of hbase-logging module, so for other sub modules we just need to depend on the test jar of hbase-logging module at test scope to output the log to console, without placing a log4j.properties in the test resources as they all (almost) have the same content. And this test module will not be included in the assembly tarball so it will not mess up the binary distribution.
Ban direct commons-logging dependency, and ban commons-logging and log4j imports in non-test code, to avoid mess up the downstream users logging framework. In hbase-logging module we do need to use log4j classes and the trick is to use full class name.
Add jcl-over-slf4j and jul-to-slf4j dependencies, as some of our dependencies use jcl or jul as logging framework, we should also redirect their log message to slf4j.
Added new metric to differentiate sink startup time from last OP applied time.
Original behaviour was to always set startup time to TimestampsOfLastAppliedOp, and always show it on “status ‘replication’” command, regardless if the sink ever applied any OP.
This was confusing, specially for scenarios where cluster was just acting as source, the output could lead to wrong interpretations about sink not applying edits or replication being stuck.
With the new metric, we now compare the two metrics values, assuming that if both are the same, there's never been any OP shipped to the given sink, so output would reflect it more clearly, to something as for example:
SINK: TimeStampStarted=Thu Dec 06 23:59:47 GMT 2018, Waiting for OPs...
HBase ships ZooKeeper 3.5.x. Was the EOL'd 3.4.x. 3.5.x client can talk to 3.4.x ensemble.
The ZooKeeper project has built a FAQ that documents known issues and work-arounds when upgrading existing deployments.
Add backoff. Avoid retrying every 100ms.
Runs ‘catalogjanitor_run’ and ‘hbck_chore_run’ inline with the loading of the ‘HBCK Report’ page.
Pass ‘?cache=true’ to skip inline invocation of ‘catalogjanitor_run’ and ‘hbck_chore_run’ drawing the page.
Introduced a general ‘local region’ at master side to store the procedure data, etc.
The hfile of this region will be stored on the root fs while the wal will be stored on the wal fs. This issue supercedes part of the code for HBASE-23326, as now we store the data in ‘MasterData’ directory instead of ‘MasterProcs’.
The old hfiles will be moved to the global hfile archived directory with the suffix $-masterlocalhfile-$. The wal files will be moved to the global old wal directory with the suffix $masterlocalwal$. The TimeToLiveMasterLocalStoreHFileCleaner and TimeToLiveMasterLocalStoreWALCleaner are configured by default for cleaning the old hfiles and wal files, and the default TTLs are both 7 days.
Relocate test-only REST RemoteHTable and RemoteAdmin from src/ to test/. And mark them as InterfaceAudience.Private.
Config key: hbase.regionserver.slowlog.systable.enabled Default value: false
This config can be enabled if hbase.regionserver.slowlog.buffer.enabled is already enabled. While hbase.regionserver.slowlog.buffer.enabled ensures that any slow/large RPC logs with complete details are written to ring buffer available at each RegionServer, hbase.regionserver.slowlog.systable.enabled would ensure that all such logs are also persisted in new system table hbase:slowlog. Operator can scan hbase:slowlog with filters to retrieve specific attribute matching records and this table would be useful to capture historical performance of slowness of RPC calls with detailed analysis.
hbase:slowlog consists of single ColumnFamily info. info consists of multiple qualifiers similar to the attributes available to query as part of Admin API: get_slowlog_responses.
One example of a row from hbase:slowlog scan result (Attached a sample screenshot in the Jira) :
\x024\xC1\x06X\x81\xF6\xEC column=info:call_details, timestamp=2020-05-16T14:59:58.764Z, value=Scan(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$ScanRequest)
\x024\xC1\x06X\x81\xF6\xEC column=info:client_address, timestamp=2020-05-16T14:59:58.764Z, value=172.20.10.2:57348
\x024\xC1\x06X\x81\xF6\xEC column=info:method_name, timestamp=2020-05-16T14:59:58.764Z, value=Scan
\x024\xC1\x06X\x81\xF6\xEC column=info:param, timestamp=2020-05-16T14:59:58.764Z, value=region { type: REGION_NAME value: “cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.” } scan { a ttribute { name: “_isolationlevel_” value: “\x5C000” } start_row: “cccccccc” time_range { from: 0 to: 9223372036854775807 } max_versions: 1 cache_blocks: true max_result_size: 2 097152 caching: 2147483647 include_stop_row: false } number_of_rows: 2147483647 close_scanner: false client_handles_partials: true client_handles_heartbeats: true track_scan_met rics: false
\x024\xC1\x06X\x81\xF6\xEC column=info:processing_time, timestamp=2020-05-16T14:59:58.764Z, value=24
\x024\xC1\x06X\x81\xF6\xEC column=info:queue_time, timestamp=2020-05-16T14:59:58.764Z, value=0
\x024\xC1\x06X\x81\xF6\xEC column=info:region_name, timestamp=2020-05-16T14:59:58.764Z, value=cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.
\x024\xC1\x06X\x81\xF6\xEC column=info:response_size, timestamp=2020-05-16T14:59:58.764Z, value=211227
\x024\xC1\x06X\x81\xF6\xEC column=info:server_class, timestamp=2020-05-16T14:59:58.764Z, value=HRegionServer
\x024\xC1\x06X\x81\xF6\xEC column=info:start_time, timestamp=2020-05-16T14:59:58.764Z, value=1589640743932
\x024\xC1\x06X\x81\xF6\xEC column=info:type, timestamp=2020-05-16T14:59:58.764Z, value=ALL
\x024\xC1\x06X\x81\xF6\xEC column=info:username, timestamp=2020-05-16T14:59:58.764Z, value=vjasani
HBASE-24271 makes changes the the default conf/hbase-site.xml such that bin/hbase will run directly out of the binary tarball or a compiled source tree without any configuration modifications vs. Hadoop 2.8+. This changes our long-standing history of shipping no configured values in conf/hbase-site.xml, so existing processes that assume this file is empty of configuration properties may require attention.
Use Slf4jRequestLog instead of the log4j HttpRequestLogAppender in HttpServer.
The request log is disabled by default in conf/log4j.properties by the following lines:
log4j.category.http.requests=INFO,NullAppender log4j.additivity.http.requests=false
Change the ‘NullAppender’ to what ever you want if you want to enable request log.
Notice that, the logger name for master status http server is ‘http.requests.master’, and for region server it is ‘http.requests.regionserver’
Use a empty string to represent no column specified for deleteall in shell mode. useage:
deleteall ‘test’,‘r1’,'',12345 deleteall ‘test’, {ROWPREFIXFILTER => ‘prefix’}, '', 12345
Added a new hbase-asyncfs module to hold the asynchronous dfs output stream implementation for implementing WAL.
Make the logic of the versions chosen more reasonable for raw scan, to avoid lose result when using filter.
Moved to hbase-thirdparty 3.3.0.
This feature enables the HBase Web UI‘s to accept a ‘proxyuser’ via the HTTP Request’s query string. When the parameter `hbase.security.authentication.spnego.kerberos.proxyuser.enable` is set to `true` in hbase-site.xml (default is `false`), the HBase UI will attempt to impersonate the user specified by the query parameter “doAs”. This query parameter is checked case-insensitively. When this option is not provided, the user who executed the request is the “real” user and there is no ability to execute impersonation against the WebUI.
For example, if the user “bob” with Kerberos credentials executes a request against the WebUI with this feature enabled and a query string which includes `doAs=alice`, the HBase UI will treat this request as executed as `alice`, not `bob`.
The standard Hadoop proxyuser configuration properties to limit users who may impersonate others apply to this change (e.g. to enable `bob` to impersonate `alice`). See the Hadoop documentation for more information on how to configure these proxyuser rules.
bin/hbase will now dynamically select a Garbage Collector implementation based on the detected JVM version. JDKs 8,9,10 use -XX:+UseConcMarkSweepGC, while JDK11+ use -XX:+UseG1GC.
Notice a slight compatibility change. Previously, the garbage collector choice would always be appended to a user-provided value for HBASE_OPTS. As of this change, this setting will only be applied when HBASE_OPTS is unset. That means that operators who provide a value for this variable will now need to also specify the collector. This is especially important for those on JDK8, where the vm default GC is not the recommended ConcMarkSweep.
Default value: false Description: If value is true, RegionServer will abort batch requests of Put/Delete with number of rows in a batch operation exceeding threshold defined by value of config: hbase.rpc.rows.warning.threshold.
StochasticLoadBalancer functional improvement:
StochasticLoadBalancer would rebalance the cluster if there are any idle RegionServers in the cluster (RegionServer having no region), while other RegionServers have at least 1 region available.
user or admin can now use hbase shell > rename_rsgroup ‘oldname’, ‘newname’ to rename rsgroup.
Add hadoop-3.2.0 and hadoop-3.2.1 in hadoop check and when ‘--quick-hadoopcheck’ we will only check hadoop-3.2.1.
Notice that, for aligning the personality scripts across all the active branches, we will commit the patch to all active branches, but the hadoop-3.2.x support in hadoopcheck is only applied to branch-2.2+.
`-PrunSmallTests` now pass on JDK11 when using `-Phadoop.profile=3.0`.
Tests that fail because a process -- RegionServer or Master -- called System.exit, will now instead throw an exception.
Hadoop hosts have had their ulimit -u raised from 10000 to 30000 (per user, by INFRA). The Docker build container has had its limit raised from 10000 to 12500.
Support RSGroup renaming in core codebase. New API Admin#renameRSGroup(String, String) is introduced in 3.0.0.
The Canary tool now offers a WebUI when run in region mode (the default mode). It is enabled by default, and by default, it binds to 0.0.0.0:16050. This can be overridden by setting hbase.canary.info.bindAddress and hbase.canary.info.port. To disable entirely, set the port to -1.
Pass --threads=2 building on jenkins. It shortens nightly build times by about ~25%.
It works by running module build/test in parallel when dependencies allow. Upping the forkcount beyond the pom default of 0.25C would have us broach our CPU budget on jenkins when two modules are running in parallel (2 modules at 0.25% of CPU each makes 0.5C and on jenkins, hadoop nodes run two jenkins executors per host). Higher forkcounts also seems to threaten build stability.
For running tests locally, to go faster, up fork count.
$ x=“0.5C” ; mvn --threads=2 -Dsurefire.firstPartForkCount=$x -Dsurefire.secondPartForkCount=$x test -PrunAllTests
You could up the x from 0.5C to 1.0C but YMMV (On overcommitted hardware, tests start bombing out pretty soon after startup). You could try upping thread count but on occasion are likely to overcommit hardware.
Start docker with upped ulimit for nproc passing ‘--ulimit nproc=12500’. It was 10000, the default, but made it 12500. Then, set PROC_LIMIT in hbase-personality so when yetus runs, it is w/ the new 12500 value.
Pass -T2 to mvn. Makes it so we do two modules-at-a-time dependencies willing. Helps speed build and testing. Doubles the resource usage when running modules in parallel.
Master & RegionService now support refresh policy authorization defined in hbase-policy.xml without restarting service. To refresh policy, please execute hbase shell command: update_config or update_config_all after policy file updated and synced on all nodes.
This change modifies the default acquisition policy for the region's close lock in order to prevent observed starvation of close requests. The new boolean configuration parameter ‘hbase.regionserver.fair.region.close.lock’ controls the lock acquisition policy: if true, the lock is created in fair mode (default); if false, the lock is created in nonfair mode (the old default).
The PrimaryRegionCountSkewCostFunction for the StochasticLoadBalancer is only needed when the read replicas feature is enabled. With this change, that function now properly indicates that it is not needed when the read replica feature is off.
If this improvement is not available, operators with clusters that are not using the read replica feature should manually disable it by setting hbase.master.balancer.stochastic.primaryRegionCountCost to 0.0 in hbase-site.xml for all HBase Masters.
Now AsyncFSWAL can also be used against the directory which has EC enabled. Need to make sure you also make use of the hadoop 3.x client as the option is only available in hadoop 3.x.
Branches-2.3+ use maven 3.5.3 building. Older branches use 3.5.4 still.
Our ‘Build Artifacts’ have a machine directory under which we emit vitals on the host the build was run on. We used to emit the result of ‘ulimit -l’ as a file named ‘ulimit-l’. This has been hijacked to instead emit result of running ‘ulimit -a’ which includes stat on ulimit -l.
ColumnFamilyDescriptor new builder API:
/\*\*
\* Retain all versions for a given TTL(retentionInterval), and then only a specific number
\* of versions(versionAfterInterval) after that interval elapses.
\*
\* @param retentionInterval Retain all versions for this interval
\* @param versionAfterInterval Retain no of versions to retain after retentionInterval
\*/
public ModifyableColumnFamilyDescriptor setVersionsWithTimeToLive(
final int retentionInterval, final int versionAfterInterval)
org.apache.hadoop.hbase.types.PBType is marked as deprecated without any replacement. It will be moved to hbase-example module and marked as IA.Private in 3.0.0. This is a mistake as it should not be part of our public API. Users who depend on this class should just copy the code your own code base.
Expose file system level read metrics for RegionServer.
If the HBase RS runs on top of HDFS, calculate the aggregation of ReadStatistics of each HdfsFileInputStream. These metrics include: (1) total number of bytes read from HDFS. (2) total number of bytes read from local DataNode. (3) total number of bytes read locally through short-circuit read. (4) total number of bytes read locally through zero-copy read.
Because HDFS ReadStatistics is calculated per input stream, it is not feasible to update the aggregated number in real time. Instead, the metrics are updated when an input stream is closed.
Admin can determine which tables go to which rsgroup by script (setting hbase.rsgroup.table.mapping.script with local filystem path) on Master side which aims to lighten the burden of admin operations. Note, since HBase 3+, rsgroup can be specified in TableDescriptor as well, if clients specify this, master will skip the determination from script.
Here is a simple example of script: {code}
#!/bin/bash namespace=$1 tablename=$2 if [[ $namespace == test ]]; then echo test elif [[ $tablename == *foo* ]]; then echo other else echo default fi {code}
MiniZKCluster now puts up its standalone node listening on loopback/127.0.0.1 rather than “localhost”.
Bumped hadoop-two.version to 2.10.0, which means we will drop the support for hadoop-2.8.x and hadoop-2.9.x.
Change timestamp display to be ISO8601 when toString on Cell and outputting in shell....
User used to see....
column=table:state, timestamp=1583967620343 .....
... but now sees:
column=table:state, timestamp=2020-03-11T23:00:20.343Z ....
merge_region shell command can now be used to merge more than 2 regions as well. It takes a list of regions as comma separated values or as an array of regions, and not just 2 regions. The full regionnames and encoded regionnames are continued to be accepted.
Rebuild our Dockerfile with support for multiple JDK versions. Use multiple stages in the Jenkinsfile instead of yetus's multijdk because of YETUS-953. Run those multiple stages in parallel to speed up results.
Note that multiple stages means multiple Yetus invocations means multiple comments on the PreCommit. This should become more obvious to users once we can make use of GitHub Checks API, HBASE-23902.
get_slowlog_responses and clear_slowlog_responses are used to retrieve and clear slow RPC logs from RingBuffer maintained by RegionServers.
New Admin APIs:
List<SlowLogRecord> getSlowLogResponses(final Set<ServerName> serverNames, final SlowLogQueryFilter slowLogQueryFilter) throws IOException;
List<Boolean> clearSlowLogResponses(final Set<ServerName> serverNames) throws IOException;
Configs:
Default 256
Default false
For more details, please look into “Get Slow Response Log from shell” section from HBase book.
Down the flakey re-rerun fork count from 1.0C -- i.e. a fork per CPU -- to 0.25C. On a recent run, the machine had 16 cores. 0.25 is 4 cores. We'd hardcoded fork count at 3 previous to changes made by parent.
Add a checkAndMutate(row, filter) method in the AsyncTable interface and the Table interface.
This method atomically checks if the row matches the specified filter. If it does, it adds the Put/Delete/RowMutations.
This is a fluent style API, the code is like:
For Table interface: {code} table.checkAndMutate(row, filter).thenPut(put); {code}
For AsyncTable interface: {code} table.checkAndMutate(row, filter).thenPut(put) .thenAccept(succ -> { if (succ) { System.out.println(“Check and put succeeded”); } else { System.out.println(“Check and put failed”); } }); {code}
The Jira Precommit job (https://builds.apache.org/job/PreCommit-HBASE-Build/) will now look for a file within the source tree (dev-support/jenkins_precommit_jira_yetus.sh) instead of depending on a script section embedded in the job.
Changed flakey list reporting to show 5 rather than 10 items. Also changed the second and first part fort counts to be 1C rather than hardcoded 3.
HBASE-23554 | Major | Encoded regionname to regionname utility
Adds shell command regioninfo:
hbase(main):001:0\> regioninfo '0e6aa5c19ae2b2627649dc7708ce27d0'
{ENCODED =\> 0e6aa5c19ae2b2627649dc7708ce27d0, NAME =\> 'TestTable,,1575941375972.0e6aa5c19ae2b2627649dc7708ce27d0.', STARTKEY =\> '', ENDKEY =\> '00000000000000000000299441'}
Took 0.4737 seconds
This JIRA adds a new configuration - `hbase.rs.cachecompactedblocksonwrite.threshold`. This configuration is the maximum total size (in bytes) of the compacted files below which the configuration `hbase.rs.cachecompactedblocksonwrite` is honoured. If the total size of the compacted fies exceeds this threshold, even when `hbase.rs.cachecompactedblocksonwrite` is enabled, the data blocks are not cached. Caching index and bloom blocks is not affected by this configuration (user configuration is always honoured).
Default value of this configuration is Long.MAX_VALUE. This means whatever the total size of the compacted files, it wil be cached.
Implements authorization for the HBase Web UI by limiting access to certain endpoints which could be used to extract sensitive information from HBase.
Access to these restricted endpoints can be limited to a group of administrators, identified either by a list of users (hbase.security.authentication.spnego.admin.users) or by a list of groups (hbase.security.authentication.spnego.admin.groups). By default, neither of these values are set which will preserve backwards compatibility (allowing all authenticated users to access all endpoints).
Further, users who have sensitive information in the HBase service configuration can set hbase.security.authentication.ui.config.protected to true which will treat the configuration endpoint as a protected, admin-only resource. By default, all authenticated users may access the configuration endpoint.
Enables master based registry as the default registry used by clients to fetch connection metadata. Refer to the section “Master Registry” in the client documentation for more details and advantages of this implementation over the default Zookeeper based registry.
Configuration parameter that controls the registry in use: hbase.client.registry.impl
Where to set this: HBase client configuration (hbase-site.xml)
Possible values:
org.apache.hadoop.hbase.client.ZKConnectionRegistry (For ZK based registry implementation)org.apache.hadoop.hbase.client.MasterRegistry (New, for master based registry implementation)Notes on defaults:
This feature has been back ported to 2.3.0 and later releases. MasterRegistry can be enabled by setting the following client configuration.
<property> <name>hbase.client.registry.impl</name> <value>org.apache.hadoop.hbase.client.MasterRegistry</value> </property>
caffeine: 2.6.2 => 2.8.1 commons-codec: 1.10 => 1.13 commons-io: 2.5 => 2.6 disrupter: 3.3.6 => 3.4.2 httpcore: 4.4.6 => 4.4.13 jackson: 2.9.10 => 2.10.1 jackson.databind: 2.9.10.1 => 2.10.1 jetty: 9.3.27.v20190418 => 9.3.28.v20191105 protobuf.plugin: 0.5.0 => 0.6.1 zookeeper: 3.4.10 => 3.4.14 slf4j: 1.7.25 => 1.7.30 rat: 0.12 => 0.13 asciidoctor: 1.5.5 => 1.5.8 asciidoctor.pdf: 1.5.0-alpha.15 => 1.5.0-rc.2 error-prone: 2.3.3 => 2.3.4
This change introduces an internal abstraction layer which allows for new SASL-based authentication mechanisms to be used inside HBase services. All existing SASL-based authentication mechanism were ported to the new abstraction, making no external change in runtime semantics, client API, or RPC serialization format.
Developers familiar with extending HBase can implement authentication mechanism beyond simple Kerberos and DelegationTokens which authenticate HBase users against some other user database. HBase service authentication (Master to/from RegionServer) continue to operate solely over Kerberos.
Introduce a new hbase-assembly/src/main/assembly/hadoop-three-compat.xml for build with hadoop 3.x.
Add a new config to hbase-default.xml
<property> <name>hbase.procedure.store.region.hfilecleaner.plugins</name> <value>org.apache.hadoop.hbase.master.cleaner.TimeToLiveHFileCleaner</value> <description>A comma-separated list of BaseHFileCleanerDelegate invoked by the RegionProcedureStore HFileCleaner service. These HFiles cleaners are called in order, so put the cleaner that prunes the most files in front. To implement your own BaseHFileCleanerDelegate, just put it in HBase's classpath and add the fully qualified class name here. Always add the above default hfile cleaners in the list as they will be overwritten in hbase-site.xml.</description> </property>
It will share the same TTL with other HFileCleaners. And you can also implement your own cleaner and change this property to enable it.
Updated parent pom to Apache version 22.
This issues fixes an issue with Bulk Loading on installations with Kerberos enabled and more than a single RegionServer. When multiple tables are involved in hosting a table‘s regions which are being bulk-loaded into, all but the RegionServer hosting the table’s first Region will “leak” one DistributedFileSystem object onto the heap, never freeing that memory. Eventually, with enough bulk loads, this will create a situation for RegionServers where they have no free heap space and will either spend all time in JVM GC, lose their ZK session, or crash with an OutOfMemoryError.
The only mitigation for this issue is to periodically restart RegionServers. All earlier versions of HBase 2.x are subject to this issue (2.0.x, <=2.1.8, <=2.2.3)
Add a new feature to improve MTTR which have 3 steps to failover:
Compared to DLS(distributed log split), this feature will reduce region open time significantly.
Config hbase.wal.split.to.hfile to true to enable this featue.
Changed the logging in hbase-zookeeper to use built-in formatting
From the PR:
"Yes. The two create the same output... I just wrote a small test suite to increase my confidence on that. I generated many tens of millions of random byte patterns and compared the output of the two algorithms. They came back identical every time.
“Just in case any inquiring minds would like to know, there is no longer an encoding required when generating the strings. The JDK implementation specifically specifies that strings returned are StandardCharsets.ISO_8859_1. This does not change anything because UTF8 and ISO_8859 overlap for the limited character set (64 characters) the encoding uses.”
Set hbase.balancer.max.balancing to a int value which <=0 will disable region balance throttling.
If cacheOnWrite is enabled during flush or compaction, index and bloom blocks(with data blocks) would be automatically cached during write.
If a RegionServer reports a Region as OPEN in disagreement with Master's status on the Region, the Master now tells the RegionServer to silently close the Region.
Makes it so the recently added HBCKServerCrashProcedure -- the SCP that gets invoked when an operator schedules an SCP via hbck2 scheduleRecoveries command -- now works the same as SCP EXCEPT if master knows nothing of the scheduled servername. In this latter case, HBCKSCP will do a full scan of hbase:meta looking for instances of the passed servername. If any found it will attempt cleanup of hbase:meta references by reassigning any found OPEN or OPENING and by closing any in CLOSING state.
Used to fix instances of what the ‘HBCK Report’ page shows as ‘Unknown Servers’.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.HFileProcedurePrettyPrinter to run the tool.
RegionsRecoveryChore introduced as part of HBASE-22460 tries to reopen regions based on config: hbase.regions.recovery.store.file.ref.count. Region reopen needs to take into consideration all compacted away store files that belong to the region and not store files(non-compacted).
Fixed this bug as part of this Jira. Updated description for corresponding configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
Very large number of ref count on a compacted store file indicates that it is a ref leak on that object(compacted store file). Such files can not be removed after it is invalidated via compaction. Only way to recover in such scenario is to reopen the region which can release all resources, like the refcount, leases, etc. This config represents Store files Ref Count threshold value considered for reopening regions. Any region with compacted store files ref count > this value would be eligible for reopening by master. Here, we get the max refCount among all refCounts on all compacted away store files that belong to a particular region. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.WALProcedurePrettyPrinter to run the tool.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.RegionProcedureStorePerformanceEvaluation to run the tool.
Use a region based procedure store to replace the old customized WAL based procedure store. The procedure data migration is done automatically during upgrading. After upgrading, the MasterProcWALs directory will be deleted and a new MasterProc directory will be created. And notice that a region will still write WAL so we still have WAL files and they will be moved to the oldWALs directory. The file name is mostly like a normal WAL file, and the only difference is that it is ended with “$masterproc$”.
Bumped surefire plugin to 3.0.0-M4
Now AsyncFSWAL also supports Durability.FSYNC_WAL.
The configuration ‘hbase.rs.cacheblocksonwrite’ was used to enable caching the blocks on write. But purposefully we were not caching the blocks when we do compaction (since it may be very aggressive) as the caching happens as and when the writer completes a block. In cloud environments since they have bigger sized caches - though they try to enable ‘hbase.rs.prefetchblocksonopen’ (non - aggressive way of caching the blocks proactively on reader creation) it does not help them because it takes time to cache the compacted blocks. This feature creates a new configuration ‘hbase.rs.cachecompactedblocksonwrite’ which when set to ‘true’ will enable the blocks created out of compaction. Remember that since it is aggressive caching the user should be having enough cache space - if not it may lead to other active blocks getting evicted. From the shell this can be enabled by using the option per Column Family also by using the below format {code} create ‘t1’, ‘f1’, {NUMREGIONS => 15, SPLITALGO => ‘HexStringSplit’, CONFIGURATION => {‘hbase.rs.cachecompactedblocksonwrite’ => ‘true’}} {code}
Users of the MOB feature can now use the mobrefs utility to get statistics about data in the MOB system and verify the health of backing files on HDFS.
HADOOP_CLASSPATH=/etc/hbase/conf:$(hbase mapredcp) yarn jar \
/some/path/to/hbase-shaded-mapreduce.jar mobrefs mobrefs-report-output some_table foo
See javadocs of the class MobRefReporter for more details.
the reference guide has added some information about MOB internals and troubleshooting.
The reference guide now includes a walk through of disabling the MOB feature if needed while maintaining availability.
Fixed unbalanced braces in string representation within HBase shell
The default rpc timeout for ReplicationSourceShipper#shipEdits is 60s, when bulkload replication enabled, timeout exception may be occurred. Now we can conf the timeout value through replication.source.shipedits.timeout, and it’s adaptive.
The newer HBase Thrift SPNEGO configs should not be required. The hbase.thrift.spnego.keytab.file and hbase.thrift.spnego.principal configs will fall back to the hbase.thrift.keytab.file and hbase.thrift.kerberos.principal original configs. The older configs will log a deprecation warning. It is preferred to new the newer SPNEGO configurations.
With BinaryComponentCompartor applications will be able to design diverse and powerful set of filters for rows and columns. See https://issues.apache.org/jira/browse/HBASE-22969 for example. In general, the comparator can be used with any filter taking ByteArrayComparable. As of now, following filters take ByteArrayComparable:
Adds a .editorconfig file with configurations populated by IntelliJ, based on our checkstyle configuration. There's lots of IntelliJ-specific configs in here that I assume are not replicated to Eclipse or Netbeans users. Any devs using those tools should push whatever updates they see fit, but please start with the checkstyle configs as the origin of truth.
An hbck2 scheduleRecoveries will run a subclass of ServerCrashProcedure which asks Master what Regions were on the dead Server but it will also do a hbase:meta table scan to see if any vestiges of the old Server remain (for the case where an SCP failed mid-point leaving references in place or where Master and hbase:meta deviated in accounting).
If holes in hbase:meta, hbck2 fixMeta now will update Master in-memory state so you do not need to restart master just so you can assign the new hole-bridging regions.
hbck2 scheduleRecoveries will now run a SCP that also looks in hbase:meta for any references to the scheduled server -- not just consult Master in-memory state -- just in case vestiges of the server are leftover in hbase:meta
HBase internal chores now log a moving average of how long execution of each chore takes at INFO level for the logger org.apache.hadoop.hbase.ScheduledChore.
Such messages will happen at most once per five minutes.
CleanerChore delegate initialization is now logged at INFO level instead of DEBUG
The ‘Procedures & Locks’ tab in Master UI only displays problematic Procedures now (RUNNABLE, WAITING-TIMEOUT, etc.). It no longer notes procedures whose state is SUCCESS.
the Apache HBase REST Proxy now uses Jackson Databind version 2.9.10.1 to address the following CVEs
Users of prior releases with Jackson Databind 2.9.10 are advised to either upgrade to this release or to upgrade their local Jackson Databind jar directly.
The MOB compaction process in the HBase Master now logs more about its activity.
In the event that you run into the problems described in HBASE-22075, there is a new HFileCleanerDelegate that will stop all removal of MOB hfiles from the archive area. It can be configured by adding org.apache.hadoop.hbase.mob.ManualMobMaintHFileCleaner to the list configured for hbase.master.hfilecleaner.plugins. This new cleaner delegate will cause your archive area to grow unbounded; you will have to manually prune files which may be prohibitively complex. Consider if your use case will allow you to mitigate by disabling mob compactions instead.
Caveats:
%hbase_root%/archive that accumulates files over time. You will have to determine which of these files are safe or not to remove.TRACE logging, the CleanerChore logger will include archive retention decision justifications.Adds per-user metrics for reads/writes to each RegionServer. These metrics are exported by default. hbase.regionserver.user.metrics.enabled can be used to disable the feature if desired for any reason.
Leaked store files can not be removed even after it is invalidated via compaction. A reasonable mitigation for a reader reference leak would be a fast reopen of the region on the same server.
Configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
This config represents Store files Ref Count threshold value considered for reopening regions. Any region with store files ref count > this value would be eligible for reopening by master. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Added a comment to make clear that read/write success counts are tallying column family success counts, not region success counts.
Additionally, the region read and write latencies previously only stored the latencies of the last column family of the region reads/writes. This has been fixed by using a map of each region to a list of read and write latency values.
Changes the message on the FNFE exception thrown when the file a Reference points to is missing; the message now includes detail on Reference as well as pointed-to file so can connect how FNFE relates to region open.
Use ‘totalRowActionRequestCount’ to calculate QPS on web UI.
Downstream users who wish to programmatically check the health of their HBase cluster may now rely on a public interface derived from the previously private implementation of the canary cli tool. The interface is named Canary and can be found in the user facing javadocs.
Downstream users who previously relied on the invoking the canary via the Java classname (either on the command line or programmatically) will need to change how they do so because the non-public implementation has moved.
Since 2.0.0，when one regionserver crashed and back online again, AssignmentManager will retain the region locations and try assign the regions to this regionserver(same host:port with the crashed one) again. But for 1.x.x, the behavior is round-robin assignment for the regions belong to the crashed regionserver. This jira change the “retain” assignment to round-robin assignment, which is same with 1.x.x version. This change will make the failover faster and improve availability.
Remove backward compatibility from `truncate` and `truncate_preserve` shell commands. This means that these commands from HBase Clients are not compatible with pre-0.99 HBase clusters.
giving the region mover “unload” command a region server name that isn‘t recognized by the cluster results in a "I don’t know about that host" message instead of a NPE.
set log level to DEBUG if you'd like the region mover to log the set of region server names it got back from the cluster.
Added a new IOEngine type for Bucket cache ie Persistent memory. In order to use BC over pmem configure IOEngine as <property> <name>hbase.bucketcache.ioengine</name> <value> pmem:///path in persistent memory </value> </property>
By default, snapshot auto cleanup based on TTL would be enabled for any new cluster. At any point in time, if snapshot cleanup is supposed to be stopped due to some snapshot restore activity or any other reason, it is advisable to disable it using shell command: hbase> snapshot_cleanup_switch false
We can re-enable it using: hbase> snapshot_cleanup_switch true
We can query whether snapshot auto cleanup is enabled for cluster using: hbase> snapshot_cleanup_enabled
Adds fix of overlaps to the fixMeta hbck service method. Uses the bulk-merge facility. Merges a max of 10 at a time. Set hbase.master.metafixer.max.merge.count to higher if you want to do more than 10 in the one go.
This issue adds via its subtasks:
* An ‘HBCK Report’ page to the Master UI added by HBASE-22527+HBASE-22709+HBASE-22723+ (since 2.1.6, 2.2.1, 2.3.0). Lists consistency or anomalies found via new hbase:meta consistency checking extensions added to CatalogJanitor (holes, overlaps, bad servers) and by a new ‘HBCK chore’ that runs at a lesser periodicity that will note filesystem orphans and overlaps as well as the following conditions: ** Master thought this region opened, but no regionserver reported it. ** Master thought this region opened on Server1, but regionserver reported Server2 ** More than one regionservers reported opened this region Both chores can be triggered from the shell to regenerate ‘new’ reports. * Means of scheduling a ServerCrashProcedure (HBASE-21393). * An ‘offline’ hbase:meta rebuild (HBASE-22680). * Offline replace of hbase.version and hbase.id * Documentation on how to use completebulkload tool to ‘adopt’ orphaned data found by new HBCK2 ‘filesystem’ check (see below) and ‘HBCK chore’ (HBASE-22859) * A ‘holes’ and ‘overlaps’ fix that runs in the master that uses new bulk-merge facility to collapse many overlaps in the one go. * hbase-operator-tools HBCK2 client tool got a bunch of additions: ** A specialized ‘fix’ for the case where operators ran old hbck ‘offlinemeta’ repair and destroyed their hbase:meta; it ties together holes in meta with orphaned data in the fs (HBASE-22567) ** A ‘filesystem’ command that reports on orphan data as well as bad references and hlinks with a ‘fix’ for the latter two options (based on hbck1 facility updated). ** Adds back the ‘replication’ fix facility from hbck1 (HBASE-22717)
The compound result is that hbck2 is now in excess of hbck1 abilities. The provided functionality is disaggregated as per the hbck2 philosophy of providing ‘plumbing’ rather than ‘porcelain’ so there is work to do still adding fix-it playbooks, scripting across outages, and automation.
HBASE-21879 introduces a utility class (org.apache.hadoop.hbase.io.ByteBuffAllocator) used for allocating/freeing ByteBuffers from/to NIO ByteBuffer pool, when BucketCache enabled with file or mmap engine, we will use this ByteBuffer pool to avoid temp ByteBuffer allocation a lot.
Introduces hbtop that‘s a real-time monitoring tool for HBase like Unix’s top command. See the ref guide for the details: https://hbase.apache.org/book.html#hbtop
Before this issue, read path was 100% offheap when block is in the BucketCache. But if a cache miss, then the RS needs to read the block via an on-heap API which causes high young-GC pressure.
This issue adds reading the block via offheap even if reading the block from filesystem directly. It requires hadoop version(>=2.9.3) but can also work with older hadoop versions (all works but we continue to read block onheap). It also requires HBASE-21946 which is not yet in place as of this writing/hbase-2.3.0.
We have written a careful doc about the implementation, performance and practice here: https://docs.google.com/document/d/1xSy9axGxafoH-Qc17zbD2Bd--rWjjI00xTWQZ8ZwI_E/edit#heading=h.nch5d72p27ex
Extends StochasticLoadBalancer to support user-provided cost function. These are loaded in addition to the default set of cost functions. Custom function implementations must extend StochasticLoadBalancer$CostFunction. Enable any additional functions by placing them on the master class path and configuring hbase.master.balancer.stochastic.additionalCostFunctions with a comma-separated list of fully-qualified class names.
Replace the ForkJoinPool in CleanerChore by ThreadPoolExecutor which can limit the spawn thread size and avoid the master GC frequently. The replacement is an internal implementation in CleanerChore, so no config key change, the upstream users can just upgrade the hbase master without any other change.
Introduced a new config key for the snapshot taking/restoring operations at master side: hbase.master.executor.snapshot.threads, its default value is 3. means we can have 3 snapshot operations running at the same time.
Add serveral API in TimeRange class for avoiding using the deprecated TimeRange constructor: * TimeRange#from: Represents the time interval [minStamp, Long.MAX_VALUE) * TimeRange#until: Represents the time interval [0, maxStamp) * TimeRange#between: Represents the time interval [minStamp, maxStamp)
Provide a public method in MultiRowRangeFilter class to speed the requirement of filtering with multiple row prefixes, it will expand the row prefixes as multiple rowkey ranges by MultiRowRangeFilter, it's more efficient. {code} public MultiRowRangeFilter(byte[][] rowKeyPrefixes); {code}
Update the base docker image to ubuntu 18.04 for the find flaky tests jenkins job.
Adds a fixMeta method to hbck Service. Fixes holes in hbase:meta. Follow-up to fix overlaps. See HBASE-22567 also.
Follow-on is adding a client-side to hbase-operator-tools that can exploit this new addition (HBASE-22825)
Changes merge so you can merge more than two regions at a time. Currently only available inside HBase. HBASE-22827, a follow-on, is about exposing the facility in the Admin API (and then via the shell).
New shaded artifact for testing: hbase-shaded-testing-util.
After HBASE-22776, the steps to config user scan snapshot feature is as followings:
We found a critical bug which can lead to WAL corruption when Durability.ASYNC_WAL is used. The reason is that we release a ByteBuffer before actually persist the content into WAL file.
The problem maybe lead to several errors, for example, ArrayIndexOfOutBounds when replaying WAL. This is because that the ByteBuffer is reused by others.
ERROR org.apache.hadoop.hbase.executor.EventHandler: Caught throwable while processing event RS_LOG_REPLAY java.lang.ArrayIndexOutOfBoundsException: 18056 at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1365) at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1358) at org.apache.hadoop.hbase.PrivateCellUtil.matchingFamily(PrivateCellUtil.java:735) at org.apache.hadoop.hbase.CellUtil.matchingFamily(CellUtil.java:816) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEditFamily(WALEdit.java:143) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEdit(WALEdit.java:148) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:297) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:195) at org.apache.hadoop.hbase.regionserver.SplitLogWorker$1.exec(SplitLogWorker.java:100)
And may even cause segmentation fault and crash the JVM directly. You will see a hs_err_pidXXX.log file and usually the problem is SIGSEGV. This is usually because that the ByteBuffer has already been returned to the OS and used for other purpose.
The problem has been reported several times in the past and this time Wellington Ramos Chevreuil provided the full logs and deeply analyzed the logs so we can find the root cause. And Lijin Bin figured out that the problem may only happen when Durability.ASYNC_WAL is used. Thanks to them.
The problem only effects the 2.x releases, all users are highly recommand to upgrade to a release which has this fix in, especially that if you use Durability.ASYNC_WAL.
Add a new method runHbckChore in Hbck interface and a new shell cmd hbck_chore_run to request HBCK chore to run at master side.
Adds a “CatalogJanitor hbase:meta Consistency Issues” section to the new ‘HBCK Report’ page added by HBASE-22709. This section is empty unless the most recent CatalogJanitor scan turned up problems. If so, will show table of issues found.
When CatalogJanitor runs, it now checks for holes, overlaps, empty info:regioninfo columns and bad servers. Dumps findings into log. Follow-up adds report to new ‘HBCK Report’ linked off the Master UI.
NOTE: All features but the badserver check made it into branch-2.1 and branch-2.0 backports.
The misspelled BufferedMutatorParams.opertationTimeout method has been marked as deprecated, and will be removed in 4.0.0. Please use the BufferedMutatorParams.operationTimeout method instead.
If a table user scan snapshots of the table, please config the following table scheme attribute to make granted users' ACLs are added to hfiles: alter ‘t1’, CONFIGURATION => {‘hbase.user.scan.snapshot.enable’ => ‘true’}
This feature is enabled by default. And the hbck chore run per 60 minutes by default. You can config “hbase.master.hbck.checker.interval” to a value lesser than or equal to 0 for disabling the chore.
Notice: the config “hbase.master.hbck.checker.interval” was renamed to “hbase.master.hbck.chore.interval” in HBASE-22737.
The HFileCleaner will clean the empty directories under archive, but if enable user scan snaphot feature, the user ACLs are set at there directories, so please config the following cleaner to make the directories with user ACLs not be cleaned: hbase.master.hfilecleaner.plugins=org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclCleaner
Upgrade jackson databind dependency to 2.9.9.1 due to CVEs
https://nvd.nist.gov/vuln/detail/CVE-2019-12814
https://nvd.nist.gov/vuln/detail/CVE-2019-12384
Add a new master web UI to show the potentially problematic opened regions. There are three case:
Feature: Take a Snapshot With TTL for auto-cleanup
Attribute:
Configs:
Default Snapshot TTL:
If Snapshot cleanup is supposed to be stopped due to some snapshot restore activity, disable it with config:
For more details, see the section “Take a Snapshot With TTL” in the HBase Reference Guide.
The config point “hbase.offheapcache.minblocksize” was wrong and is now deprecated. The new config point is “hbase.blockcache.minblocksize”.
OfflineMetaRepair is no longer supported in HBase-2+. Please refer to https://hbase.apache.org/book.html#HBCK2
This tool is deprecated in 2.x and will be removed in 3.0.
Mark the Hbck#scheduleServerCrashProcedure(List<HBaseProtos.ServerName> serverNames) as deprecated. Use Hbck#scheduleServerCrashProcedures(List<ServerName> serverNames) instead.
In HBASE-20734 we moved the recovered.edits onto the wal file system but when constructing the directory we missed the BASE_NAMESPACE_DIR(‘data’). So when using the default config, you will find that there are lots of new directories at the same level with the ‘data’ directory.
In this issue, we add the BASE_NAMESPACE_DIR back, and also try our best to clean up the wrong directories. But we can only clean up the region level directories, so if you want a clean fs layout on HDFS you still need to manually delete the empty directories at the same level with ‘data’.
The effect versions are 2.2.0, 2.1.[1-5], 1.4.[8-10], 1.3.[3-5].
Add a coprocessor to set HDFS acls to make hbase granted users with READ permission have the access to scan snapshots. To use this feature, please make sure the HDFS config is set: dfs.namenode.acls.enabled=true fs.permissions.umask-mode=027
and set the HBase config: hbase.coprocessor.master.classes=“org.apache.hadoop.hbase.security.access.AccessController,org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclController” hbase.user.scan.snapshot.enable=true
hbase.regionserver.compaction.check.period is used for controlling how often the compaction checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
hbase.regionserver.flush.check.period is used for controlling how ofter the flush checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
When run with JDK11 HBase now uses more recent version of the jaxws reference implementation (v2.3.2).
Added completebulkload short name for BulkLoadHFilesTool to bin/hbase.
Change the default hadoop-3 version to 3.1.2. Drop the support for the releases which are effected by CVE-2018-8029, see this email https://lists.apache.org/thread.html/3d6831c3893cd27b6850aea2feff7d536888286d588e703c6ffd2e82@%3Cuser.hadoop.apache.org%3E
This change exposes the aggregate count of store reader references for a given store as ‘storeRefCount’ in region metrics and ClusterStatus.
The HBase “source checksum” now uses SHA512 instead of MD5.
The CellUtil.setTimestamp method changes to be an API with audience LimitedPrivate(COPROC) in HBase 3.0. With that designation the API should remain stable within a given minor release line, but may change between minor releases.
Previously, this method was deprecated in HBase 2.0 for removal in HBase 3.0. Deprecation messages in HBase 2.y releases have been updated to indicate the expected API audience change.
The access method was used to the HttpServerFunctionalTest class as a common place.
The class LossyCounting was unintentionally marked Public but was never intended to be part of our public API. This oversight has been corrected and LossyCounting is now marked as Private and going forward may be subject to additional breaking changes or removal without notice. If you have taken a dependency on this class we recommend cloning it locally into your project before upgrading to this release.
Warnings for level headings are corrected in the book for the HBase Incompatibilities section.
Add hadoop 3.0.3, 3.1.1 3.1.2 in our hadoop check jobs.
The DumpReplicationQueues tool will now list replication queues sorted in chronological order.
Support get|set LogLevel in secure(kerberized) environment.
Fixes a formatting issue in the administration section of the book, where listing indentation were a little bit off.
This change adds the new method listNamespaces to the Admin interface, which can be used to retrieve a list of the namespaces present in the schema as an unprivileged operation. Formerly the only available method for accomplishing this was listNamespaceDescriptors, which requires GLOBAL CREATE or ADMIN permissions.
Now the default hadoop-two.version has been changed to 2.8.5, and all hadoop versions before 2.8.2(exclude) will not be supported any more.
Removed extra + in HRegion, HStore and LoadIncrementalHFiles for branch-2 and HRegion and HStore for branch-1.
Updated metrics core from 3.2.1 to 3.2.6.
The rubocop definition for the maximum method length was set to 75.
Fixes the formatting of the “Voting on Release Candidates” to actually show the quote and code formatting of the RAT check.
The rubocop configuration in the hbase-shell module now allows a line length with 100 characters, instead of 80 as before. For everything before 2.1.5 this change introduces rubocop itself.
This change adds new conditions for rolling the WAL for when syncs on the HDFS writer pipeline are perceived to be slow.
As before the configuration parameter hbase.regionserver.wal.slowsync.ms sets the slow sync warning threshold.
If we encounter hbase.regionserver.wal.slowsync.roll.threshold number of slow syncs (default 100) within the interval defined by hbase.regionserver.wal.slowsync.roll.interval.ms (default 1 minute), we will request a WAL roll.
Or, if the time for any sync exceeds the threshold set by hbase.regionserver.wal.roll.on.sync.ms (default 10 seconds) we will request a WAL roll immediately.
Operators can monitor how often these new thresholds result in a WAL roll by looking at newly added metrics to the WAL related metric group: * slowSyncRollRequest - How many times a roll was requested due to sync too slow on the write pipeline.
Additionally, as a part of this change there are also additional metrics for existing reasons for a WAL roll: * errorRollRequest - How many times a roll was requested due to I/O or other errors. * sizeRollRequest - How many times a roll was requested due to file size roll threshold.
MajorCompactorTTL Tool allows to compact all regions in a table that have been TTLed out. This saves space on DFS and is useful for tables which are similar to time series data. This is typically scheduled to run frequently (say via cron) to cleanup old data on an ongoing basis.
RSGroupMajorCompactionTTL tool is similar to MajorCompactorTTL but runs at a region server group level. If multiple tables in an rsgroup are similar to time-series data, then it runs a single command to clean them up. As more tables are added/removed from rsgroup, it's easy to have a single command to take care of all of them.
This change allows the system and superusers to initiate compactions, even when a space quota violation policy disallows compactions from happening. The original intent behind disallowing of compactions was to prevent end-user compactions from creating undue I/O load, not disallowing *any* compaction in the system.
Maven project integration for Eclipse has been isolated into a maven profile to ensure it only is active when in an Eclipse project.
Things should continue to behave the same for Eclipse users. If something should go wrong folks should manually activate the eclipse-specific profile.
Deprecated Preemptive Fail Fast related constants in HConstants, the support of this feature will be removed in 3.0.0 so use these constants will have no effect for 3.0.0+ releases. And the constants will be kept till 4.0.0.
Users can use ‘hbase.client.perserver.requests.threshold’ to control the number of concurrent requests to the same region server. Please see the release note of HBASE-16388 for more details.
Adds new configuration hbase.client.failure.map.cleanup.interval which defaults to ten minutes.
The default version of JRuby shipped with HBase has been updated to the JRuby 9.1.17.0 release.
For details on changes see the release notes for JRuby 9.1.17.0
Add below method in Table interface:
RegionLocator getRegionLocator() throws IOException;
Add below methods in AsyncTable interface:
AsyncTableRegionLocator getRegionLocator(); CompletableFuture<TableDescriptor> getDescriptor();
LruBlockCache uses the Segmented LRU (SLRU) policy to capture frequency and recency of the working set. It achieves concurrency by using an O(n) background thread to prioritize the entries and evict. Accessing an entry is O(1) by a hash table lookup, recording its logical access time, and setting a frequency flag. A write is performed in O(1) time by updating the hash table and triggering an async eviction thread. This provides ideal concurrency and minimizes the latencies by penalizing the thread instead of the caller. However the policy does not age the frequencies and may not be resilient to various workload patterns.
This change introduces a new L1 policy, TinyLfuBlockCache, which records the frequency in a counting sketch, ages periodically by halving the counters, and orders entries by SLRU. An entry is discarded by comparing the frequency of the new arrival to the SLRU's victim, and keeping the one with the highest frequency. This allows the operations to be performed in O(1) time and, though the use of a compact sketch, a much larger history is retained beyond the current working set. In a variety of real world traces the policy had near optimal hit rates.
New configuration variable hfile.block.cache.policy sets the eviction policy for the L1 block cache. The default is “LRU” (LruBlockCache). Set to “TinyLFU” to use TinyLfuBlockCache instead.
Introduced
Future<Void> createTableAsync(TableDescriptor);
Introduced these methods: void move(byte[]); void move(byte[], ServerName); Future<Void> splitRegionAsync(byte[]);
These methods are deprecated: void move(byte[], byte[])
Add a new jenkins file for running pre commit check for GitHub PR.
Add cloneSnapshot/restoreSnapshot with acl methods in AsyncAdmin.
When insufficient permissions, you now get:
HTTP/1.1 403 Forbidden
on the HTTP side, and in the message
Forbidden org.apache.hadoop.hbase.security.AccessDeniedException: org.apache.hadoop.hbase.security.AccessDeniedException: Insufficient permissions for user ‘myuser',action: get, tableName:mytable, family:cf. at org.apache.ranger.authorization.hbase.RangerAuthorizationCoprocessor.authorizeAccess(RangerAuthorizationCoprocessor.java:547) and the rest of the ADE stack
Now we will sort the javac WARNING/ERROR before generating diff in pre-commit so we can get a stable output for the error prone. The downside is that we just sort the output lexicographically so the line number will also be sorted lexicographically, which is a bit strange to human.
Exposes a new configuration property “zookeeper.multi.max.size” which dictates the maximum size of deletes that HBase will make to ZooKeeper in a single RPC. This property defaults to 1MB, which should fall beneath the default ZooKeeper limit of 2MB, controlled by “jute.maxbuffer”.
Fixed awkward dependency issue that prevented site building.
HBase 2.1.4 shipped with an early version of this fix that incorrectly altered the libraries included in our binary assembly for using Apache Hadoop 2.7 (the current build default Hadoop version for 2.1.z). For folks running out of the box against a Hadoop 2.7 cluster (or folks who skip the installation step of replacing the bundled Hadoop libraries) this will result in a failure at Region Server startup due to a missing class definition. e.g.:
2019-03-27 09:02:05,779 ERROR [main] regionserver.HRegionServer: Failed construction RegionServer java.lang.NoClassDefFoundError: org/apache/htrace/SamplerBuilder at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:644) at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:628) at org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:149) at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2667) at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:93) at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2701) at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2683) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:372) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:171) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:356) at org.apache.hadoop.fs.Path.getFileSystem(Path.java:295) at org.apache.hadoop.hbase.util.CommonFSUtils.getRootDir(CommonFSUtils.java:362) at org.apache.hadoop.hbase.util.CommonFSUtils.isValidWALRootDir(CommonFSUtils.java:411) at org.apache.hadoop.hbase.util.CommonFSUtils.getWALRootDir(CommonFSUtils.java:387) at org.apache.hadoop.hbase.regionserver.HRegionServer.initializeFileSystem(HRegionServer.java:704) at org.apache.hadoop.hbase.regionserver.HRegionServer.<init>(HRegionServer.java:613) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.hadoop.hbase.regionserver.HRegionServer.constructRegionServer(HRegionServer.java:3029) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.start(HRegionServerCommandLine.java:63) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.run(HRegionServerCommandLine.java:87) at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) at org.apache.hadoop.hbase.util.ServerCommandLine.doMain(ServerCommandLine.java:149) at org.apache.hadoop.hbase.regionserver.HRegionServer.main(HRegionServer.java:3047) Caused by: java.lang.ClassNotFoundException: org.apache.htrace.SamplerBuilder at java.net.URLClassLoader.findClass(URLClassLoader.java:381) at java.lang.ClassLoader.loadClass(ClassLoader.java:424) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349) at java.lang.ClassLoader.loadClass(ClassLoader.java:357) ... 26 more
Workaround via any one of the following:
hadoop executable is in the PATH seen at Region Server startup and that you are not using the HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP bypass.htrace-core-3.1.0-incubating.jar available to all Region Servers via the HBASE_CLASSPATH environment variable.htrace-core-3.1.0-incubating.jar available to all Region Servers by copying it into the directory ${HBASE_HOME}/lib/client-facing-thirdparty/.Add a listTableDescriptors(List<TableName>) method in the AsyncAdmin interface, to align with the Admin interface.
Deprecate Admin.deleteSnapshot(byte[]), please use the String version instead.
Add a mergeRegionsAsync(byte[][], boolean) method in the AsyncAdmin interface.
Instead of using assert, now we will throw IllegalArgumentException when you want to merge less than 2 regions at client side. And also, at master side, instead of using assert, now we will throw DoNotRetryIOException if you want merge more than 2 regions, since we only support merging two regions at once for now.
Add drainXXX parameter for balancerSwitch/splitSwitch/mergeSwitch methods in the AsyncAdmin interface, which has the same meaning with the synchronous parameter for these methods in the Admin interface.
As of HBase 3.0, the ByteBufferUtils class is now marked as a Private API for internal project use only. Downstream users are advised that it no longer has any compatibility promises across releases.
As of earlier HBase release lines the class is now marked as deprecated to call attention to this planned transition.
bulkload (HFileOutputFormat2) support config the compression on client ,you can set the job configuration “hbase.mapreduce.hfileoutputformat.compression” override the auto-detection of the target table's compression
Add a cloneSnapshotAsync method with restoreAcl parameter. Deprecated restoreSnapshotAsync method as it just ignores the failsafe configuration. Make snapshotAsync method returns a Future<Void>. Deprecated the snapshot related methods which take a ‘byte[]’ as the snapshot name. Use default methods to reduce the code base for implementation classes.
Deprecated AsyncTable.isTableAvailable(TableName, byte[][]).
After HBASE-21871, we can specify a peer table name with --peerTableName in VerifyReplication tool like the following: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable 5 TestTable
In addition, we can compare any 2 tables in any remote clusters with specifying both peerId and --peerTableName.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable zk1,zk2,zk3:2181/hbase TestTable
Adds below flush, split, and compaction metrics
HBASE-21481 improves the quality of access control, by strengthening the protection of super users's privileges.
Now we have four types of RIT procedure metrics, assign, unassign, move, reopen. The meaning of assign/unassign is changed, as we will not increase the unassign metric and then the assign metric when moving a region. Also introduced two new procedure metrics, open and close, which are used to track the open/close region calls to region server. We may send open/close multiple times to finish a RIT since we may retry multiple times.
Problem: This is an old problem since HBASE-2231. The compaction event marker was only writed to WAL. But after flush, the WAL may be archived, which means an useful compaction event marker be deleted, too. So the compacted store files cannot be archived when region open and replay WAL.
Solution: After this jira, the compaction event tracker will be writed to HFile. When region open and load store files, read the compaction evnet tracker from HFile and archive the compacted store files which still exist.
HBase contains two quota scopes: MACHINE and CLUSTER. Before this patch, set quota operations did not expose scope option to client api and use MACHINE as default, CLUSTER scope can not be set and used. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’
This issue implements CLUSTER scope in a simple way: For user, namespace, user over namespace quota, use [ClusterLimit / RSNum] as machine limit. For table and user over table quota, use [ClusterLimit / TotalTableRegionNum * MachineTableRegionNum] as machine limit. After this patch, user can set CLUSTER scope quota, but MACHINE is still default if user ignore scope. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’ set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => MACHINE set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => CLUSTER
Change spotbugs version to 3.1.11.
This modifies “status ‘replication’” output, fixing inconsistencies on the reporting times and ages of last shipped edits, as well as wrong calculation of replication lags.
It also introduces additional info for each recovery queue, which was not accounted by this command before.
The new output for “status ‘replication’” command is explained in details below: a) Source started, target stopped, no edits arrived on source yet: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ... b) Source started, target stopped, add edit on source: ... Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:21:00 GMT 2018, Replication Lag=2459 ... c) Source started, target stopped, edit added on source, restart source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 Recovered Queue: 1-hbase01.home,16020,1542784524057 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:23:00 GMT 2018, Replication Lag=201495 ... d) Source started, target stopped, add edit on source, restart source, add another edit on source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=6349 Recovered Queue: 1-hbase01.home,16020,1542782758742 No Ops shipped since last restart, SizeOfLogQueue=0, TimeStampOfLastArrivedInSource=Wed Nov 21 06:53:05 GMT 2018, Replication Lag=569394 ... e) Source started, target stopped, add edit on source, restart source, add another edit on source, start target: ... SOURCE: PeerID=1 Normal Queue: 1 AgeOfLastShippedOp=30000, TimeStampOfLastShippedOp=Wed Nov 21 07:07:58 GMT 2018, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=0 ... f) Source started, target stopped, add edit on source, restart source, restart target: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ...
Remove bloom filter type ROWPREFIX_DELIMITED. May add it back when find a better solution.
Support enable or disable exceed throttle quota. Exceed throttle quota means, user can over consume user/namespace/table quota if region server has additional available quota because other users don't consume at the same time. Use the following shell commands to enable/disable exceed throttle quota: enable_exceed_throttle_quota disable_exceed_throttle_quota There are two limits when enable exceed throttle quota:
Remove jackson dependencies from most hbase modules except hbase-rest, use shaded gson instead. The output json will be a bit different since jackson can use getter/setter, but gson will always use the fields.
Mark HConstants.META_QOS as deprecated. It is for internal use only, which is the highest priority. You should not try to set a priority greater than or equal to this value, although it is no harm but also useless.
This patch adds the ability to disable split and/or merge for a table (By default, split and merge are enabled for a table).
Allows shell to set Scan options previously not exposed. See additions as part of the scan help by typing following hbase shell:
hbase> help ‘scan’
We can specify peerQuorumAddress instead of peerId in VerifyReplication tool. So it no longer requires peerId to be setup when using this tool.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication zk1,zk2,zk3:2181/hbase testTable
Introduce a VerifyWALEntriesReplicationEndpoint which replicates nothing but only verifies if all the cells are valid. It can be used to capture bugs for writing WAL, as most times we will not read the WALs again after writing it if there are no region server crashes.
Introduced an new config key in this issue: hbase.regionserver.inmemory.compaction.pool.size. the default value would be 10. you can configure this to set the pool size of in-memory compaction pool. Note that all memstores in one region server will share the same pool, so if you have many regions in one region server, you need to set this larger to compact faster for better read performance.
Make StoppedRpcClientException extend DoNotRetryIOException.
To implement user permission control in Precedure V2, move grant and revoke method from AccessController to master firstly. Mark AccessController#grant and AccessController#revoke as deprecated and please use Admin#grant and Admin#revoke instead.
IMPORTANT: Due to security issues, all users who use hbase thrift should avoid using releases which do not have this fix.
The effect releases are: 2.1.x: 2.1.2 and below 2.0.x: 2.0.4 and below 1.x: 1.4.x and below
If you are using the effect releases above, please consider upgrading to a newer release ASAP.
For users who have configured hbase.bucketcache.ioengine with either the file:, files:, or mmap: prefix, and configured it to be persistent via the hbase.bucketcache.persistent.path property, the serialization format of the bucket cache has changed between versions. The old state will not be read during startup, and there is currently no migration path. The impact is expected to be minimal, however, since the cache will rebuild over time as access patterns dictate.
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Fixes a couple of bugs in ZooKeeper interaction. Firstly, zk sync() call that is used to sync the lagging followers with leader so that the client sees a consistent snapshot state was actually asynchronous under the hood. We make it synchronous for correctness. Second, zookeeper events are now processed in a separate thread rather than doing it in the thread context of zookeeper client connection. This decoupling frees up client connection quickly and avoids deadlocks.
Update our package version numbers throughout the Dockerfiles to be pinned to their epic:upstream-version components only. Previously we'd specify the full debian package version number, including the debian-revision. This lead to instability as debian packaging details changed. See also man deb-version
Adds a new metric where we collect the number of read requests (tracked per row) whether the row was fetched completely from memstore or it was pulled from files and memstore. The metric is now collected under the mbean for Tables and under the mbean for regions. Under table mbean ie.- ‘name": "Hadoop:service=HBase,name=RegionServer,sub=Tables’ The new metrics will be listed as {code} “Namespace_default_table_t3_columnfamily_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_columnfamily_f1_metric_mixedRowReadsCount”: 1, {code} Where the format is Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_mixedRowReadsCount {code}
The same one under the region ie. “name”: “Hadoop:service=HBase,name=RegionServer,sub=Regions”, comes as {code} “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_mixedRowReadsCount”: 1, {code} where Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_mixedRowReadsCount This is also an aggregate against every store the number of reads that happened purely from the memstore or it was a mixed read that happened from memstore and file.
This adds [-h|-help] options to rowcounter. Passing either -h or -help will print rowcounter guide as below:
$hbase rowcounter -h
usage: hbase rowcounter <tablename> [options] [<column1> <column2>...] Options: --starttime=<arg> starting time filter to start counting rows from. --endtime=<arg> end time filter limit, to only count rows up to this timestamp. --range=<arg> [startKey],[endKey][;[startKey],[endKey]...]] --expectedCount=<arg> expected number of rows to be count. For performance, consider the following configuration properties: -Dhbase.client.scanner.caching=100 -Dmapreduce.map.speculative=false
CI coverage has been extended to include Hadoop 3.2.x for HBase 2.2+.
Adds being able to edit hbase:meta table schema. For example,
hbase(main):006:0> alter ‘hbase:meta’, {NAME => ‘info’, DATA_BLOCK_ENCODING => ‘ROW_INDEX_V1’} Updating all regions with the new schema... All regions updated. Done. Took 1.2138 seconds
You can even add columnfamilies. Howevert, you cannot delete any of the core hbase:meta column families such as ‘info’ and ‘table’.
This ticket summarizes significant improvements and expansion to the metrics surface area. Interested users should review the individual sub-tasks.
Adds backoff in ServerCrashProcedure wait on WAL split to complete if large backlog of files to split (Its possible to avoid SCP blocking, waiting on WALs to split if you use procedure-based splitting -- set ‘hbase.split.wal.zk.coordinated’ to false to enable procedure based wal splitting.)
Notice this has changed log level for mismatching row keys, originally those were being logged at INFO level, now it's logged at DEBUG level. This is consistent with the logging of mismatching cells. Also, for missing row keys, it now logs row key values in human readable format, making it more meaningful for operators troubleshooting mismatches.
Introduce a new config hbase.replication.drop.on.deleted.columnfamily, default is false. When config to true, the replication will drop the edits for columnfamily that has been deleted from the replication source and target.
This change extends the Normalizer with a handful of new configurations. The configuration points supported are:
hbase.normalizer.split.enabled Whether to split a region as part of normalization. Default: true.hbase.normalizer.merge.enabled Whether to merge a region as part of normalization. Default true.hbase.normalizer.min.region.count The minimum number of regions in a table to consider it for merge normalization. Default: 3.hbase.normalizer.merge.min_region_age.days The minimum age for a region to be considered for a merge, in days. Default: 3.hbase.normalizer.merge.min_region_size.mb The minimum size for a region to be considered for a merge, in whole MBs. Default: 1.Add a hbase-logging module, put the log4j related code in this module only so other modules do not need to depend on log4j at compile scope. See the comments of Log4jUtils and InternalLog4jUtils for more details.
Add a log4j.properties to the test jar of hbase-logging module, so for other sub modules we just need to depend on the test jar of hbase-logging module at test scope to output the log to console, without placing a log4j.properties in the test resources as they all (almost) have the same content. And this test module will not be included in the assembly tarball so it will not mess up the binary distribution.
Ban direct commons-logging dependency, and ban commons-logging and log4j imports in non-test code, to avoid mess up the downstream users logging framework. In hbase-logging module we do need to use log4j classes and the trick is to use full class name.
Add jcl-over-slf4j and jul-to-slf4j dependencies, as some of our dependencies use jcl or jul as logging framework, we should also redirect their log message to slf4j.
Added new metric to differentiate sink startup time from last OP applied time.
Original behaviour was to always set startup time to TimestampsOfLastAppliedOp, and always show it on “status ‘replication’” command, regardless if the sink ever applied any OP.
This was confusing, specially for scenarios where cluster was just acting as source, the output could lead to wrong interpretations about sink not applying edits or replication being stuck.
With the new metric, we now compare the two metrics values, assuming that if both are the same, there's never been any OP shipped to the given sink, so output would reflect it more clearly, to something as for example:
SINK: TimeStampStarted=Thu Dec 06 23:59:47 GMT 2018, Waiting for OPs...
HBase ships ZooKeeper 3.5.x. Was the EOL'd 3.4.x. 3.5.x client can talk to 3.4.x ensemble.
The ZooKeeper project has built a FAQ that documents known issues and work-arounds when upgrading existing deployments.
Add backoff. Avoid retrying every 100ms.
Runs ‘catalogjanitor_run’ and ‘hbck_chore_run’ inline with the loading of the ‘HBCK Report’ page.
Pass ‘?cache=true’ to skip inline invocation of ‘catalogjanitor_run’ and ‘hbck_chore_run’ drawing the page.
Introduced a general ‘local region’ at master side to store the procedure data, etc.
The hfile of this region will be stored on the root fs while the wal will be stored on the wal fs. This issue supercedes part of the code for HBASE-23326, as now we store the data in ‘MasterData’ directory instead of ‘MasterProcs’.
The old hfiles will be moved to the global hfile archived directory with the suffix $-masterlocalhfile-$. The wal files will be moved to the global old wal directory with the suffix $masterlocalwal$. The TimeToLiveMasterLocalStoreHFileCleaner and TimeToLiveMasterLocalStoreWALCleaner are configured by default for cleaning the old hfiles and wal files, and the default TTLs are both 7 days.
Relocate test-only REST RemoteHTable and RemoteAdmin from src/ to test/. And mark them as InterfaceAudience.Private.
Config key: hbase.regionserver.slowlog.systable.enabled Default value: false
This config can be enabled if hbase.regionserver.slowlog.buffer.enabled is already enabled. While hbase.regionserver.slowlog.buffer.enabled ensures that any slow/large RPC logs with complete details are written to ring buffer available at each RegionServer, hbase.regionserver.slowlog.systable.enabled would ensure that all such logs are also persisted in new system table hbase:slowlog. Operator can scan hbase:slowlog with filters to retrieve specific attribute matching records and this table would be useful to capture historical performance of slowness of RPC calls with detailed analysis.
hbase:slowlog consists of single ColumnFamily info. info consists of multiple qualifiers similar to the attributes available to query as part of Admin API: get_slowlog_responses.
One example of a row from hbase:slowlog scan result (Attached a sample screenshot in the Jira) :
\x024\xC1\x06X\x81\xF6\xEC column=info:call_details, timestamp=2020-05-16T14:59:58.764Z, value=Scan(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$ScanRequest)
\x024\xC1\x06X\x81\xF6\xEC column=info:client_address, timestamp=2020-05-16T14:59:58.764Z, value=172.20.10.2:57348
\x024\xC1\x06X\x81\xF6\xEC column=info:method_name, timestamp=2020-05-16T14:59:58.764Z, value=Scan
\x024\xC1\x06X\x81\xF6\xEC column=info:param, timestamp=2020-05-16T14:59:58.764Z, value=region { type: REGION_NAME value: “cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.” } scan { a ttribute { name: “_isolationlevel_” value: “\x5C000” } start_row: “cccccccc” time_range { from: 0 to: 9223372036854775807 } max_versions: 1 cache_blocks: true max_result_size: 2 097152 caching: 2147483647 include_stop_row: false } number_of_rows: 2147483647 close_scanner: false client_handles_partials: true client_handles_heartbeats: true track_scan_met rics: false
\x024\xC1\x06X\x81\xF6\xEC column=info:processing_time, timestamp=2020-05-16T14:59:58.764Z, value=24
\x024\xC1\x06X\x81\xF6\xEC column=info:queue_time, timestamp=2020-05-16T14:59:58.764Z, value=0
\x024\xC1\x06X\x81\xF6\xEC column=info:region_name, timestamp=2020-05-16T14:59:58.764Z, value=cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.
\x024\xC1\x06X\x81\xF6\xEC column=info:response_size, timestamp=2020-05-16T14:59:58.764Z, value=211227
\x024\xC1\x06X\x81\xF6\xEC column=info:server_class, timestamp=2020-05-16T14:59:58.764Z, value=HRegionServer
\x024\xC1\x06X\x81\xF6\xEC column=info:start_time, timestamp=2020-05-16T14:59:58.764Z, value=1589640743932
\x024\xC1\x06X\x81\xF6\xEC column=info:type, timestamp=2020-05-16T14:59:58.764Z, value=ALL
\x024\xC1\x06X\x81\xF6\xEC column=info:username, timestamp=2020-05-16T14:59:58.764Z, value=vjasani
HBASE-24271 makes changes the the default conf/hbase-site.xml such that bin/hbase will run directly out of the binary tarball or a compiled source tree without any configuration modifications vs. Hadoop 2.8+. This changes our long-standing history of shipping no configured values in conf/hbase-site.xml, so existing processes that assume this file is empty of configuration properties may require attention.
Use Slf4jRequestLog instead of the log4j HttpRequestLogAppender in HttpServer.
The request log is disabled by default in conf/log4j.properties by the following lines:
log4j.category.http.requests=INFO,NullAppender log4j.additivity.http.requests=false
Change the ‘NullAppender’ to what ever you want if you want to enable request log.
Notice that, the logger name for master status http server is ‘http.requests.master’, and for region server it is ‘http.requests.regionserver’
Use a empty string to represent no column specified for deleteall in shell mode. useage:
deleteall ‘test’,‘r1’,'',12345 deleteall ‘test’, {ROWPREFIXFILTER => ‘prefix’}, '', 12345
Added a new hbase-asyncfs module to hold the asynchronous dfs output stream implementation for implementing WAL.
Make the logic of the versions chosen more reasonable for raw scan, to avoid lose result when using filter.
Moved to hbase-thirdparty 3.3.0.
This feature enables the HBase Web UI‘s to accept a ‘proxyuser’ via the HTTP Request’s query string. When the parameter `hbase.security.authentication.spnego.kerberos.proxyuser.enable` is set to `true` in hbase-site.xml (default is `false`), the HBase UI will attempt to impersonate the user specified by the query parameter “doAs”. This query parameter is checked case-insensitively. When this option is not provided, the user who executed the request is the “real” user and there is no ability to execute impersonation against the WebUI.
For example, if the user “bob” with Kerberos credentials executes a request against the WebUI with this feature enabled and a query string which includes `doAs=alice`, the HBase UI will treat this request as executed as `alice`, not `bob`.
The standard Hadoop proxyuser configuration properties to limit users who may impersonate others apply to this change (e.g. to enable `bob` to impersonate `alice`). See the Hadoop documentation for more information on how to configure these proxyuser rules.
bin/hbase will now dynamically select a Garbage Collector implementation based on the detected JVM version. JDKs 8,9,10 use -XX:+UseConcMarkSweepGC, while JDK11+ use -XX:+UseG1GC.
Notice a slight compatibility change. Previously, the garbage collector choice would always be appended to a user-provided value for HBASE_OPTS. As of this change, this setting will only be applied when HBASE_OPTS is unset. That means that operators who provide a value for this variable will now need to also specify the collector. This is especially important for those on JDK8, where the vm default GC is not the recommended ConcMarkSweep.
Default value: false Description: If value is true, RegionServer will abort batch requests of Put/Delete with number of rows in a batch operation exceeding threshold defined by value of config: hbase.rpc.rows.warning.threshold.
StochasticLoadBalancer functional improvement:
StochasticLoadBalancer would rebalance the cluster if there are any idle RegionServers in the cluster (RegionServer having no region), while other RegionServers have at least 1 region available.
user or admin can now use hbase shell > rename_rsgroup ‘oldname’, ‘newname’ to rename rsgroup.
Add hadoop-3.2.0 and hadoop-3.2.1 in hadoop check and when ‘--quick-hadoopcheck’ we will only check hadoop-3.2.1.
Notice that, for aligning the personality scripts across all the active branches, we will commit the patch to all active branches, but the hadoop-3.2.x support in hadoopcheck is only applied to branch-2.2+.
`-PrunSmallTests` now pass on JDK11 when using `-Phadoop.profile=3.0`.
Tests that fail because a process -- RegionServer or Master -- called System.exit, will now instead throw an exception.
Hadoop hosts have had their ulimit -u raised from 10000 to 30000 (per user, by INFRA). The Docker build container has had its limit raised from 10000 to 12500.
Support RSGroup renaming in core codebase. New API Admin#renameRSGroup(String, String) is introduced in 3.0.0.
The Canary tool now offers a WebUI when run in region mode (the default mode). It is enabled by default, and by default, it binds to 0.0.0.0:16050. This can be overridden by setting hbase.canary.info.bindAddress and hbase.canary.info.port. To disable entirely, set the port to -1.
Pass --threads=2 building on jenkins. It shortens nightly build times by about ~25%.
It works by running module build/test in parallel when dependencies allow. Upping the forkcount beyond the pom default of 0.25C would have us broach our CPU budget on jenkins when two modules are running in parallel (2 modules at 0.25% of CPU each makes 0.5C and on jenkins, hadoop nodes run two jenkins executors per host). Higher forkcounts also seems to threaten build stability.
For running tests locally, to go faster, up fork count.
$ x=“0.5C” ; mvn --threads=2 -Dsurefire.firstPartForkCount=$x -Dsurefire.secondPartForkCount=$x test -PrunAllTests
You could up the x from 0.5C to 1.0C but YMMV (On overcommitted hardware, tests start bombing out pretty soon after startup). You could try upping thread count but on occasion are likely to overcommit hardware.
Start docker with upped ulimit for nproc passing ‘--ulimit nproc=12500’. It was 10000, the default, but made it 12500. Then, set PROC_LIMIT in hbase-personality so when yetus runs, it is w/ the new 12500 value.
Pass -T2 to mvn. Makes it so we do two modules-at-a-time dependencies willing. Helps speed build and testing. Doubles the resource usage when running modules in parallel.
Master & RegionService now support refresh policy authorization defined in hbase-policy.xml without restarting service. To refresh policy, please execute hbase shell command: update_config or update_config_all after policy file updated and synced on all nodes.
This change modifies the default acquisition policy for the region's close lock in order to prevent observed starvation of close requests. The new boolean configuration parameter ‘hbase.regionserver.fair.region.close.lock’ controls the lock acquisition policy: if true, the lock is created in fair mode (default); if false, the lock is created in nonfair mode (the old default).
The PrimaryRegionCountSkewCostFunction for the StochasticLoadBalancer is only needed when the read replicas feature is enabled. With this change, that function now properly indicates that it is not needed when the read replica feature is off.
If this improvement is not available, operators with clusters that are not using the read replica feature should manually disable it by setting hbase.master.balancer.stochastic.primaryRegionCountCost to 0.0 in hbase-site.xml for all HBase Masters.
Now AsyncFSWAL can also be used against the directory which has EC enabled. Need to make sure you also make use of the hadoop 3.x client as the option is only available in hadoop 3.x.
Branches-2.3+ use maven 3.5.3 building. Older branches use 3.5.4 still.
Our ‘Build Artifacts’ have a machine directory under which we emit vitals on the host the build was run on. We used to emit the result of ‘ulimit -l’ as a file named ‘ulimit-l’. This has been hijacked to instead emit result of running ‘ulimit -a’ which includes stat on ulimit -l.
ColumnFamilyDescriptor new builder API:
/\*\*
\* Retain all versions for a given TTL(retentionInterval), and then only a specific number
\* of versions(versionAfterInterval) after that interval elapses.
\*
\* @param retentionInterval Retain all versions for this interval
\* @param versionAfterInterval Retain no of versions to retain after retentionInterval
\*/
public ModifyableColumnFamilyDescriptor setVersionsWithTimeToLive(
final int retentionInterval, final int versionAfterInterval)
org.apache.hadoop.hbase.types.PBType is marked as deprecated without any replacement. It will be moved to hbase-example module and marked as IA.Private in 3.0.0. This is a mistake as it should not be part of our public API. Users who depend on this class should just copy the code your own code base.
Expose file system level read metrics for RegionServer.
If the HBase RS runs on top of HDFS, calculate the aggregation of ReadStatistics of each HdfsFileInputStream. These metrics include: (1) total number of bytes read from HDFS. (2) total number of bytes read from local DataNode. (3) total number of bytes read locally through short-circuit read. (4) total number of bytes read locally through zero-copy read.
Because HDFS ReadStatistics is calculated per input stream, it is not feasible to update the aggregated number in real time. Instead, the metrics are updated when an input stream is closed.
Admin can determine which tables go to which rsgroup by script (setting hbase.rsgroup.table.mapping.script with local filystem path) on Master side which aims to lighten the burden of admin operations. Note, since HBase 3+, rsgroup can be specified in TableDescriptor as well, if clients specify this, master will skip the determination from script.
Here is a simple example of script: {code}
#!/bin/bash namespace=$1 tablename=$2 if [[ $namespace == test ]]; then echo test elif [[ $tablename == *foo* ]]; then echo other else echo default fi {code}
MiniZKCluster now puts up its standalone node listening on loopback/127.0.0.1 rather than “localhost”.
Bumped hadoop-two.version to 2.10.0, which means we will drop the support for hadoop-2.8.x and hadoop-2.9.x.
Change timestamp display to be ISO8601 when toString on Cell and outputting in shell....
User used to see....
column=table:state, timestamp=1583967620343 .....
... but now sees:
column=table:state, timestamp=2020-03-11T23:00:20.343Z ....
merge_region shell command can now be used to merge more than 2 regions as well. It takes a list of regions as comma separated values or as an array of regions, and not just 2 regions. The full regionnames and encoded regionnames are continued to be accepted.
Rebuild our Dockerfile with support for multiple JDK versions. Use multiple stages in the Jenkinsfile instead of yetus's multijdk because of YETUS-953. Run those multiple stages in parallel to speed up results.
Note that multiple stages means multiple Yetus invocations means multiple comments on the PreCommit. This should become more obvious to users once we can make use of GitHub Checks API, HBASE-23902.
get_slowlog_responses and clear_slowlog_responses are used to retrieve and clear slow RPC logs from RingBuffer maintained by RegionServers.
New Admin APIs:
List<SlowLogRecord> getSlowLogResponses(final Set<ServerName> serverNames, final SlowLogQueryFilter slowLogQueryFilter) throws IOException;
List<Boolean> clearSlowLogResponses(final Set<ServerName> serverNames) throws IOException;
Configs:
Default 256
Default false
For more details, please look into “Get Slow Response Log from shell” section from HBase book.
Down the flakey re-rerun fork count from 1.0C -- i.e. a fork per CPU -- to 0.25C. On a recent run, the machine had 16 cores. 0.25 is 4 cores. We'd hardcoded fork count at 3 previous to changes made by parent.
Add a checkAndMutate(row, filter) method in the AsyncTable interface and the Table interface.
This method atomically checks if the row matches the specified filter. If it does, it adds the Put/Delete/RowMutations.
This is a fluent style API, the code is like:
For Table interface: {code} table.checkAndMutate(row, filter).thenPut(put); {code}
For AsyncTable interface: {code} table.checkAndMutate(row, filter).thenPut(put) .thenAccept(succ -> { if (succ) { System.out.println(“Check and put succeeded”); } else { System.out.println(“Check and put failed”); } }); {code}
The Jira Precommit job (https://builds.apache.org/job/PreCommit-HBASE-Build/) will now look for a file within the source tree (dev-support/jenkins_precommit_jira_yetus.sh) instead of depending on a script section embedded in the job.
Changed flakey list reporting to show 5 rather than 10 items. Also changed the second and first part fort counts to be 1C rather than hardcoded 3.
HBASE-23554 | Major | Encoded regionname to regionname utility
Adds shell command regioninfo:
hbase(main):001:0\> regioninfo '0e6aa5c19ae2b2627649dc7708ce27d0'
{ENCODED =\> 0e6aa5c19ae2b2627649dc7708ce27d0, NAME =\> 'TestTable,,1575941375972.0e6aa5c19ae2b2627649dc7708ce27d0.', STARTKEY =\> '', ENDKEY =\> '00000000000000000000299441'}
Took 0.4737 seconds
This JIRA adds a new configuration - `hbase.rs.cachecompactedblocksonwrite.threshold`. This configuration is the maximum total size (in bytes) of the compacted files below which the configuration `hbase.rs.cachecompactedblocksonwrite` is honoured. If the total size of the compacted fies exceeds this threshold, even when `hbase.rs.cachecompactedblocksonwrite` is enabled, the data blocks are not cached. Caching index and bloom blocks is not affected by this configuration (user configuration is always honoured).
Default value of this configuration is Long.MAX_VALUE. This means whatever the total size of the compacted files, it wil be cached.
Implements authorization for the HBase Web UI by limiting access to certain endpoints which could be used to extract sensitive information from HBase.
Access to these restricted endpoints can be limited to a group of administrators, identified either by a list of users (hbase.security.authentication.spnego.admin.users) or by a list of groups (hbase.security.authentication.spnego.admin.groups). By default, neither of these values are set which will preserve backwards compatibility (allowing all authenticated users to access all endpoints).
Further, users who have sensitive information in the HBase service configuration can set hbase.security.authentication.ui.config.protected to true which will treat the configuration endpoint as a protected, admin-only resource. By default, all authenticated users may access the configuration endpoint.
Enables master based registry as the default registry used by clients to fetch connection metadata. Refer to the section “Master Registry” in the client documentation for more details and advantages of this implementation over the default Zookeeper based registry.
Configuration parameter that controls the registry in use: hbase.client.registry.impl
Where to set this: HBase client configuration (hbase-site.xml)
Possible values:
org.apache.hadoop.hbase.client.ZKConnectionRegistry (For ZK based registry implementation)org.apache.hadoop.hbase.client.MasterRegistry (New, for master based registry implementation)Notes on defaults:
This feature has been back ported to 2.3.0 and later releases. MasterRegistry can be enabled by setting the following client configuration.
<property> <name>hbase.client.registry.impl</name> <value>org.apache.hadoop.hbase.client.MasterRegistry</value> </property>
caffeine: 2.6.2 => 2.8.1 commons-codec: 1.10 => 1.13 commons-io: 2.5 => 2.6 disrupter: 3.3.6 => 3.4.2 httpcore: 4.4.6 => 4.4.13 jackson: 2.9.10 => 2.10.1 jackson.databind: 2.9.10.1 => 2.10.1 jetty: 9.3.27.v20190418 => 9.3.28.v20191105 protobuf.plugin: 0.5.0 => 0.6.1 zookeeper: 3.4.10 => 3.4.14 slf4j: 1.7.25 => 1.7.30 rat: 0.12 => 0.13 asciidoctor: 1.5.5 => 1.5.8 asciidoctor.pdf: 1.5.0-alpha.15 => 1.5.0-rc.2 error-prone: 2.3.3 => 2.3.4
This change introduces an internal abstraction layer which allows for new SASL-based authentication mechanisms to be used inside HBase services. All existing SASL-based authentication mechanism were ported to the new abstraction, making no external change in runtime semantics, client API, or RPC serialization format.
Developers familiar with extending HBase can implement authentication mechanism beyond simple Kerberos and DelegationTokens which authenticate HBase users against some other user database. HBase service authentication (Master to/from RegionServer) continue to operate solely over Kerberos.
Introduce a new hbase-assembly/src/main/assembly/hadoop-three-compat.xml for build with hadoop 3.x.
Add a new config to hbase-default.xml
<property> <name>hbase.procedure.store.region.hfilecleaner.plugins</name> <value>org.apache.hadoop.hbase.master.cleaner.TimeToLiveHFileCleaner</value> <description>A comma-separated list of BaseHFileCleanerDelegate invoked by the RegionProcedureStore HFileCleaner service. These HFiles cleaners are called in order, so put the cleaner that prunes the most files in front. To implement your own BaseHFileCleanerDelegate, just put it in HBase's classpath and add the fully qualified class name here. Always add the above default hfile cleaners in the list as they will be overwritten in hbase-site.xml.</description> </property>
It will share the same TTL with other HFileCleaners. And you can also implement your own cleaner and change this property to enable it.
Updated parent pom to Apache version 22.
This issues fixes an issue with Bulk Loading on installations with Kerberos enabled and more than a single RegionServer. When multiple tables are involved in hosting a table‘s regions which are being bulk-loaded into, all but the RegionServer hosting the table’s first Region will “leak” one DistributedFileSystem object onto the heap, never freeing that memory. Eventually, with enough bulk loads, this will create a situation for RegionServers where they have no free heap space and will either spend all time in JVM GC, lose their ZK session, or crash with an OutOfMemoryError.
The only mitigation for this issue is to periodically restart RegionServers. All earlier versions of HBase 2.x are subject to this issue (2.0.x, <=2.1.8, <=2.2.3)
Add a new feature to improve MTTR which have 3 steps to failover:
Compared to DLS(distributed log split), this feature will reduce region open time significantly.
Config hbase.wal.split.to.hfile to true to enable this featue.
Changed the logging in hbase-zookeeper to use built-in formatting
From the PR:
"Yes. The two create the same output... I just wrote a small test suite to increase my confidence on that. I generated many tens of millions of random byte patterns and compared the output of the two algorithms. They came back identical every time.
“Just in case any inquiring minds would like to know, there is no longer an encoding required when generating the strings. The JDK implementation specifically specifies that strings returned are StandardCharsets.ISO_8859_1. This does not change anything because UTF8 and ISO_8859 overlap for the limited character set (64 characters) the encoding uses.”
Set hbase.balancer.max.balancing to a int value which <=0 will disable region balance throttling.
If cacheOnWrite is enabled during flush or compaction, index and bloom blocks(with data blocks) would be automatically cached during write.
If a RegionServer reports a Region as OPEN in disagreement with Master's status on the Region, the Master now tells the RegionServer to silently close the Region.
Makes it so the recently added HBCKServerCrashProcedure -- the SCP that gets invoked when an operator schedules an SCP via hbck2 scheduleRecoveries command -- now works the same as SCP EXCEPT if master knows nothing of the scheduled servername. In this latter case, HBCKSCP will do a full scan of hbase:meta looking for instances of the passed servername. If any found it will attempt cleanup of hbase:meta references by reassigning any found OPEN or OPENING and by closing any in CLOSING state.
Used to fix instances of what the ‘HBCK Report’ page shows as ‘Unknown Servers’.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.HFileProcedurePrettyPrinter to run the tool.
RegionsRecoveryChore introduced as part of HBASE-22460 tries to reopen regions based on config: hbase.regions.recovery.store.file.ref.count. Region reopen needs to take into consideration all compacted away store files that belong to the region and not store files(non-compacted).
Fixed this bug as part of this Jira. Updated description for corresponding configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
Very large number of ref count on a compacted store file indicates that it is a ref leak on that object(compacted store file). Such files can not be removed after it is invalidated via compaction. Only way to recover in such scenario is to reopen the region which can release all resources, like the refcount, leases, etc. This config represents Store files Ref Count threshold value considered for reopening regions. Any region with compacted store files ref count > this value would be eligible for reopening by master. Here, we get the max refCount among all refCounts on all compacted away store files that belong to a particular region. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.WALProcedurePrettyPrinter to run the tool.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.RegionProcedureStorePerformanceEvaluation to run the tool.
Use a region based procedure store to replace the old customized WAL based procedure store. The procedure data migration is done automatically during upgrading. After upgrading, the MasterProcWALs directory will be deleted and a new MasterProc directory will be created. And notice that a region will still write WAL so we still have WAL files and they will be moved to the oldWALs directory. The file name is mostly like a normal WAL file, and the only difference is that it is ended with “$masterproc$”.
Bumped surefire plugin to 3.0.0-M4
Now AsyncFSWAL also supports Durability.FSYNC_WAL.
The configuration ‘hbase.rs.cacheblocksonwrite’ was used to enable caching the blocks on write. But purposefully we were not caching the blocks when we do compaction (since it may be very aggressive) as the caching happens as and when the writer completes a block. In cloud environments since they have bigger sized caches - though they try to enable ‘hbase.rs.prefetchblocksonopen’ (non - aggressive way of caching the blocks proactively on reader creation) it does not help them because it takes time to cache the compacted blocks. This feature creates a new configuration ‘hbase.rs.cachecompactedblocksonwrite’ which when set to ‘true’ will enable the blocks created out of compaction. Remember that since it is aggressive caching the user should be having enough cache space - if not it may lead to other active blocks getting evicted. From the shell this can be enabled by using the option per Column Family also by using the below format {code} create ‘t1’, ‘f1’, {NUMREGIONS => 15, SPLITALGO => ‘HexStringSplit’, CONFIGURATION => {‘hbase.rs.cachecompactedblocksonwrite’ => ‘true’}} {code}
Users of the MOB feature can now use the mobrefs utility to get statistics about data in the MOB system and verify the health of backing files on HDFS.
HADOOP_CLASSPATH=/etc/hbase/conf:$(hbase mapredcp) yarn jar \
/some/path/to/hbase-shaded-mapreduce.jar mobrefs mobrefs-report-output some_table foo
See javadocs of the class MobRefReporter for more details.
the reference guide has added some information about MOB internals and troubleshooting.
The reference guide now includes a walk through of disabling the MOB feature if needed while maintaining availability.
Fixed unbalanced braces in string representation within HBase shell
The default rpc timeout for ReplicationSourceShipper#shipEdits is 60s, when bulkload replication enabled, timeout exception may be occurred. Now we can conf the timeout value through replication.source.shipedits.timeout, and it’s adaptive.
The newer HBase Thrift SPNEGO configs should not be required. The hbase.thrift.spnego.keytab.file and hbase.thrift.spnego.principal configs will fall back to the hbase.thrift.keytab.file and hbase.thrift.kerberos.principal original configs. The older configs will log a deprecation warning. It is preferred to new the newer SPNEGO configurations.
With BinaryComponentCompartor applications will be able to design diverse and powerful set of filters for rows and columns. See https://issues.apache.org/jira/browse/HBASE-22969 for example. In general, the comparator can be used with any filter taking ByteArrayComparable. As of now, following filters take ByteArrayComparable:
Adds a .editorconfig file with configurations populated by IntelliJ, based on our checkstyle configuration. There's lots of IntelliJ-specific configs in here that I assume are not replicated to Eclipse or Netbeans users. Any devs using those tools should push whatever updates they see fit, but please start with the checkstyle configs as the origin of truth.
An hbck2 scheduleRecoveries will run a subclass of ServerCrashProcedure which asks Master what Regions were on the dead Server but it will also do a hbase:meta table scan to see if any vestiges of the old Server remain (for the case where an SCP failed mid-point leaving references in place or where Master and hbase:meta deviated in accounting).
If holes in hbase:meta, hbck2 fixMeta now will update Master in-memory state so you do not need to restart master just so you can assign the new hole-bridging regions.
hbck2 scheduleRecoveries will now run a SCP that also looks in hbase:meta for any references to the scheduled server -- not just consult Master in-memory state -- just in case vestiges of the server are leftover in hbase:meta
HBase internal chores now log a moving average of how long execution of each chore takes at INFO level for the logger org.apache.hadoop.hbase.ScheduledChore.
Such messages will happen at most once per five minutes.
CleanerChore delegate initialization is now logged at INFO level instead of DEBUG
The ‘Procedures & Locks’ tab in Master UI only displays problematic Procedures now (RUNNABLE, WAITING-TIMEOUT, etc.). It no longer notes procedures whose state is SUCCESS.
the Apache HBase REST Proxy now uses Jackson Databind version 2.9.10.1 to address the following CVEs
Users of prior releases with Jackson Databind 2.9.10 are advised to either upgrade to this release or to upgrade their local Jackson Databind jar directly.
The MOB compaction process in the HBase Master now logs more about its activity.
In the event that you run into the problems described in HBASE-22075, there is a new HFileCleanerDelegate that will stop all removal of MOB hfiles from the archive area. It can be configured by adding org.apache.hadoop.hbase.mob.ManualMobMaintHFileCleaner to the list configured for hbase.master.hfilecleaner.plugins. This new cleaner delegate will cause your archive area to grow unbounded; you will have to manually prune files which may be prohibitively complex. Consider if your use case will allow you to mitigate by disabling mob compactions instead.
Caveats:
%hbase_root%/archive that accumulates files over time. You will have to determine which of these files are safe or not to remove.TRACE logging, the CleanerChore logger will include archive retention decision justifications.Adds per-user metrics for reads/writes to each RegionServer. These metrics are exported by default. hbase.regionserver.user.metrics.enabled can be used to disable the feature if desired for any reason.
Leaked store files can not be removed even after it is invalidated via compaction. A reasonable mitigation for a reader reference leak would be a fast reopen of the region on the same server.
Configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
This config represents Store files Ref Count threshold value considered for reopening regions. Any region with store files ref count > this value would be eligible for reopening by master. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Added a comment to make clear that read/write success counts are tallying column family success counts, not region success counts.
Additionally, the region read and write latencies previously only stored the latencies of the last column family of the region reads/writes. This has been fixed by using a map of each region to a list of read and write latency values.
Changes the message on the FNFE exception thrown when the file a Reference points to is missing; the message now includes detail on Reference as well as pointed-to file so can connect how FNFE relates to region open.
Use ‘totalRowActionRequestCount’ to calculate QPS on web UI.
Downstream users who wish to programmatically check the health of their HBase cluster may now rely on a public interface derived from the previously private implementation of the canary cli tool. The interface is named Canary and can be found in the user facing javadocs.
Downstream users who previously relied on the invoking the canary via the Java classname (either on the command line or programmatically) will need to change how they do so because the non-public implementation has moved.
Since 2.0.0，when one regionserver crashed and back online again, AssignmentManager will retain the region locations and try assign the regions to this regionserver(same host:port with the crashed one) again. But for 1.x.x, the behavior is round-robin assignment for the regions belong to the crashed regionserver. This jira change the “retain” assignment to round-robin assignment, which is same with 1.x.x version. This change will make the failover faster and improve availability.
Remove backward compatibility from `truncate` and `truncate_preserve` shell commands. This means that these commands from HBase Clients are not compatible with pre-0.99 HBase clusters.
giving the region mover “unload” command a region server name that isn‘t recognized by the cluster results in a "I don’t know about that host" message instead of a NPE.
set log level to DEBUG if you'd like the region mover to log the set of region server names it got back from the cluster.
Added a new IOEngine type for Bucket cache ie Persistent memory. In order to use BC over pmem configure IOEngine as <property> <name>hbase.bucketcache.ioengine</name> <value> pmem:///path in persistent memory </value> </property>
By default, snapshot auto cleanup based on TTL would be enabled for any new cluster. At any point in time, if snapshot cleanup is supposed to be stopped due to some snapshot restore activity or any other reason, it is advisable to disable it using shell command: hbase> snapshot_cleanup_switch false
We can re-enable it using: hbase> snapshot_cleanup_switch true
We can query whether snapshot auto cleanup is enabled for cluster using: hbase> snapshot_cleanup_enabled
Adds fix of overlaps to the fixMeta hbck service method. Uses the bulk-merge facility. Merges a max of 10 at a time. Set hbase.master.metafixer.max.merge.count to higher if you want to do more than 10 in the one go.
This issue adds via its subtasks:
* An ‘HBCK Report’ page to the Master UI added by HBASE-22527+HBASE-22709+HBASE-22723+ (since 2.1.6, 2.2.1, 2.3.0). Lists consistency or anomalies found via new hbase:meta consistency checking extensions added to CatalogJanitor (holes, overlaps, bad servers) and by a new ‘HBCK chore’ that runs at a lesser periodicity that will note filesystem orphans and overlaps as well as the following conditions: ** Master thought this region opened, but no regionserver reported it. ** Master thought this region opened on Server1, but regionserver reported Server2 ** More than one regionservers reported opened this region Both chores can be triggered from the shell to regenerate ‘new’ reports. * Means of scheduling a ServerCrashProcedure (HBASE-21393). * An ‘offline’ hbase:meta rebuild (HBASE-22680). * Offline replace of hbase.version and hbase.id * Documentation on how to use completebulkload tool to ‘adopt’ orphaned data found by new HBCK2 ‘filesystem’ check (see below) and ‘HBCK chore’ (HBASE-22859) * A ‘holes’ and ‘overlaps’ fix that runs in the master that uses new bulk-merge facility to collapse many overlaps in the one go. * hbase-operator-tools HBCK2 client tool got a bunch of additions: ** A specialized ‘fix’ for the case where operators ran old hbck ‘offlinemeta’ repair and destroyed their hbase:meta; it ties together holes in meta with orphaned data in the fs (HBASE-22567) ** A ‘filesystem’ command that reports on orphan data as well as bad references and hlinks with a ‘fix’ for the latter two options (based on hbck1 facility updated). ** Adds back the ‘replication’ fix facility from hbck1 (HBASE-22717)
The compound result is that hbck2 is now in excess of hbck1 abilities. The provided functionality is disaggregated as per the hbck2 philosophy of providing ‘plumbing’ rather than ‘porcelain’ so there is work to do still adding fix-it playbooks, scripting across outages, and automation.
HBASE-21879 introduces a utility class (org.apache.hadoop.hbase.io.ByteBuffAllocator) used for allocating/freeing ByteBuffers from/to NIO ByteBuffer pool, when BucketCache enabled with file or mmap engine, we will use this ByteBuffer pool to avoid temp ByteBuffer allocation a lot.
Introduces hbtop that‘s a real-time monitoring tool for HBase like Unix’s top command. See the ref guide for the details: https://hbase.apache.org/book.html#hbtop
Before this issue, read path was 100% offheap when block is in the BucketCache. But if a cache miss, then the RS needs to read the block via an on-heap API which causes high young-GC pressure.
This issue adds reading the block via offheap even if reading the block from filesystem directly. It requires hadoop version(>=2.9.3) but can also work with older hadoop versions (all works but we continue to read block onheap). It also requires HBASE-21946 which is not yet in place as of this writing/hbase-2.3.0.
We have written a careful doc about the implementation, performance and practice here: https://docs.google.com/document/d/1xSy9axGxafoH-Qc17zbD2Bd--rWjjI00xTWQZ8ZwI_E/edit#heading=h.nch5d72p27ex
Extends StochasticLoadBalancer to support user-provided cost function. These are loaded in addition to the default set of cost functions. Custom function implementations must extend StochasticLoadBalancer$CostFunction. Enable any additional functions by placing them on the master class path and configuring hbase.master.balancer.stochastic.additionalCostFunctions with a comma-separated list of fully-qualified class names.
Replace the ForkJoinPool in CleanerChore by ThreadPoolExecutor which can limit the spawn thread size and avoid the master GC frequently. The replacement is an internal implementation in CleanerChore, so no config key change, the upstream users can just upgrade the hbase master without any other change.
Introduced a new config key for the snapshot taking/restoring operations at master side: hbase.master.executor.snapshot.threads, its default value is 3. means we can have 3 snapshot operations running at the same time.
Add serveral API in TimeRange class for avoiding using the deprecated TimeRange constructor: * TimeRange#from: Represents the time interval [minStamp, Long.MAX_VALUE) * TimeRange#until: Represents the time interval [0, maxStamp) * TimeRange#between: Represents the time interval [minStamp, maxStamp)
Provide a public method in MultiRowRangeFilter class to speed the requirement of filtering with multiple row prefixes, it will expand the row prefixes as multiple rowkey ranges by MultiRowRangeFilter, it's more efficient. {code} public MultiRowRangeFilter(byte[][] rowKeyPrefixes); {code}
Update the base docker image to ubuntu 18.04 for the find flaky tests jenkins job.
Adds a fixMeta method to hbck Service. Fixes holes in hbase:meta. Follow-up to fix overlaps. See HBASE-22567 also.
Follow-on is adding a client-side to hbase-operator-tools that can exploit this new addition (HBASE-22825)
Changes merge so you can merge more than two regions at a time. Currently only available inside HBase. HBASE-22827, a follow-on, is about exposing the facility in the Admin API (and then via the shell).
New shaded artifact for testing: hbase-shaded-testing-util.
After HBASE-22776, the steps to config user scan snapshot feature is as followings:
We found a critical bug which can lead to WAL corruption when Durability.ASYNC_WAL is used. The reason is that we release a ByteBuffer before actually persist the content into WAL file.
The problem maybe lead to several errors, for example, ArrayIndexOfOutBounds when replaying WAL. This is because that the ByteBuffer is reused by others.
ERROR org.apache.hadoop.hbase.executor.EventHandler: Caught throwable while processing event RS_LOG_REPLAY java.lang.ArrayIndexOutOfBoundsException: 18056 at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1365) at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1358) at org.apache.hadoop.hbase.PrivateCellUtil.matchingFamily(PrivateCellUtil.java:735) at org.apache.hadoop.hbase.CellUtil.matchingFamily(CellUtil.java:816) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEditFamily(WALEdit.java:143) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEdit(WALEdit.java:148) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:297) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:195) at org.apache.hadoop.hbase.regionserver.SplitLogWorker$1.exec(SplitLogWorker.java:100)
And may even cause segmentation fault and crash the JVM directly. You will see a hs_err_pidXXX.log file and usually the problem is SIGSEGV. This is usually because that the ByteBuffer has already been returned to the OS and used for other purpose.
The problem has been reported several times in the past and this time Wellington Ramos Chevreuil provided the full logs and deeply analyzed the logs so we can find the root cause. And Lijin Bin figured out that the problem may only happen when Durability.ASYNC_WAL is used. Thanks to them.
The problem only effects the 2.x releases, all users are highly recommand to upgrade to a release which has this fix in, especially that if you use Durability.ASYNC_WAL.
Add a new method runHbckChore in Hbck interface and a new shell cmd hbck_chore_run to request HBCK chore to run at master side.
Adds a “CatalogJanitor hbase:meta Consistency Issues” section to the new ‘HBCK Report’ page added by HBASE-22709. This section is empty unless the most recent CatalogJanitor scan turned up problems. If so, will show table of issues found.
When CatalogJanitor runs, it now checks for holes, overlaps, empty info:regioninfo columns and bad servers. Dumps findings into log. Follow-up adds report to new ‘HBCK Report’ linked off the Master UI.
NOTE: All features but the badserver check made it into branch-2.1 and branch-2.0 backports.
The misspelled BufferedMutatorParams.opertationTimeout method has been marked as deprecated, and will be removed in 4.0.0. Please use the BufferedMutatorParams.operationTimeout method instead.
If a table user scan snapshots of the table, please config the following table scheme attribute to make granted users' ACLs are added to hfiles: alter ‘t1’, CONFIGURATION => {‘hbase.user.scan.snapshot.enable’ => ‘true’}
This feature is enabled by default. And the hbck chore run per 60 minutes by default. You can config “hbase.master.hbck.checker.interval” to a value lesser than or equal to 0 for disabling the chore.
Notice: the config “hbase.master.hbck.checker.interval” was renamed to “hbase.master.hbck.chore.interval” in HBASE-22737.
The HFileCleaner will clean the empty directories under archive, but if enable user scan snaphot feature, the user ACLs are set at there directories, so please config the following cleaner to make the directories with user ACLs not be cleaned: hbase.master.hfilecleaner.plugins=org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclCleaner
Upgrade jackson databind dependency to 2.9.9.1 due to CVEs
https://nvd.nist.gov/vuln/detail/CVE-2019-12814
https://nvd.nist.gov/vuln/detail/CVE-2019-12384
Add a new master web UI to show the potentially problematic opened regions. There are three case:
Feature: Take a Snapshot With TTL for auto-cleanup
Attribute:
Configs:
Default Snapshot TTL:
If Snapshot cleanup is supposed to be stopped due to some snapshot restore activity, disable it with config:
For more details, see the section “Take a Snapshot With TTL” in the HBase Reference Guide.
The config point “hbase.offheapcache.minblocksize” was wrong and is now deprecated. The new config point is “hbase.blockcache.minblocksize”.
OfflineMetaRepair is no longer supported in HBase-2+. Please refer to https://hbase.apache.org/book.html#HBCK2
This tool is deprecated in 2.x and will be removed in 3.0.
Mark the Hbck#scheduleServerCrashProcedure(List<HBaseProtos.ServerName> serverNames) as deprecated. Use Hbck#scheduleServerCrashProcedures(List<ServerName> serverNames) instead.
In HBASE-20734 we moved the recovered.edits onto the wal file system but when constructing the directory we missed the BASE_NAMESPACE_DIR(‘data’). So when using the default config, you will find that there are lots of new directories at the same level with the ‘data’ directory.
In this issue, we add the BASE_NAMESPACE_DIR back, and also try our best to clean up the wrong directories. But we can only clean up the region level directories, so if you want a clean fs layout on HDFS you still need to manually delete the empty directories at the same level with ‘data’.
The effect versions are 2.2.0, 2.1.[1-5], 1.4.[8-10], 1.3.[3-5].
Add a coprocessor to set HDFS acls to make hbase granted users with READ permission have the access to scan snapshots. To use this feature, please make sure the HDFS config is set: dfs.namenode.acls.enabled=true fs.permissions.umask-mode=027
and set the HBase config: hbase.coprocessor.master.classes=“org.apache.hadoop.hbase.security.access.AccessController,org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclController” hbase.user.scan.snapshot.enable=true
hbase.regionserver.compaction.check.period is used for controlling how often the compaction checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
hbase.regionserver.flush.check.period is used for controlling how ofter the flush checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
When run with JDK11 HBase now uses more recent version of the jaxws reference implementation (v2.3.2).
Added completebulkload short name for BulkLoadHFilesTool to bin/hbase.
Change the default hadoop-3 version to 3.1.2. Drop the support for the releases which are effected by CVE-2018-8029, see this email https://lists.apache.org/thread.html/3d6831c3893cd27b6850aea2feff7d536888286d588e703c6ffd2e82@%3Cuser.hadoop.apache.org%3E
This change exposes the aggregate count of store reader references for a given store as ‘storeRefCount’ in region metrics and ClusterStatus.
The HBase “source checksum” now uses SHA512 instead of MD5.
The CellUtil.setTimestamp method changes to be an API with audience LimitedPrivate(COPROC) in HBase 3.0. With that designation the API should remain stable within a given minor release line, but may change between minor releases.
Previously, this method was deprecated in HBase 2.0 for removal in HBase 3.0. Deprecation messages in HBase 2.y releases have been updated to indicate the expected API audience change.
The access method was used to the HttpServerFunctionalTest class as a common place.
The class LossyCounting was unintentionally marked Public but was never intended to be part of our public API. This oversight has been corrected and LossyCounting is now marked as Private and going forward may be subject to additional breaking changes or removal without notice. If you have taken a dependency on this class we recommend cloning it locally into your project before upgrading to this release.
Warnings for level headings are corrected in the book for the HBase Incompatibilities section.
Add hadoop 3.0.3, 3.1.1 3.1.2 in our hadoop check jobs.
The DumpReplicationQueues tool will now list replication queues sorted in chronological order.
Support get|set LogLevel in secure(kerberized) environment.
Fixes a formatting issue in the administration section of the book, where listing indentation were a little bit off.
This change adds the new method listNamespaces to the Admin interface, which can be used to retrieve a list of the namespaces present in the schema as an unprivileged operation. Formerly the only available method for accomplishing this was listNamespaceDescriptors, which requires GLOBAL CREATE or ADMIN permissions.
Now the default hadoop-two.version has been changed to 2.8.5, and all hadoop versions before 2.8.2(exclude) will not be supported any more.
Removed extra + in HRegion, HStore and LoadIncrementalHFiles for branch-2 and HRegion and HStore for branch-1.
Updated metrics core from 3.2.1 to 3.2.6.
The rubocop definition for the maximum method length was set to 75.
Fixes the formatting of the “Voting on Release Candidates” to actually show the quote and code formatting of the RAT check.
The rubocop configuration in the hbase-shell module now allows a line length with 100 characters, instead of 80 as before. For everything before 2.1.5 this change introduces rubocop itself.
This change adds new conditions for rolling the WAL for when syncs on the HDFS writer pipeline are perceived to be slow.
As before the configuration parameter hbase.regionserver.wal.slowsync.ms sets the slow sync warning threshold.
If we encounter hbase.regionserver.wal.slowsync.roll.threshold number of slow syncs (default 100) within the interval defined by hbase.regionserver.wal.slowsync.roll.interval.ms (default 1 minute), we will request a WAL roll.
Or, if the time for any sync exceeds the threshold set by hbase.regionserver.wal.roll.on.sync.ms (default 10 seconds) we will request a WAL roll immediately.
Operators can monitor how often these new thresholds result in a WAL roll by looking at newly added metrics to the WAL related metric group: * slowSyncRollRequest - How many times a roll was requested due to sync too slow on the write pipeline.
Additionally, as a part of this change there are also additional metrics for existing reasons for a WAL roll: * errorRollRequest - How many times a roll was requested due to I/O or other errors. * sizeRollRequest - How many times a roll was requested due to file size roll threshold.
MajorCompactorTTL Tool allows to compact all regions in a table that have been TTLed out. This saves space on DFS and is useful for tables which are similar to time series data. This is typically scheduled to run frequently (say via cron) to cleanup old data on an ongoing basis.
RSGroupMajorCompactionTTL tool is similar to MajorCompactorTTL but runs at a region server group level. If multiple tables in an rsgroup are similar to time-series data, then it runs a single command to clean them up. As more tables are added/removed from rsgroup, it's easy to have a single command to take care of all of them.
This change allows the system and superusers to initiate compactions, even when a space quota violation policy disallows compactions from happening. The original intent behind disallowing of compactions was to prevent end-user compactions from creating undue I/O load, not disallowing *any* compaction in the system.
Maven project integration for Eclipse has been isolated into a maven profile to ensure it only is active when in an Eclipse project.
Things should continue to behave the same for Eclipse users. If something should go wrong folks should manually activate the eclipse-specific profile.
Deprecated Preemptive Fail Fast related constants in HConstants, the support of this feature will be removed in 3.0.0 so use these constants will have no effect for 3.0.0+ releases. And the constants will be kept till 4.0.0.
Users can use ‘hbase.client.perserver.requests.threshold’ to control the number of concurrent requests to the same region server. Please see the release note of HBASE-16388 for more details.
Adds new configuration hbase.client.failure.map.cleanup.interval which defaults to ten minutes.
The default version of JRuby shipped with HBase has been updated to the JRuby 9.1.17.0 release.
For details on changes see the release notes for JRuby 9.1.17.0
Add below method in Table interface:
RegionLocator getRegionLocator() throws IOException;
Add below methods in AsyncTable interface:
AsyncTableRegionLocator getRegionLocator(); CompletableFuture<TableDescriptor> getDescriptor();
LruBlockCache uses the Segmented LRU (SLRU) policy to capture frequency and recency of the working set. It achieves concurrency by using an O(n) background thread to prioritize the entries and evict. Accessing an entry is O(1) by a hash table lookup, recording its logical access time, and setting a frequency flag. A write is performed in O(1) time by updating the hash table and triggering an async eviction thread. This provides ideal concurrency and minimizes the latencies by penalizing the thread instead of the caller. However the policy does not age the frequencies and may not be resilient to various workload patterns.
This change introduces a new L1 policy, TinyLfuBlockCache, which records the frequency in a counting sketch, ages periodically by halving the counters, and orders entries by SLRU. An entry is discarded by comparing the frequency of the new arrival to the SLRU's victim, and keeping the one with the highest frequency. This allows the operations to be performed in O(1) time and, though the use of a compact sketch, a much larger history is retained beyond the current working set. In a variety of real world traces the policy had near optimal hit rates.
New configuration variable hfile.block.cache.policy sets the eviction policy for the L1 block cache. The default is “LRU” (LruBlockCache). Set to “TinyLFU” to use TinyLfuBlockCache instead.
Introduced
Future<Void> createTableAsync(TableDescriptor);
Introduced these methods: void move(byte[]); void move(byte[], ServerName); Future<Void> splitRegionAsync(byte[]);
These methods are deprecated: void move(byte[], byte[])
Add a new jenkins file for running pre commit check for GitHub PR.
Add cloneSnapshot/restoreSnapshot with acl methods in AsyncAdmin.
When insufficient permissions, you now get:
HTTP/1.1 403 Forbidden
on the HTTP side, and in the message
Forbidden org.apache.hadoop.hbase.security.AccessDeniedException: org.apache.hadoop.hbase.security.AccessDeniedException: Insufficient permissions for user ‘myuser',action: get, tableName:mytable, family:cf. at org.apache.ranger.authorization.hbase.RangerAuthorizationCoprocessor.authorizeAccess(RangerAuthorizationCoprocessor.java:547) and the rest of the ADE stack
Now we will sort the javac WARNING/ERROR before generating diff in pre-commit so we can get a stable output for the error prone. The downside is that we just sort the output lexicographically so the line number will also be sorted lexicographically, which is a bit strange to human.
Exposes a new configuration property “zookeeper.multi.max.size” which dictates the maximum size of deletes that HBase will make to ZooKeeper in a single RPC. This property defaults to 1MB, which should fall beneath the default ZooKeeper limit of 2MB, controlled by “jute.maxbuffer”.
Fixed awkward dependency issue that prevented site building.
HBase 2.1.4 shipped with an early version of this fix that incorrectly altered the libraries included in our binary assembly for using Apache Hadoop 2.7 (the current build default Hadoop version for 2.1.z). For folks running out of the box against a Hadoop 2.7 cluster (or folks who skip the installation step of replacing the bundled Hadoop libraries) this will result in a failure at Region Server startup due to a missing class definition. e.g.:
2019-03-27 09:02:05,779 ERROR [main] regionserver.HRegionServer: Failed construction RegionServer java.lang.NoClassDefFoundError: org/apache/htrace/SamplerBuilder at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:644) at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:628) at org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:149) at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2667) at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:93) at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2701) at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2683) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:372) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:171) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:356) at org.apache.hadoop.fs.Path.getFileSystem(Path.java:295) at org.apache.hadoop.hbase.util.CommonFSUtils.getRootDir(CommonFSUtils.java:362) at org.apache.hadoop.hbase.util.CommonFSUtils.isValidWALRootDir(CommonFSUtils.java:411) at org.apache.hadoop.hbase.util.CommonFSUtils.getWALRootDir(CommonFSUtils.java:387) at org.apache.hadoop.hbase.regionserver.HRegionServer.initializeFileSystem(HRegionServer.java:704) at org.apache.hadoop.hbase.regionserver.HRegionServer.<init>(HRegionServer.java:613) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.hadoop.hbase.regionserver.HRegionServer.constructRegionServer(HRegionServer.java:3029) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.start(HRegionServerCommandLine.java:63) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.run(HRegionServerCommandLine.java:87) at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) at org.apache.hadoop.hbase.util.ServerCommandLine.doMain(ServerCommandLine.java:149) at org.apache.hadoop.hbase.regionserver.HRegionServer.main(HRegionServer.java:3047) Caused by: java.lang.ClassNotFoundException: org.apache.htrace.SamplerBuilder at java.net.URLClassLoader.findClass(URLClassLoader.java:381) at java.lang.ClassLoader.loadClass(ClassLoader.java:424) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349) at java.lang.ClassLoader.loadClass(ClassLoader.java:357) ... 26 more
Workaround via any one of the following:
hadoop executable is in the PATH seen at Region Server startup and that you are not using the HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP bypass.htrace-core-3.1.0-incubating.jar available to all Region Servers via the HBASE_CLASSPATH environment variable.htrace-core-3.1.0-incubating.jar available to all Region Servers by copying it into the directory ${HBASE_HOME}/lib/client-facing-thirdparty/.Add a listTableDescriptors(List<TableName>) method in the AsyncAdmin interface, to align with the Admin interface.
Deprecate Admin.deleteSnapshot(byte[]), please use the String version instead.
Add a mergeRegionsAsync(byte[][], boolean) method in the AsyncAdmin interface.
Instead of using assert, now we will throw IllegalArgumentException when you want to merge less than 2 regions at client side. And also, at master side, instead of using assert, now we will throw DoNotRetryIOException if you want merge more than 2 regions, since we only support merging two regions at once for now.
Add drainXXX parameter for balancerSwitch/splitSwitch/mergeSwitch methods in the AsyncAdmin interface, which has the same meaning with the synchronous parameter for these methods in the Admin interface.
As of HBase 3.0, the ByteBufferUtils class is now marked as a Private API for internal project use only. Downstream users are advised that it no longer has any compatibility promises across releases.
As of earlier HBase release lines the class is now marked as deprecated to call attention to this planned transition.
bulkload (HFileOutputFormat2) support config the compression on client ,you can set the job configuration “hbase.mapreduce.hfileoutputformat.compression” override the auto-detection of the target table's compression
Add a cloneSnapshotAsync method with restoreAcl parameter. Deprecated restoreSnapshotAsync method as it just ignores the failsafe configuration. Make snapshotAsync method returns a Future<Void>. Deprecated the snapshot related methods which take a ‘byte[]’ as the snapshot name. Use default methods to reduce the code base for implementation classes.
Deprecated AsyncTable.isTableAvailable(TableName, byte[][]).
After HBASE-21871, we can specify a peer table name with --peerTableName in VerifyReplication tool like the following: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable 5 TestTable
In addition, we can compare any 2 tables in any remote clusters with specifying both peerId and --peerTableName.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable zk1,zk2,zk3:2181/hbase TestTable
Adds below flush, split, and compaction metrics
HBASE-21481 improves the quality of access control, by strengthening the protection of super users's privileges.
Now we have four types of RIT procedure metrics, assign, unassign, move, reopen. The meaning of assign/unassign is changed, as we will not increase the unassign metric and then the assign metric when moving a region. Also introduced two new procedure metrics, open and close, which are used to track the open/close region calls to region server. We may send open/close multiple times to finish a RIT since we may retry multiple times.
Problem: This is an old problem since HBASE-2231. The compaction event marker was only writed to WAL. But after flush, the WAL may be archived, which means an useful compaction event marker be deleted, too. So the compacted store files cannot be archived when region open and replay WAL.
Solution: After this jira, the compaction event tracker will be writed to HFile. When region open and load store files, read the compaction evnet tracker from HFile and archive the compacted store files which still exist.
HBase contains two quota scopes: MACHINE and CLUSTER. Before this patch, set quota operations did not expose scope option to client api and use MACHINE as default, CLUSTER scope can not be set and used. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’
This issue implements CLUSTER scope in a simple way: For user, namespace, user over namespace quota, use [ClusterLimit / RSNum] as machine limit. For table and user over table quota, use [ClusterLimit / TotalTableRegionNum * MachineTableRegionNum] as machine limit. After this patch, user can set CLUSTER scope quota, but MACHINE is still default if user ignore scope. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’ set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => MACHINE set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => CLUSTER
Change spotbugs version to 3.1.11.
This modifies “status ‘replication’” output, fixing inconsistencies on the reporting times and ages of last shipped edits, as well as wrong calculation of replication lags.
It also introduces additional info for each recovery queue, which was not accounted by this command before.
The new output for “status ‘replication’” command is explained in details below: a) Source started, target stopped, no edits arrived on source yet: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ... b) Source started, target stopped, add edit on source: ... Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:21:00 GMT 2018, Replication Lag=2459 ... c) Source started, target stopped, edit added on source, restart source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 Recovered Queue: 1-hbase01.home,16020,1542784524057 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:23:00 GMT 2018, Replication Lag=201495 ... d) Source started, target stopped, add edit on source, restart source, add another edit on source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=6349 Recovered Queue: 1-hbase01.home,16020,1542782758742 No Ops shipped since last restart, SizeOfLogQueue=0, TimeStampOfLastArrivedInSource=Wed Nov 21 06:53:05 GMT 2018, Replication Lag=569394 ... e) Source started, target stopped, add edit on source, restart source, add another edit on source, start target: ... SOURCE: PeerID=1 Normal Queue: 1 AgeOfLastShippedOp=30000, TimeStampOfLastShippedOp=Wed Nov 21 07:07:58 GMT 2018, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=0 ... f) Source started, target stopped, add edit on source, restart source, restart target: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ...
Remove bloom filter type ROWPREFIX_DELIMITED. May add it back when find a better solution.
Support enable or disable exceed throttle quota. Exceed throttle quota means, user can over consume user/namespace/table quota if region server has additional available quota because other users don't consume at the same time. Use the following shell commands to enable/disable exceed throttle quota: enable_exceed_throttle_quota disable_exceed_throttle_quota There are two limits when enable exceed throttle quota:
Remove jackson dependencies from most hbase modules except hbase-rest, use shaded gson instead. The output json will be a bit different since jackson can use getter/setter, but gson will always use the fields.
Mark HConstants.META_QOS as deprecated. It is for internal use only, which is the highest priority. You should not try to set a priority greater than or equal to this value, although it is no harm but also useless.
This patch adds the ability to disable split and/or merge for a table (By default, split and merge are enabled for a table).
Allows shell to set Scan options previously not exposed. See additions as part of the scan help by typing following hbase shell:
hbase> help ‘scan’
We can specify peerQuorumAddress instead of peerId in VerifyReplication tool. So it no longer requires peerId to be setup when using this tool.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication zk1,zk2,zk3:2181/hbase testTable
Introduce a VerifyWALEntriesReplicationEndpoint which replicates nothing but only verifies if all the cells are valid. It can be used to capture bugs for writing WAL, as most times we will not read the WALs again after writing it if there are no region server crashes.
Introduced an new config key in this issue: hbase.regionserver.inmemory.compaction.pool.size. the default value would be 10. you can configure this to set the pool size of in-memory compaction pool. Note that all memstores in one region server will share the same pool, so if you have many regions in one region server, you need to set this larger to compact faster for better read performance.
Make StoppedRpcClientException extend DoNotRetryIOException.
To implement user permission control in Precedure V2, move grant and revoke method from AccessController to master firstly. Mark AccessController#grant and AccessController#revoke as deprecated and please use Admin#grant and Admin#revoke instead.
IMPORTANT: Due to security issues, all users who use hbase thrift should avoid using releases which do not have this fix.
The effect releases are: 2.1.x: 2.1.2 and below 2.0.x: 2.0.4 and below 1.x: 1.4.x and below
If you are using the effect releases above, please consider upgrading to a newer release ASAP.
For users who have configured hbase.bucketcache.ioengine with either the file:, files:, or mmap: prefix, and configured it to be persistent via the hbase.bucketcache.persistent.path property, the serialization format of the bucket cache has changed between versions. The old state will not be read during startup, and there is currently no migration path. The impact is expected to be minimal, however, since the cache will rebuild over time as access patterns dictate.
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Update our package version numbers throughout the Dockerfiles to be pinned to their epic:upstream-version components only. Previously we'd specify the full debian package version number, including the debian-revision. This lead to instability as debian packaging details changed. See also man deb-version
Adds a new metric where we collect the number of read requests (tracked per row) whether the row was fetched completely from memstore or it was pulled from files and memstore. The metric is now collected under the mbean for Tables and under the mbean for regions. Under table mbean ie.- ‘name": "Hadoop:service=HBase,name=RegionServer,sub=Tables’ The new metrics will be listed as {code} “Namespace_default_table_t3_columnfamily_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_columnfamily_f1_metric_mixedRowReadsCount”: 1, {code} Where the format is Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename>_table_<tableName>_columnfamily_<columnfamilyname>_metric_mixedRowReadsCount {code}
The same one under the region ie. “name”: “Hadoop:service=HBase,name=RegionServer,sub=Regions”, comes as {code} “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_memstoreOnlyRowReadsCount”: 5, “Namespace_default_table_t3_region_75a7846f4ac4a2805071a855f7d0dbdc_store_f1_metric_mixedRowReadsCount”: 1, {code} where Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_memstoreOnlyRowReadsCount Namespace_<namespacename_table_<tableName>_region_<regionName>_store_<storeName>_metric_mixedRowReadsCount This is also an aggregate against every store the number of reads that happened purely from the memstore or it was a mixed read that happened from memstore and file.
This adds [-h|-help] options to rowcounter. Passing either -h or -help will print rowcounter guide as below:
$hbase rowcounter -h
usage: hbase rowcounter <tablename> [options] [<column1> <column2>...] Options: --starttime=<arg> starting time filter to start counting rows from. --endtime=<arg> end time filter limit, to only count rows up to this timestamp. --range=<arg> [startKey],[endKey][;[startKey],[endKey]...]] --expectedCount=<arg> expected number of rows to be count. For performance, consider the following configuration properties: -Dhbase.client.scanner.caching=100 -Dmapreduce.map.speculative=false
CI coverage has been extended to include Hadoop 3.2.x for HBase 2.2+.
Adds being able to edit hbase:meta table schema. For example,
hbase(main):006:0> alter ‘hbase:meta’, {NAME => ‘info’, DATA_BLOCK_ENCODING => ‘ROW_INDEX_V1’} Updating all regions with the new schema... All regions updated. Done. Took 1.2138 seconds
You can even add columnfamilies. Howevert, you cannot delete any of the core hbase:meta column families such as ‘info’ and ‘table’.
This ticket summarizes significant improvements and expansion to the metrics surface area. Interested users should review the individual sub-tasks.
Adds backoff in ServerCrashProcedure wait on WAL split to complete if large backlog of files to split (Its possible to avoid SCP blocking, waiting on WALs to split if you use procedure-based splitting -- set ‘hbase.split.wal.zk.coordinated’ to false to enable procedure based wal splitting.)
Notice this has changed log level for mismatching row keys, originally those were being logged at INFO level, now it's logged at DEBUG level. This is consistent with the logging of mismatching cells. Also, for missing row keys, it now logs row key values in human readable format, making it more meaningful for operators troubleshooting mismatches.
Introduce a new config hbase.replication.drop.on.deleted.columnfamily, default is false. When config to true, the replication will drop the edits for columnfamily that has been deleted from the replication source and target.
This change extends the Normalizer with a handful of new configurations. The configuration points supported are:
hbase.normalizer.split.enabled Whether to split a region as part of normalization. Default: true.hbase.normalizer.merge.enabled Whether to merge a region as part of normalization. Default true.hbase.normalizer.min.region.count The minimum number of regions in a table to consider it for merge normalization. Default: 3.hbase.normalizer.merge.min_region_age.days The minimum age for a region to be considered for a merge, in days. Default: 3.hbase.normalizer.merge.min_region_size.mb The minimum size for a region to be considered for a merge, in whole MBs. Default: 1.Add a hbase-logging module, put the log4j related code in this module only so other modules do not need to depend on log4j at compile scope. See the comments of Log4jUtils and InternalLog4jUtils for more details.
Add a log4j.properties to the test jar of hbase-logging module, so for other sub modules we just need to depend on the test jar of hbase-logging module at test scope to output the log to console, without placing a log4j.properties in the test resources as they all (almost) have the same content. And this test module will not be included in the assembly tarball so it will not mess up the binary distribution.
Ban direct commons-logging dependency, and ban commons-logging and log4j imports in non-test code, to avoid mess up the downstream users logging framework. In hbase-logging module we do need to use log4j classes and the trick is to use full class name.
Add jcl-over-slf4j and jul-to-slf4j dependencies, as some of our dependencies use jcl or jul as logging framework, we should also redirect their log message to slf4j.
Added new metric to differentiate sink startup time from last OP applied time.
Original behaviour was to always set startup time to TimestampsOfLastAppliedOp, and always show it on “status ‘replication’” command, regardless if the sink ever applied any OP.
This was confusing, specially for scenarios where cluster was just acting as source, the output could lead to wrong interpretations about sink not applying edits or replication being stuck.
With the new metric, we now compare the two metrics values, assuming that if both are the same, there's never been any OP shipped to the given sink, so output would reflect it more clearly, to something as for example:
SINK: TimeStampStarted=Thu Dec 06 23:59:47 GMT 2018, Waiting for OPs...
HBase ships ZooKeeper 3.5.x. Was the EOL'd 3.4.x. 3.5.x client can talk to 3.4.x ensemble.
The ZooKeeper project has built a FAQ that documents known issues and work-arounds when upgrading existing deployments.
Add backoff. Avoid retrying every 100ms.
Runs ‘catalogjanitor_run’ and ‘hbck_chore_run’ inline with the loading of the ‘HBCK Report’ page.
Pass ‘?cache=true’ to skip inline invocation of ‘catalogjanitor_run’ and ‘hbck_chore_run’ drawing the page.
Introduced a general ‘local region’ at master side to store the procedure data, etc.
The hfile of this region will be stored on the root fs while the wal will be stored on the wal fs. This issue supercedes part of the code for HBASE-23326, as now we store the data in ‘MasterData’ directory instead of ‘MasterProcs’.
The old hfiles will be moved to the global hfile archived directory with the suffix $-masterlocalhfile-$. The wal files will be moved to the global old wal directory with the suffix $masterlocalwal$. The TimeToLiveMasterLocalStoreHFileCleaner and TimeToLiveMasterLocalStoreWALCleaner are configured by default for cleaning the old hfiles and wal files, and the default TTLs are both 7 days.
Relocate test-only REST RemoteHTable and RemoteAdmin from src/ to test/. And mark them as InterfaceAudience.Private.
Config key: hbase.regionserver.slowlog.systable.enabled Default value: false
This config can be enabled if hbase.regionserver.slowlog.buffer.enabled is already enabled. While hbase.regionserver.slowlog.buffer.enabled ensures that any slow/large RPC logs with complete details are written to ring buffer available at each RegionServer, hbase.regionserver.slowlog.systable.enabled would ensure that all such logs are also persisted in new system table hbase:slowlog. Operator can scan hbase:slowlog with filters to retrieve specific attribute matching records and this table would be useful to capture historical performance of slowness of RPC calls with detailed analysis.
hbase:slowlog consists of single ColumnFamily info. info consists of multiple qualifiers similar to the attributes available to query as part of Admin API: get_slowlog_responses.
One example of a row from hbase:slowlog scan result (Attached a sample screenshot in the Jira) :
\x024\xC1\x06X\x81\xF6\xEC column=info:call_details, timestamp=2020-05-16T14:59:58.764Z, value=Scan(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$ScanRequest)
\x024\xC1\x06X\x81\xF6\xEC column=info:client_address, timestamp=2020-05-16T14:59:58.764Z, value=172.20.10.2:57348
\x024\xC1\x06X\x81\xF6\xEC column=info:method_name, timestamp=2020-05-16T14:59:58.764Z, value=Scan
\x024\xC1\x06X\x81\xF6\xEC column=info:param, timestamp=2020-05-16T14:59:58.764Z, value=region { type: REGION_NAME value: “cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.” } scan { a ttribute { name: “_isolationlevel_” value: “\x5C000” } start_row: “cccccccc” time_range { from: 0 to: 9223372036854775807 } max_versions: 1 cache_blocks: true max_result_size: 2 097152 caching: 2147483647 include_stop_row: false } number_of_rows: 2147483647 close_scanner: false client_handles_partials: true client_handles_heartbeats: true track_scan_met rics: false
\x024\xC1\x06X\x81\xF6\xEC column=info:processing_time, timestamp=2020-05-16T14:59:58.764Z, value=24
\x024\xC1\x06X\x81\xF6\xEC column=info:queue_time, timestamp=2020-05-16T14:59:58.764Z, value=0
\x024\xC1\x06X\x81\xF6\xEC column=info:region_name, timestamp=2020-05-16T14:59:58.764Z, value=cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.
\x024\xC1\x06X\x81\xF6\xEC column=info:response_size, timestamp=2020-05-16T14:59:58.764Z, value=211227
\x024\xC1\x06X\x81\xF6\xEC column=info:server_class, timestamp=2020-05-16T14:59:58.764Z, value=HRegionServer
\x024\xC1\x06X\x81\xF6\xEC column=info:start_time, timestamp=2020-05-16T14:59:58.764Z, value=1589640743932
\x024\xC1\x06X\x81\xF6\xEC column=info:type, timestamp=2020-05-16T14:59:58.764Z, value=ALL
\x024\xC1\x06X\x81\xF6\xEC column=info:username, timestamp=2020-05-16T14:59:58.764Z, value=vjasani
HBASE-24271 makes changes the the default conf/hbase-site.xml such that bin/hbase will run directly out of the binary tarball or a compiled source tree without any configuration modifications vs. Hadoop 2.8+. This changes our long-standing history of shipping no configured values in conf/hbase-site.xml, so existing processes that assume this file is empty of configuration properties may require attention.
Use Slf4jRequestLog instead of the log4j HttpRequestLogAppender in HttpServer.
The request log is disabled by default in conf/log4j.properties by the following lines:
log4j.category.http.requests=INFO,NullAppender log4j.additivity.http.requests=false
Change the ‘NullAppender’ to what ever you want if you want to enable request log.
Notice that, the logger name for master status http server is ‘http.requests.master’, and for region server it is ‘http.requests.regionserver’
Use a empty string to represent no column specified for deleteall in shell mode. useage:
deleteall ‘test’,‘r1’,'',12345 deleteall ‘test’, {ROWPREFIXFILTER => ‘prefix’}, '', 12345
Added a new hbase-asyncfs module to hold the asynchronous dfs output stream implementation for implementing WAL.
Make the logic of the versions chosen more reasonable for raw scan, to avoid lose result when using filter.
Moved to hbase-thirdparty 3.3.0.
This feature enables the HBase Web UI‘s to accept a ‘proxyuser’ via the HTTP Request’s query string. When the parameter `hbase.security.authentication.spnego.kerberos.proxyuser.enable` is set to `true` in hbase-site.xml (default is `false`), the HBase UI will attempt to impersonate the user specified by the query parameter “doAs”. This query parameter is checked case-insensitively. When this option is not provided, the user who executed the request is the “real” user and there is no ability to execute impersonation against the WebUI.
For example, if the user “bob” with Kerberos credentials executes a request against the WebUI with this feature enabled and a query string which includes `doAs=alice`, the HBase UI will treat this request as executed as `alice`, not `bob`.
The standard Hadoop proxyuser configuration properties to limit users who may impersonate others apply to this change (e.g. to enable `bob` to impersonate `alice`). See the Hadoop documentation for more information on how to configure these proxyuser rules.
bin/hbase will now dynamically select a Garbage Collector implementation based on the detected JVM version. JDKs 8,9,10 use -XX:+UseConcMarkSweepGC, while JDK11+ use -XX:+UseG1GC.
Notice a slight compatibility change. Previously, the garbage collector choice would always be appended to a user-provided value for HBASE_OPTS. As of this change, this setting will only be applied when HBASE_OPTS is unset. That means that operators who provide a value for this variable will now need to also specify the collector. This is especially important for those on JDK8, where the vm default GC is not the recommended ConcMarkSweep.
Default value: false Description: If value is true, RegionServer will abort batch requests of Put/Delete with number of rows in a batch operation exceeding threshold defined by value of config: hbase.rpc.rows.warning.threshold.
StochasticLoadBalancer functional improvement:
StochasticLoadBalancer would rebalance the cluster if there are any idle RegionServers in the cluster (RegionServer having no region), while other RegionServers have at least 1 region available.
user or admin can now use hbase shell > rename_rsgroup ‘oldname’, ‘newname’ to rename rsgroup.
Add hadoop-3.2.0 and hadoop-3.2.1 in hadoop check and when ‘--quick-hadoopcheck’ we will only check hadoop-3.2.1.
Notice that, for aligning the personality scripts across all the active branches, we will commit the patch to all active branches, but the hadoop-3.2.x support in hadoopcheck is only applied to branch-2.2+.
`-PrunSmallTests` now pass on JDK11 when using `-Phadoop.profile=3.0`.
Tests that fail because a process -- RegionServer or Master -- called System.exit, will now instead throw an exception.
Hadoop hosts have had their ulimit -u raised from 10000 to 30000 (per user, by INFRA). The Docker build container has had its limit raised from 10000 to 12500.
Support RSGroup renaming in core codebase. New API Admin#renameRSGroup(String, String) is introduced in 3.0.0.
The Canary tool now offers a WebUI when run in region mode (the default mode). It is enabled by default, and by default, it binds to 0.0.0.0:16050. This can be overridden by setting hbase.canary.info.bindAddress and hbase.canary.info.port. To disable entirely, set the port to -1.
Pass --threads=2 building on jenkins. It shortens nightly build times by about ~25%.
It works by running module build/test in parallel when dependencies allow. Upping the forkcount beyond the pom default of 0.25C would have us broach our CPU budget on jenkins when two modules are running in parallel (2 modules at 0.25% of CPU each makes 0.5C and on jenkins, hadoop nodes run two jenkins executors per host). Higher forkcounts also seems to threaten build stability.
For running tests locally, to go faster, up fork count.
$ x=“0.5C” ; mvn --threads=2 -Dsurefire.firstPartForkCount=$x -Dsurefire.secondPartForkCount=$x test -PrunAllTests
You could up the x from 0.5C to 1.0C but YMMV (On overcommitted hardware, tests start bombing out pretty soon after startup). You could try upping thread count but on occasion are likely to overcommit hardware.
Start docker with upped ulimit for nproc passing ‘--ulimit nproc=12500’. It was 10000, the default, but made it 12500. Then, set PROC_LIMIT in hbase-personality so when yetus runs, it is w/ the new 12500 value.
Pass -T2 to mvn. Makes it so we do two modules-at-a-time dependencies willing. Helps speed build and testing. Doubles the resource usage when running modules in parallel.
Master & RegionService now support refresh policy authorization defined in hbase-policy.xml without restarting service. To refresh policy, please execute hbase shell command: update_config or update_config_all after policy file updated and synced on all nodes.
This change modifies the default acquisition policy for the region's close lock in order to prevent observed starvation of close requests. The new boolean configuration parameter ‘hbase.regionserver.fair.region.close.lock’ controls the lock acquisition policy: if true, the lock is created in fair mode (default); if false, the lock is created in nonfair mode (the old default).
The PrimaryRegionCountSkewCostFunction for the StochasticLoadBalancer is only needed when the read replicas feature is enabled. With this change, that function now properly indicates that it is not needed when the read replica feature is off.
If this improvement is not available, operators with clusters that are not using the read replica feature should manually disable it by setting hbase.master.balancer.stochastic.primaryRegionCountCost to 0.0 in hbase-site.xml for all HBase Masters.
Now AsyncFSWAL can also be used against the directory which has EC enabled. Need to make sure you also make use of the hadoop 3.x client as the option is only available in hadoop 3.x.
Branches-2.3+ use maven 3.5.3 building. Older branches use 3.5.4 still.
Our ‘Build Artifacts’ have a machine directory under which we emit vitals on the host the build was run on. We used to emit the result of ‘ulimit -l’ as a file named ‘ulimit-l’. This has been hijacked to instead emit result of running ‘ulimit -a’ which includes stat on ulimit -l.
ColumnFamilyDescriptor new builder API:
/\*\*
\* Retain all versions for a given TTL(retentionInterval), and then only a specific number
\* of versions(versionAfterInterval) after that interval elapses.
\*
\* @param retentionInterval Retain all versions for this interval
\* @param versionAfterInterval Retain no of versions to retain after retentionInterval
\*/
public ModifyableColumnFamilyDescriptor setVersionsWithTimeToLive(
final int retentionInterval, final int versionAfterInterval)
org.apache.hadoop.hbase.types.PBType is marked as deprecated without any replacement. It will be moved to hbase-example module and marked as IA.Private in 3.0.0. This is a mistake as it should not be part of our public API. Users who depend on this class should just copy the code your own code base.
Expose file system level read metrics for RegionServer.
If the HBase RS runs on top of HDFS, calculate the aggregation of ReadStatistics of each HdfsFileInputStream. These metrics include: (1) total number of bytes read from HDFS. (2) total number of bytes read from local DataNode. (3) total number of bytes read locally through short-circuit read. (4) total number of bytes read locally through zero-copy read.
Because HDFS ReadStatistics is calculated per input stream, it is not feasible to update the aggregated number in real time. Instead, the metrics are updated when an input stream is closed.
Admin can determine which tables go to which rsgroup by script (setting hbase.rsgroup.table.mapping.script with local filystem path) on Master side which aims to lighten the burden of admin operations. Note, since HBase 3+, rsgroup can be specified in TableDescriptor as well, if clients specify this, master will skip the determination from script.
Here is a simple example of script: {code}
#!/bin/bash namespace=$1 tablename=$2 if [[ $namespace == test ]]; then echo test elif [[ $tablename == *foo* ]]; then echo other else echo default fi {code}
MiniZKCluster now puts up its standalone node listening on loopback/127.0.0.1 rather than “localhost”.
Bumped hadoop-two.version to 2.10.0, which means we will drop the support for hadoop-2.8.x and hadoop-2.9.x.
Change timestamp display to be ISO8601 when toString on Cell and outputting in shell....
User used to see....
column=table:state, timestamp=1583967620343 .....
... but now sees:
column=table:state, timestamp=2020-03-11T23:00:20.343Z ....
merge_region shell command can now be used to merge more than 2 regions as well. It takes a list of regions as comma separated values or as an array of regions, and not just 2 regions. The full regionnames and encoded regionnames are continued to be accepted.
Rebuild our Dockerfile with support for multiple JDK versions. Use multiple stages in the Jenkinsfile instead of yetus's multijdk because of YETUS-953. Run those multiple stages in parallel to speed up results.
Note that multiple stages means multiple Yetus invocations means multiple comments on the PreCommit. This should become more obvious to users once we can make use of GitHub Checks API, HBASE-23902.
get_slowlog_responses and clear_slowlog_responses are used to retrieve and clear slow RPC logs from RingBuffer maintained by RegionServers.
New Admin APIs:
List<SlowLogRecord> getSlowLogResponses(final Set<ServerName> serverNames, final SlowLogQueryFilter slowLogQueryFilter) throws IOException;
List<Boolean> clearSlowLogResponses(final Set<ServerName> serverNames) throws IOException;
Configs:
Default 256
Default false
For more details, please look into “Get Slow Response Log from shell” section from HBase book.
Down the flakey re-rerun fork count from 1.0C -- i.e. a fork per CPU -- to 0.25C. On a recent run, the machine had 16 cores. 0.25 is 4 cores. We'd hardcoded fork count at 3 previous to changes made by parent.
Add a checkAndMutate(row, filter) method in the AsyncTable interface and the Table interface.
This method atomically checks if the row matches the specified filter. If it does, it adds the Put/Delete/RowMutations.
This is a fluent style API, the code is like:
For Table interface: {code} table.checkAndMutate(row, filter).thenPut(put); {code}
For AsyncTable interface: {code} table.checkAndMutate(row, filter).thenPut(put) .thenAccept(succ -> { if (succ) { System.out.println(“Check and put succeeded”); } else { System.out.println(“Check and put failed”); } }); {code}
The Jira Precommit job (https://builds.apache.org/job/PreCommit-HBASE-Build/) will now look for a file within the source tree (dev-support/jenkins_precommit_jira_yetus.sh) instead of depending on a script section embedded in the job.
Changed flakey list reporting to show 5 rather than 10 items. Also changed the second and first part fort counts to be 1C rather than hardcoded 3.
HBASE-23554 | Major | Encoded regionname to regionname utility
Adds shell command regioninfo:
hbase(main):001:0\> regioninfo '0e6aa5c19ae2b2627649dc7708ce27d0'
{ENCODED =\> 0e6aa5c19ae2b2627649dc7708ce27d0, NAME =\> 'TestTable,,1575941375972.0e6aa5c19ae2b2627649dc7708ce27d0.', STARTKEY =\> '', ENDKEY =\> '00000000000000000000299441'}
Took 0.4737 seconds
This JIRA adds a new configuration - `hbase.rs.cachecompactedblocksonwrite.threshold`. This configuration is the maximum total size (in bytes) of the compacted files below which the configuration `hbase.rs.cachecompactedblocksonwrite` is honoured. If the total size of the compacted fies exceeds this threshold, even when `hbase.rs.cachecompactedblocksonwrite` is enabled, the data blocks are not cached. Caching index and bloom blocks is not affected by this configuration (user configuration is always honoured).
Default value of this configuration is Long.MAX_VALUE. This means whatever the total size of the compacted files, it wil be cached.
Implements authorization for the HBase Web UI by limiting access to certain endpoints which could be used to extract sensitive information from HBase.
Access to these restricted endpoints can be limited to a group of administrators, identified either by a list of users (hbase.security.authentication.spnego.admin.users) or by a list of groups (hbase.security.authentication.spnego.admin.groups). By default, neither of these values are set which will preserve backwards compatibility (allowing all authenticated users to access all endpoints).
Further, users who have sensitive information in the HBase service configuration can set hbase.security.authentication.ui.config.protected to true which will treat the configuration endpoint as a protected, admin-only resource. By default, all authenticated users may access the configuration endpoint.
Enables master based registry as the default registry used by clients to fetch connection metadata. Refer to the section “Master Registry” in the client documentation for more details and advantages of this implementation over the default Zookeeper based registry.
Configuration parameter that controls the registry in use: hbase.client.registry.impl
Where to set this: HBase client configuration (hbase-site.xml)
Possible values:
org.apache.hadoop.hbase.client.ZKConnectionRegistry (For ZK based registry implementation)org.apache.hadoop.hbase.client.MasterRegistry (New, for master based registry implementation)Notes on defaults:
This feature has been back ported to 2.3.0 and later releases. MasterRegistry can be enabled by setting the following client configuration.
<property> <name>hbase.client.registry.impl</name> <value>org.apache.hadoop.hbase.client.MasterRegistry</value> </property>
caffeine: 2.6.2 => 2.8.1 commons-codec: 1.10 => 1.13 commons-io: 2.5 => 2.6 disrupter: 3.3.6 => 3.4.2 httpcore: 4.4.6 => 4.4.13 jackson: 2.9.10 => 2.10.1 jackson.databind: 2.9.10.1 => 2.10.1 jetty: 9.3.27.v20190418 => 9.3.28.v20191105 protobuf.plugin: 0.5.0 => 0.6.1 zookeeper: 3.4.10 => 3.4.14 slf4j: 1.7.25 => 1.7.30 rat: 0.12 => 0.13 asciidoctor: 1.5.5 => 1.5.8 asciidoctor.pdf: 1.5.0-alpha.15 => 1.5.0-rc.2 error-prone: 2.3.3 => 2.3.4
This change introduces an internal abstraction layer which allows for new SASL-based authentication mechanisms to be used inside HBase services. All existing SASL-based authentication mechanism were ported to the new abstraction, making no external change in runtime semantics, client API, or RPC serialization format.
Developers familiar with extending HBase can implement authentication mechanism beyond simple Kerberos and DelegationTokens which authenticate HBase users against some other user database. HBase service authentication (Master to/from RegionServer) continue to operate solely over Kerberos.
Introduce a new hbase-assembly/src/main/assembly/hadoop-three-compat.xml for build with hadoop 3.x.
Add a new config to hbase-default.xml
<property> <name>hbase.procedure.store.region.hfilecleaner.plugins</name> <value>org.apache.hadoop.hbase.master.cleaner.TimeToLiveHFileCleaner</value> <description>A comma-separated list of BaseHFileCleanerDelegate invoked by the RegionProcedureStore HFileCleaner service. These HFiles cleaners are called in order, so put the cleaner that prunes the most files in front. To implement your own BaseHFileCleanerDelegate, just put it in HBase's classpath and add the fully qualified class name here. Always add the above default hfile cleaners in the list as they will be overwritten in hbase-site.xml.</description> </property>
It will share the same TTL with other HFileCleaners. And you can also implement your own cleaner and change this property to enable it.
Updated parent pom to Apache version 22.
This issues fixes an issue with Bulk Loading on installations with Kerberos enabled and more than a single RegionServer. When multiple tables are involved in hosting a table‘s regions which are being bulk-loaded into, all but the RegionServer hosting the table’s first Region will “leak” one DistributedFileSystem object onto the heap, never freeing that memory. Eventually, with enough bulk loads, this will create a situation for RegionServers where they have no free heap space and will either spend all time in JVM GC, lose their ZK session, or crash with an OutOfMemoryError.
The only mitigation for this issue is to periodically restart RegionServers. All earlier versions of HBase 2.x are subject to this issue (2.0.x, <=2.1.8, <=2.2.3)
Add a new feature to improve MTTR which have 3 steps to failover:
Compared to DLS(distributed log split), this feature will reduce region open time significantly.
Config hbase.wal.split.to.hfile to true to enable this featue.
Changed the logging in hbase-zookeeper to use built-in formatting
From the PR:
"Yes. The two create the same output... I just wrote a small test suite to increase my confidence on that. I generated many tens of millions of random byte patterns and compared the output of the two algorithms. They came back identical every time.
“Just in case any inquiring minds would like to know, there is no longer an encoding required when generating the strings. The JDK implementation specifically specifies that strings returned are StandardCharsets.ISO_8859_1. This does not change anything because UTF8 and ISO_8859 overlap for the limited character set (64 characters) the encoding uses.”
Set hbase.balancer.max.balancing to a int value which <=0 will disable region balance throttling.
If cacheOnWrite is enabled during flush or compaction, index and bloom blocks(with data blocks) would be automatically cached during write.
If a RegionServer reports a Region as OPEN in disagreement with Master's status on the Region, the Master now tells the RegionServer to silently close the Region.
Makes it so the recently added HBCKServerCrashProcedure -- the SCP that gets invoked when an operator schedules an SCP via hbck2 scheduleRecoveries command -- now works the same as SCP EXCEPT if master knows nothing of the scheduled servername. In this latter case, HBCKSCP will do a full scan of hbase:meta looking for instances of the passed servername. If any found it will attempt cleanup of hbase:meta references by reassigning any found OPEN or OPENING and by closing any in CLOSING state.
Used to fix instances of what the ‘HBCK Report’ page shows as ‘Unknown Servers’.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.HFileProcedurePrettyPrinter to run the tool.
RegionsRecoveryChore introduced as part of HBASE-22460 tries to reopen regions based on config: hbase.regions.recovery.store.file.ref.count. Region reopen needs to take into consideration all compacted away store files that belong to the region and not store files(non-compacted).
Fixed this bug as part of this Jira. Updated description for corresponding configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
Very large number of ref count on a compacted store file indicates that it is a ref leak on that object(compacted store file). Such files can not be removed after it is invalidated via compaction. Only way to recover in such scenario is to reopen the region which can release all resources, like the refcount, leases, etc. This config represents Store files Ref Count threshold value considered for reopening regions. Any region with compacted store files ref count > this value would be eligible for reopening by master. Here, we get the max refCount among all refCounts on all compacted away store files that belong to a particular region. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.WALProcedurePrettyPrinter to run the tool.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.RegionProcedureStorePerformanceEvaluation to run the tool.
Use a region based procedure store to replace the old customized WAL based procedure store. The procedure data migration is done automatically during upgrading. After upgrading, the MasterProcWALs directory will be deleted and a new MasterProc directory will be created. And notice that a region will still write WAL so we still have WAL files and they will be moved to the oldWALs directory. The file name is mostly like a normal WAL file, and the only difference is that it is ended with “$masterproc$”.
Bumped surefire plugin to 3.0.0-M4
Now AsyncFSWAL also supports Durability.FSYNC_WAL.
The configuration ‘hbase.rs.cacheblocksonwrite’ was used to enable caching the blocks on write. But purposefully we were not caching the blocks when we do compaction (since it may be very aggressive) as the caching happens as and when the writer completes a block. In cloud environments since they have bigger sized caches - though they try to enable ‘hbase.rs.prefetchblocksonopen’ (non - aggressive way of caching the blocks proactively on reader creation) it does not help them because it takes time to cache the compacted blocks. This feature creates a new configuration ‘hbase.rs.cachecompactedblocksonwrite’ which when set to ‘true’ will enable the blocks created out of compaction. Remember that since it is aggressive caching the user should be having enough cache space - if not it may lead to other active blocks getting evicted. From the shell this can be enabled by using the option per Column Family also by using the below format {code} create ‘t1’, ‘f1’, {NUMREGIONS => 15, SPLITALGO => ‘HexStringSplit’, CONFIGURATION => {‘hbase.rs.cachecompactedblocksonwrite’ => ‘true’}} {code}
Users of the MOB feature can now use the mobrefs utility to get statistics about data in the MOB system and verify the health of backing files on HDFS.
HADOOP_CLASSPATH=/etc/hbase/conf:$(hbase mapredcp) yarn jar \
/some/path/to/hbase-shaded-mapreduce.jar mobrefs mobrefs-report-output some_table foo
See javadocs of the class MobRefReporter for more details.
the reference guide has added some information about MOB internals and troubleshooting.
The reference guide now includes a walk through of disabling the MOB feature if needed while maintaining availability.
Fixed unbalanced braces in string representation within HBase shell
The default rpc timeout for ReplicationSourceShipper#shipEdits is 60s, when bulkload replication enabled, timeout exception may be occurred. Now we can conf the timeout value through replication.source.shipedits.timeout, and it’s adaptive.
The newer HBase Thrift SPNEGO configs should not be required. The hbase.thrift.spnego.keytab.file and hbase.thrift.spnego.principal configs will fall back to the hbase.thrift.keytab.file and hbase.thrift.kerberos.principal original configs. The older configs will log a deprecation warning. It is preferred to new the newer SPNEGO configurations.
With BinaryComponentCompartor applications will be able to design diverse and powerful set of filters for rows and columns. See https://issues.apache.org/jira/browse/HBASE-22969 for example. In general, the comparator can be used with any filter taking ByteArrayComparable. As of now, following filters take ByteArrayComparable:
Adds a .editorconfig file with configurations populated by IntelliJ, based on our checkstyle configuration. There's lots of IntelliJ-specific configs in here that I assume are not replicated to Eclipse or Netbeans users. Any devs using those tools should push whatever updates they see fit, but please start with the checkstyle configs as the origin of truth.
An hbck2 scheduleRecoveries will run a subclass of ServerCrashProcedure which asks Master what Regions were on the dead Server but it will also do a hbase:meta table scan to see if any vestiges of the old Server remain (for the case where an SCP failed mid-point leaving references in place or where Master and hbase:meta deviated in accounting).
If holes in hbase:meta, hbck2 fixMeta now will update Master in-memory state so you do not need to restart master just so you can assign the new hole-bridging regions.
hbck2 scheduleRecoveries will now run a SCP that also looks in hbase:meta for any references to the scheduled server -- not just consult Master in-memory state -- just in case vestiges of the server are leftover in hbase:meta
HBase internal chores now log a moving average of how long execution of each chore takes at INFO level for the logger org.apache.hadoop.hbase.ScheduledChore.
Such messages will happen at most once per five minutes.
CleanerChore delegate initialization is now logged at INFO level instead of DEBUG
The ‘Procedures & Locks’ tab in Master UI only displays problematic Procedures now (RUNNABLE, WAITING-TIMEOUT, etc.). It no longer notes procedures whose state is SUCCESS.
the Apache HBase REST Proxy now uses Jackson Databind version 2.9.10.1 to address the following CVEs
Users of prior releases with Jackson Databind 2.9.10 are advised to either upgrade to this release or to upgrade their local Jackson Databind jar directly.
The MOB compaction process in the HBase Master now logs more about its activity.
In the event that you run into the problems described in HBASE-22075, there is a new HFileCleanerDelegate that will stop all removal of MOB hfiles from the archive area. It can be configured by adding org.apache.hadoop.hbase.mob.ManualMobMaintHFileCleaner to the list configured for hbase.master.hfilecleaner.plugins. This new cleaner delegate will cause your archive area to grow unbounded; you will have to manually prune files which may be prohibitively complex. Consider if your use case will allow you to mitigate by disabling mob compactions instead.
Caveats:
%hbase_root%/archive that accumulates files over time. You will have to determine which of these files are safe or not to remove.TRACE logging, the CleanerChore logger will include archive retention decision justifications.Adds per-user metrics for reads/writes to each RegionServer. These metrics are exported by default. hbase.regionserver.user.metrics.enabled can be used to disable the feature if desired for any reason.
Leaked store files can not be removed even after it is invalidated via compaction. A reasonable mitigation for a reader reference leak would be a fast reopen of the region on the same server.
Configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
This config represents Store files Ref Count threshold value considered for reopening regions. Any region with store files ref count > this value would be eligible for reopening by master. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Added a comment to make clear that read/write success counts are tallying column family success counts, not region success counts.
Additionally, the region read and write latencies previously only stored the latencies of the last column family of the region reads/writes. This has been fixed by using a map of each region to a list of read and write latency values.
Changes the message on the FNFE exception thrown when the file a Reference points to is missing; the message now includes detail on Reference as well as pointed-to file so can connect how FNFE relates to region open.
Use ‘totalRowActionRequestCount’ to calculate QPS on web UI.
Downstream users who wish to programmatically check the health of their HBase cluster may now rely on a public interface derived from the previously private implementation of the canary cli tool. The interface is named Canary and can be found in the user facing javadocs.
Downstream users who previously relied on the invoking the canary via the Java classname (either on the command line or programmatically) will need to change how they do so because the non-public implementation has moved.
Since 2.0.0，when one regionserver crashed and back online again, AssignmentManager will retain the region locations and try assign the regions to this regionserver(same host:port with the crashed one) again. But for 1.x.x, the behavior is round-robin assignment for the regions belong to the crashed regionserver. This jira change the “retain” assignment to round-robin assignment, which is same with 1.x.x version. This change will make the failover faster and improve availability.
Remove backward compatibility from `truncate` and `truncate_preserve` shell commands. This means that these commands from HBase Clients are not compatible with pre-0.99 HBase clusters.
giving the region mover “unload” command a region server name that isn‘t recognized by the cluster results in a "I don’t know about that host" message instead of a NPE.
set log level to DEBUG if you'd like the region mover to log the set of region server names it got back from the cluster.
Added a new IOEngine type for Bucket cache ie Persistent memory. In order to use BC over pmem configure IOEngine as <property> <name>hbase.bucketcache.ioengine</name> <value> pmem:///path in persistent memory </value> </property>
By default, snapshot auto cleanup based on TTL would be enabled for any new cluster. At any point in time, if snapshot cleanup is supposed to be stopped due to some snapshot restore activity or any other reason, it is advisable to disable it using shell command: hbase> snapshot_cleanup_switch false
We can re-enable it using: hbase> snapshot_cleanup_switch true
We can query whether snapshot auto cleanup is enabled for cluster using: hbase> snapshot_cleanup_enabled
Adds fix of overlaps to the fixMeta hbck service method. Uses the bulk-merge facility. Merges a max of 10 at a time. Set hbase.master.metafixer.max.merge.count to higher if you want to do more than 10 in the one go.
This issue adds via its subtasks:
* An ‘HBCK Report’ page to the Master UI added by HBASE-22527+HBASE-22709+HBASE-22723+ (since 2.1.6, 2.2.1, 2.3.0). Lists consistency or anomalies found via new hbase:meta consistency checking extensions added to CatalogJanitor (holes, overlaps, bad servers) and by a new ‘HBCK chore’ that runs at a lesser periodicity that will note filesystem orphans and overlaps as well as the following conditions: ** Master thought this region opened, but no regionserver reported it. ** Master thought this region opened on Server1, but regionserver reported Server2 ** More than one regionservers reported opened this region Both chores can be triggered from the shell to regenerate ‘new’ reports. * Means of scheduling a ServerCrashProcedure (HBASE-21393). * An ‘offline’ hbase:meta rebuild (HBASE-22680). * Offline replace of hbase.version and hbase.id * Documentation on how to use completebulkload tool to ‘adopt’ orphaned data found by new HBCK2 ‘filesystem’ check (see below) and ‘HBCK chore’ (HBASE-22859) * A ‘holes’ and ‘overlaps’ fix that runs in the master that uses new bulk-merge facility to collapse many overlaps in the one go. * hbase-operator-tools HBCK2 client tool got a bunch of additions: ** A specialized ‘fix’ for the case where operators ran old hbck ‘offlinemeta’ repair and destroyed their hbase:meta; it ties together holes in meta with orphaned data in the fs (HBASE-22567) ** A ‘filesystem’ command that reports on orphan data as well as bad references and hlinks with a ‘fix’ for the latter two options (based on hbck1 facility updated). ** Adds back the ‘replication’ fix facility from hbck1 (HBASE-22717)
The compound result is that hbck2 is now in excess of hbck1 abilities. The provided functionality is disaggregated as per the hbck2 philosophy of providing ‘plumbing’ rather than ‘porcelain’ so there is work to do still adding fix-it playbooks, scripting across outages, and automation.
HBASE-21879 introduces a utility class (org.apache.hadoop.hbase.io.ByteBuffAllocator) used for allocating/freeing ByteBuffers from/to NIO ByteBuffer pool, when BucketCache enabled with file or mmap engine, we will use this ByteBuffer pool to avoid temp ByteBuffer allocation a lot.
Introduces hbtop that‘s a real-time monitoring tool for HBase like Unix’s top command. See the ref guide for the details: https://hbase.apache.org/book.html#hbtop
Before this issue, read path was 100% offheap when block is in the BucketCache. But if a cache miss, then the RS needs to read the block via an on-heap API which causes high young-GC pressure.
This issue adds reading the block via offheap even if reading the block from filesystem directly. It requires hadoop version(>=2.9.3) but can also work with older hadoop versions (all works but we continue to read block onheap). It also requires HBASE-21946 which is not yet in place as of this writing/hbase-2.3.0.
We have written a careful doc about the implementation, performance and practice here: https://docs.google.com/document/d/1xSy9axGxafoH-Qc17zbD2Bd--rWjjI00xTWQZ8ZwI_E/edit#heading=h.nch5d72p27ex
Extends StochasticLoadBalancer to support user-provided cost function. These are loaded in addition to the default set of cost functions. Custom function implementations must extend StochasticLoadBalancer$CostFunction. Enable any additional functions by placing them on the master class path and configuring hbase.master.balancer.stochastic.additionalCostFunctions with a comma-separated list of fully-qualified class names.
Replace the ForkJoinPool in CleanerChore by ThreadPoolExecutor which can limit the spawn thread size and avoid the master GC frequently. The replacement is an internal implementation in CleanerChore, so no config key change, the upstream users can just upgrade the hbase master without any other change.
Introduced a new config key for the snapshot taking/restoring operations at master side: hbase.master.executor.snapshot.threads, its default value is 3. means we can have 3 snapshot operations running at the same time.
Add serveral API in TimeRange class for avoiding using the deprecated TimeRange constructor: * TimeRange#from: Represents the time interval [minStamp, Long.MAX_VALUE) * TimeRange#until: Represents the time interval [0, maxStamp) * TimeRange#between: Represents the time interval [minStamp, maxStamp)
Provide a public method in MultiRowRangeFilter class to speed the requirement of filtering with multiple row prefixes, it will expand the row prefixes as multiple rowkey ranges by MultiRowRangeFilter, it's more efficient. {code} public MultiRowRangeFilter(byte[][] rowKeyPrefixes); {code}
Update the base docker image to ubuntu 18.04 for the find flaky tests jenkins job.
Adds a fixMeta method to hbck Service. Fixes holes in hbase:meta. Follow-up to fix overlaps. See HBASE-22567 also.
Follow-on is adding a client-side to hbase-operator-tools that can exploit this new addition (HBASE-22825)
Changes merge so you can merge more than two regions at a time. Currently only available inside HBase. HBASE-22827, a follow-on, is about exposing the facility in the Admin API (and then via the shell).
New shaded artifact for testing: hbase-shaded-testing-util.
After HBASE-22776, the steps to config user scan snapshot feature is as followings:
We found a critical bug which can lead to WAL corruption when Durability.ASYNC_WAL is used. The reason is that we release a ByteBuffer before actually persist the content into WAL file.
The problem maybe lead to several errors, for example, ArrayIndexOfOutBounds when replaying WAL. This is because that the ByteBuffer is reused by others.
ERROR org.apache.hadoop.hbase.executor.EventHandler: Caught throwable while processing event RS_LOG_REPLAY java.lang.ArrayIndexOutOfBoundsException: 18056 at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1365) at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1358) at org.apache.hadoop.hbase.PrivateCellUtil.matchingFamily(PrivateCellUtil.java:735) at org.apache.hadoop.hbase.CellUtil.matchingFamily(CellUtil.java:816) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEditFamily(WALEdit.java:143) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEdit(WALEdit.java:148) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:297) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:195) at org.apache.hadoop.hbase.regionserver.SplitLogWorker$1.exec(SplitLogWorker.java:100)
And may even cause segmentation fault and crash the JVM directly. You will see a hs_err_pidXXX.log file and usually the problem is SIGSEGV. This is usually because that the ByteBuffer has already been returned to the OS and used for other purpose.
The problem has been reported several times in the past and this time Wellington Ramos Chevreuil provided the full logs and deeply analyzed the logs so we can find the root cause. And Lijin Bin figured out that the problem may only happen when Durability.ASYNC_WAL is used. Thanks to them.
The problem only effects the 2.x releases, all users are highly recommand to upgrade to a release which has this fix in, especially that if you use Durability.ASYNC_WAL.
Add a new method runHbckChore in Hbck interface and a new shell cmd hbck_chore_run to request HBCK chore to run at master side.
Adds a “CatalogJanitor hbase:meta Consistency Issues” section to the new ‘HBCK Report’ page added by HBASE-22709. This section is empty unless the most recent CatalogJanitor scan turned up problems. If so, will show table of issues found.
When CatalogJanitor runs, it now checks for holes, overlaps, empty info:regioninfo columns and bad servers. Dumps findings into log. Follow-up adds report to new ‘HBCK Report’ linked off the Master UI.
NOTE: All features but the badserver check made it into branch-2.1 and branch-2.0 backports.
The misspelled BufferedMutatorParams.opertationTimeout method has been marked as deprecated, and will be removed in 4.0.0. Please use the BufferedMutatorParams.operationTimeout method instead.
If a table user scan snapshots of the table, please config the following table scheme attribute to make granted users' ACLs are added to hfiles: alter ‘t1’, CONFIGURATION => {‘hbase.user.scan.snapshot.enable’ => ‘true’}
This feature is enabled by default. And the hbck chore run per 60 minutes by default. You can config “hbase.master.hbck.checker.interval” to a value lesser than or equal to 0 for disabling the chore.
Notice: the config “hbase.master.hbck.checker.interval” was renamed to “hbase.master.hbck.chore.interval” in HBASE-22737.
The HFileCleaner will clean the empty directories under archive, but if enable user scan snaphot feature, the user ACLs are set at there directories, so please config the following cleaner to make the directories with user ACLs not be cleaned: hbase.master.hfilecleaner.plugins=org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclCleaner
Upgrade jackson databind dependency to 2.9.9.1 due to CVEs
https://nvd.nist.gov/vuln/detail/CVE-2019-12814
https://nvd.nist.gov/vuln/detail/CVE-2019-12384
Add a new master web UI to show the potentially problematic opened regions. There are three case:
Feature: Take a Snapshot With TTL for auto-cleanup
Attribute:
Configs:
Default Snapshot TTL:
If Snapshot cleanup is supposed to be stopped due to some snapshot restore activity, disable it with config:
For more details, see the section “Take a Snapshot With TTL” in the HBase Reference Guide.
The config point “hbase.offheapcache.minblocksize” was wrong and is now deprecated. The new config point is “hbase.blockcache.minblocksize”.
OfflineMetaRepair is no longer supported in HBase-2+. Please refer to https://hbase.apache.org/book.html#HBCK2
This tool is deprecated in 2.x and will be removed in 3.0.
Mark the Hbck#scheduleServerCrashProcedure(List<HBaseProtos.ServerName> serverNames) as deprecated. Use Hbck#scheduleServerCrashProcedures(List<ServerName> serverNames) instead.
In HBASE-20734 we moved the recovered.edits onto the wal file system but when constructing the directory we missed the BASE_NAMESPACE_DIR(‘data’). So when using the default config, you will find that there are lots of new directories at the same level with the ‘data’ directory.
In this issue, we add the BASE_NAMESPACE_DIR back, and also try our best to clean up the wrong directories. But we can only clean up the region level directories, so if you want a clean fs layout on HDFS you still need to manually delete the empty directories at the same level with ‘data’.
The effect versions are 2.2.0, 2.1.[1-5], 1.4.[8-10], 1.3.[3-5].
Add a coprocessor to set HDFS acls to make hbase granted users with READ permission have the access to scan snapshots. To use this feature, please make sure the HDFS config is set: dfs.namenode.acls.enabled=true fs.permissions.umask-mode=027
and set the HBase config: hbase.coprocessor.master.classes=“org.apache.hadoop.hbase.security.access.AccessController,org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclController” hbase.user.scan.snapshot.enable=true
hbase.regionserver.compaction.check.period is used for controlling how often the compaction checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
hbase.regionserver.flush.check.period is used for controlling how ofter the flush checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
When run with JDK11 HBase now uses more recent version of the jaxws reference implementation (v2.3.2).
Added completebulkload short name for BulkLoadHFilesTool to bin/hbase.
Change the default hadoop-3 version to 3.1.2. Drop the support for the releases which are effected by CVE-2018-8029, see this email https://lists.apache.org/thread.html/3d6831c3893cd27b6850aea2feff7d536888286d588e703c6ffd2e82@%3Cuser.hadoop.apache.org%3E
This change exposes the aggregate count of store reader references for a given store as ‘storeRefCount’ in region metrics and ClusterStatus.
The HBase “source checksum” now uses SHA512 instead of MD5.
The CellUtil.setTimestamp method changes to be an API with audience LimitedPrivate(COPROC) in HBase 3.0. With that designation the API should remain stable within a given minor release line, but may change between minor releases.
Previously, this method was deprecated in HBase 2.0 for removal in HBase 3.0. Deprecation messages in HBase 2.y releases have been updated to indicate the expected API audience change.
The access method was used to the HttpServerFunctionalTest class as a common place.
The class LossyCounting was unintentionally marked Public but was never intended to be part of our public API. This oversight has been corrected and LossyCounting is now marked as Private and going forward may be subject to additional breaking changes or removal without notice. If you have taken a dependency on this class we recommend cloning it locally into your project before upgrading to this release.
Warnings for level headings are corrected in the book for the HBase Incompatibilities section.
Add hadoop 3.0.3, 3.1.1 3.1.2 in our hadoop check jobs.
The DumpReplicationQueues tool will now list replication queues sorted in chronological order.
Support get|set LogLevel in secure(kerberized) environment.
Fixes a formatting issue in the administration section of the book, where listing indentation were a little bit off.
This change adds the new method listNamespaces to the Admin interface, which can be used to retrieve a list of the namespaces present in the schema as an unprivileged operation. Formerly the only available method for accomplishing this was listNamespaceDescriptors, which requires GLOBAL CREATE or ADMIN permissions.
Now the default hadoop-two.version has been changed to 2.8.5, and all hadoop versions before 2.8.2(exclude) will not be supported any more.
Removed extra + in HRegion, HStore and LoadIncrementalHFiles for branch-2 and HRegion and HStore for branch-1.
Updated metrics core from 3.2.1 to 3.2.6.
The rubocop definition for the maximum method length was set to 75.
Fixes the formatting of the “Voting on Release Candidates” to actually show the quote and code formatting of the RAT check.
The rubocop configuration in the hbase-shell module now allows a line length with 100 characters, instead of 80 as before. For everything before 2.1.5 this change introduces rubocop itself.
This change adds new conditions for rolling the WAL for when syncs on the HDFS writer pipeline are perceived to be slow.
As before the configuration parameter hbase.regionserver.wal.slowsync.ms sets the slow sync warning threshold.
If we encounter hbase.regionserver.wal.slowsync.roll.threshold number of slow syncs (default 100) within the interval defined by hbase.regionserver.wal.slowsync.roll.interval.ms (default 1 minute), we will request a WAL roll.
Or, if the time for any sync exceeds the threshold set by hbase.regionserver.wal.roll.on.sync.ms (default 10 seconds) we will request a WAL roll immediately.
Operators can monitor how often these new thresholds result in a WAL roll by looking at newly added metrics to the WAL related metric group: * slowSyncRollRequest - How many times a roll was requested due to sync too slow on the write pipeline.
Additionally, as a part of this change there are also additional metrics for existing reasons for a WAL roll: * errorRollRequest - How many times a roll was requested due to I/O or other errors. * sizeRollRequest - How many times a roll was requested due to file size roll threshold.
MajorCompactorTTL Tool allows to compact all regions in a table that have been TTLed out. This saves space on DFS and is useful for tables which are similar to time series data. This is typically scheduled to run frequently (say via cron) to cleanup old data on an ongoing basis.
RSGroupMajorCompactionTTL tool is similar to MajorCompactorTTL but runs at a region server group level. If multiple tables in an rsgroup are similar to time-series data, then it runs a single command to clean them up. As more tables are added/removed from rsgroup, it's easy to have a single command to take care of all of them.
This change allows the system and superusers to initiate compactions, even when a space quota violation policy disallows compactions from happening. The original intent behind disallowing of compactions was to prevent end-user compactions from creating undue I/O load, not disallowing *any* compaction in the system.
Maven project integration for Eclipse has been isolated into a maven profile to ensure it only is active when in an Eclipse project.
Things should continue to behave the same for Eclipse users. If something should go wrong folks should manually activate the eclipse-specific profile.
Deprecated Preemptive Fail Fast related constants in HConstants, the support of this feature will be removed in 3.0.0 so use these constants will have no effect for 3.0.0+ releases. And the constants will be kept till 4.0.0.
Users can use ‘hbase.client.perserver.requests.threshold’ to control the number of concurrent requests to the same region server. Please see the release note of HBASE-16388 for more details.
Adds new configuration hbase.client.failure.map.cleanup.interval which defaults to ten minutes.
The default version of JRuby shipped with HBase has been updated to the JRuby 9.1.17.0 release.
For details on changes see the release notes for JRuby 9.1.17.0
Add below method in Table interface:
RegionLocator getRegionLocator() throws IOException;
Add below methods in AsyncTable interface:
AsyncTableRegionLocator getRegionLocator(); CompletableFuture<TableDescriptor> getDescriptor();
LruBlockCache uses the Segmented LRU (SLRU) policy to capture frequency and recency of the working set. It achieves concurrency by using an O(n) background thread to prioritize the entries and evict. Accessing an entry is O(1) by a hash table lookup, recording its logical access time, and setting a frequency flag. A write is performed in O(1) time by updating the hash table and triggering an async eviction thread. This provides ideal concurrency and minimizes the latencies by penalizing the thread instead of the caller. However the policy does not age the frequencies and may not be resilient to various workload patterns.
This change introduces a new L1 policy, TinyLfuBlockCache, which records the frequency in a counting sketch, ages periodically by halving the counters, and orders entries by SLRU. An entry is discarded by comparing the frequency of the new arrival to the SLRU's victim, and keeping the one with the highest frequency. This allows the operations to be performed in O(1) time and, though the use of a compact sketch, a much larger history is retained beyond the current working set. In a variety of real world traces the policy had near optimal hit rates.
New configuration variable hfile.block.cache.policy sets the eviction policy for the L1 block cache. The default is “LRU” (LruBlockCache). Set to “TinyLFU” to use TinyLfuBlockCache instead.
Introduced
Future<Void> createTableAsync(TableDescriptor);
Introduced these methods: void move(byte[]); void move(byte[], ServerName); Future<Void> splitRegionAsync(byte[]);
These methods are deprecated: void move(byte[], byte[])
Add a new jenkins file for running pre commit check for GitHub PR.
Add cloneSnapshot/restoreSnapshot with acl methods in AsyncAdmin.
When insufficient permissions, you now get:
HTTP/1.1 403 Forbidden
on the HTTP side, and in the message
Forbidden org.apache.hadoop.hbase.security.AccessDeniedException: org.apache.hadoop.hbase.security.AccessDeniedException: Insufficient permissions for user ‘myuser',action: get, tableName:mytable, family:cf. at org.apache.ranger.authorization.hbase.RangerAuthorizationCoprocessor.authorizeAccess(RangerAuthorizationCoprocessor.java:547) and the rest of the ADE stack
Now we will sort the javac WARNING/ERROR before generating diff in pre-commit so we can get a stable output for the error prone. The downside is that we just sort the output lexicographically so the line number will also be sorted lexicographically, which is a bit strange to human.
Exposes a new configuration property “zookeeper.multi.max.size” which dictates the maximum size of deletes that HBase will make to ZooKeeper in a single RPC. This property defaults to 1MB, which should fall beneath the default ZooKeeper limit of 2MB, controlled by “jute.maxbuffer”.
Fixed awkward dependency issue that prevented site building.
HBase 2.1.4 shipped with an early version of this fix that incorrectly altered the libraries included in our binary assembly for using Apache Hadoop 2.7 (the current build default Hadoop version for 2.1.z). For folks running out of the box against a Hadoop 2.7 cluster (or folks who skip the installation step of replacing the bundled Hadoop libraries) this will result in a failure at Region Server startup due to a missing class definition. e.g.:
2019-03-27 09:02:05,779 ERROR [main] regionserver.HRegionServer: Failed construction RegionServer java.lang.NoClassDefFoundError: org/apache/htrace/SamplerBuilder at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:644) at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:628) at org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:149) at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2667) at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:93) at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2701) at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2683) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:372) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:171) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:356) at org.apache.hadoop.fs.Path.getFileSystem(Path.java:295) at org.apache.hadoop.hbase.util.CommonFSUtils.getRootDir(CommonFSUtils.java:362) at org.apache.hadoop.hbase.util.CommonFSUtils.isValidWALRootDir(CommonFSUtils.java:411) at org.apache.hadoop.hbase.util.CommonFSUtils.getWALRootDir(CommonFSUtils.java:387) at org.apache.hadoop.hbase.regionserver.HRegionServer.initializeFileSystem(HRegionServer.java:704) at org.apache.hadoop.hbase.regionserver.HRegionServer.<init>(HRegionServer.java:613) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.hadoop.hbase.regionserver.HRegionServer.constructRegionServer(HRegionServer.java:3029) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.start(HRegionServerCommandLine.java:63) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.run(HRegionServerCommandLine.java:87) at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) at org.apache.hadoop.hbase.util.ServerCommandLine.doMain(ServerCommandLine.java:149) at org.apache.hadoop.hbase.regionserver.HRegionServer.main(HRegionServer.java:3047) Caused by: java.lang.ClassNotFoundException: org.apache.htrace.SamplerBuilder at java.net.URLClassLoader.findClass(URLClassLoader.java:381) at java.lang.ClassLoader.loadClass(ClassLoader.java:424) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349) at java.lang.ClassLoader.loadClass(ClassLoader.java:357) ... 26 more
Workaround via any one of the following:
hadoop executable is in the PATH seen at Region Server startup and that you are not using the HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP bypass.htrace-core-3.1.0-incubating.jar available to all Region Servers via the HBASE_CLASSPATH environment variable.htrace-core-3.1.0-incubating.jar available to all Region Servers by copying it into the directory ${HBASE_HOME}/lib/client-facing-thirdparty/.Add a listTableDescriptors(List<TableName>) method in the AsyncAdmin interface, to align with the Admin interface.
Deprecate Admin.deleteSnapshot(byte[]), please use the String version instead.
Add a mergeRegionsAsync(byte[][], boolean) method in the AsyncAdmin interface.
Instead of using assert, now we will throw IllegalArgumentException when you want to merge less than 2 regions at client side. And also, at master side, instead of using assert, now we will throw DoNotRetryIOException if you want merge more than 2 regions, since we only support merging two regions at once for now.
Add drainXXX parameter for balancerSwitch/splitSwitch/mergeSwitch methods in the AsyncAdmin interface, which has the same meaning with the synchronous parameter for these methods in the Admin interface.
As of HBase 3.0, the ByteBufferUtils class is now marked as a Private API for internal project use only. Downstream users are advised that it no longer has any compatibility promises across releases.
As of earlier HBase release lines the class is now marked as deprecated to call attention to this planned transition.
bulkload (HFileOutputFormat2) support config the compression on client ,you can set the job configuration “hbase.mapreduce.hfileoutputformat.compression” override the auto-detection of the target table's compression
Add a cloneSnapshotAsync method with restoreAcl parameter. Deprecated restoreSnapshotAsync method as it just ignores the failsafe configuration. Make snapshotAsync method returns a Future<Void>. Deprecated the snapshot related methods which take a ‘byte[]’ as the snapshot name. Use default methods to reduce the code base for implementation classes.
Deprecated AsyncTable.isTableAvailable(TableName, byte[][]).
After HBASE-21871, we can specify a peer table name with --peerTableName in VerifyReplication tool like the following: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable 5 TestTable
In addition, we can compare any 2 tables in any remote clusters with specifying both peerId and --peerTableName.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable zk1,zk2,zk3:2181/hbase TestTable
Adds below flush, split, and compaction metrics
HBASE-21481 improves the quality of access control, by strengthening the protection of super users's privileges.
Now we have four types of RIT procedure metrics, assign, unassign, move, reopen. The meaning of assign/unassign is changed, as we will not increase the unassign metric and then the assign metric when moving a region. Also introduced two new procedure metrics, open and close, which are used to track the open/close region calls to region server. We may send open/close multiple times to finish a RIT since we may retry multiple times.
Problem: This is an old problem since HBASE-2231. The compaction event marker was only writed to WAL. But after flush, the WAL may be archived, which means an useful compaction event marker be deleted, too. So the compacted store files cannot be archived when region open and replay WAL.
Solution: After this jira, the compaction event tracker will be writed to HFile. When region open and load store files, read the compaction evnet tracker from HFile and archive the compacted store files which still exist.
HBase contains two quota scopes: MACHINE and CLUSTER. Before this patch, set quota operations did not expose scope option to client api and use MACHINE as default, CLUSTER scope can not be set and used. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’
This issue implements CLUSTER scope in a simple way: For user, namespace, user over namespace quota, use [ClusterLimit / RSNum] as machine limit. For table and user over table quota, use [ClusterLimit / TotalTableRegionNum * MachineTableRegionNum] as machine limit. After this patch, user can set CLUSTER scope quota, but MACHINE is still default if user ignore scope. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’ set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => MACHINE set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => CLUSTER
Change spotbugs version to 3.1.11.
This modifies “status ‘replication’” output, fixing inconsistencies on the reporting times and ages of last shipped edits, as well as wrong calculation of replication lags.
It also introduces additional info for each recovery queue, which was not accounted by this command before.
The new output for “status ‘replication’” command is explained in details below: a) Source started, target stopped, no edits arrived on source yet: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ... b) Source started, target stopped, add edit on source: ... Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:21:00 GMT 2018, Replication Lag=2459 ... c) Source started, target stopped, edit added on source, restart source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 Recovered Queue: 1-hbase01.home,16020,1542784524057 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:23:00 GMT 2018, Replication Lag=201495 ... d) Source started, target stopped, add edit on source, restart source, add another edit on source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=6349 Recovered Queue: 1-hbase01.home,16020,1542782758742 No Ops shipped since last restart, SizeOfLogQueue=0, TimeStampOfLastArrivedInSource=Wed Nov 21 06:53:05 GMT 2018, Replication Lag=569394 ... e) Source started, target stopped, add edit on source, restart source, add another edit on source, start target: ... SOURCE: PeerID=1 Normal Queue: 1 AgeOfLastShippedOp=30000, TimeStampOfLastShippedOp=Wed Nov 21 07:07:58 GMT 2018, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=0 ... f) Source started, target stopped, add edit on source, restart source, restart target: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ...
Remove bloom filter type ROWPREFIX_DELIMITED. May add it back when find a better solution.
Support enable or disable exceed throttle quota. Exceed throttle quota means, user can over consume user/namespace/table quota if region server has additional available quota because other users don't consume at the same time. Use the following shell commands to enable/disable exceed throttle quota: enable_exceed_throttle_quota disable_exceed_throttle_quota There are two limits when enable exceed throttle quota:
Remove jackson dependencies from most hbase modules except hbase-rest, use shaded gson instead. The output json will be a bit different since jackson can use getter/setter, but gson will always use the fields.
Mark HConstants.META_QOS as deprecated. It is for internal use only, which is the highest priority. You should not try to set a priority greater than or equal to this value, although it is no harm but also useless.
This patch adds the ability to disable split and/or merge for a table (By default, split and merge are enabled for a table).
Allows shell to set Scan options previously not exposed. See additions as part of the scan help by typing following hbase shell:
hbase> help ‘scan’
We can specify peerQuorumAddress instead of peerId in VerifyReplication tool. So it no longer requires peerId to be setup when using this tool.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication zk1,zk2,zk3:2181/hbase testTable
Introduce a VerifyWALEntriesReplicationEndpoint which replicates nothing but only verifies if all the cells are valid. It can be used to capture bugs for writing WAL, as most times we will not read the WALs again after writing it if there are no region server crashes.
Introduced an new config key in this issue: hbase.regionserver.inmemory.compaction.pool.size. the default value would be 10. you can configure this to set the pool size of in-memory compaction pool. Note that all memstores in one region server will share the same pool, so if you have many regions in one region server, you need to set this larger to compact faster for better read performance.
Make StoppedRpcClientException extend DoNotRetryIOException.
To implement user permission control in Precedure V2, move grant and revoke method from AccessController to master firstly. Mark AccessController#grant and AccessController#revoke as deprecated and please use Admin#grant and Admin#revoke instead.
IMPORTANT: Due to security issues, all users who use hbase thrift should avoid using releases which do not have this fix.
The effect releases are: 2.1.x: 2.1.2 and below 2.0.x: 2.0.4 and below 1.x: 1.4.x and below
If you are using the effect releases above, please consider upgrading to a newer release ASAP.
For users who have configured hbase.bucketcache.ioengine with either the file:, files:, or mmap: prefix, and configured it to be persistent via the hbase.bucketcache.persistent.path property, the serialization format of the bucket cache has changed between versions. The old state will not be read during startup, and there is currently no migration path. The impact is expected to be minimal, however, since the cache will rebuild over time as access patterns dictate.
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Adds backoff in ServerCrashProcedure wait on WAL split to complete if large backlog of files to split (Its possible to avoid SCP blocking, waiting on WALs to split if you use procedure-based splitting -- set ‘hbase.split.wal.zk.coordinated’ to false to enable procedure based wal splitting.)
Notice this has changed log level for mismatching row keys, originally those were being logged at INFO level, now it's logged at DEBUG level. This is consistent with the logging of mismatching cells. Also, for missing row keys, it now logs row key values in human readable format, making it more meaningful for operators troubleshooting mismatches.
Introduce a new config hbase.replication.drop.on.deleted.columnfamily, default is false. When config to true, the replication will drop the edits for columnfamily that has been deleted from the replication source and target.
This change extends the Normalizer with a handful of new configurations. The configuration points supported are:
hbase.normalizer.split.enabled Whether to split a region as part of normalization. Default: true.hbase.normalizer.merge.enabled Whether to merge a region as part of normalization. Default true.hbase.normalizer.min.region.count The minimum number of regions in a table to consider it for merge normalization. Default: 3.hbase.normalizer.merge.min_region_age.days The minimum age for a region to be considered for a merge, in days. Default: 3.hbase.normalizer.merge.min_region_size.mb The minimum size for a region to be considered for a merge, in whole MBs. Default: 1.Add a hbase-logging module, put the log4j related code in this module only so other modules do not need to depend on log4j at compile scope. See the comments of Log4jUtils and InternalLog4jUtils for more details.
Add a log4j.properties to the test jar of hbase-logging module, so for other sub modules we just need to depend on the test jar of hbase-logging module at test scope to output the log to console, without placing a log4j.properties in the test resources as they all (almost) have the same content. And this test module will not be included in the assembly tarball so it will not mess up the binary distribution.
Ban direct commons-logging dependency, and ban commons-logging and log4j imports in non-test code, to avoid mess up the downstream users logging framework. In hbase-logging module we do need to use log4j classes and the trick is to use full class name.
Add jcl-over-slf4j and jul-to-slf4j dependencies, as some of our dependencies use jcl or jul as logging framework, we should also redirect their log message to slf4j.
Added new metric to differentiate sink startup time from last OP applied time.
Original behaviour was to always set startup time to TimestampsOfLastAppliedOp, and always show it on “status ‘replication’” command, regardless if the sink ever applied any OP.
This was confusing, specially for scenarios where cluster was just acting as source, the output could lead to wrong interpretations about sink not applying edits or replication being stuck.
With the new metric, we now compare the two metrics values, assuming that if both are the same, there's never been any OP shipped to the given sink, so output would reflect it more clearly, to something as for example:
SINK: TimeStampStarted=Thu Dec 06 23:59:47 GMT 2018, Waiting for OPs...
HBase ships ZooKeeper 3.5.x. Was the EOL'd 3.4.x. 3.5.x client can talk to 3.4.x ensemble.
The ZooKeeper project has built a FAQ that documents known issues and work-arounds when upgrading existing deployments.
Add backoff. Avoid retrying every 100ms.
Runs ‘catalogjanitor_run’ and ‘hbck_chore_run’ inline with the loading of the ‘HBCK Report’ page.
Pass ‘?cache=true’ to skip inline invocation of ‘catalogjanitor_run’ and ‘hbck_chore_run’ drawing the page.
Introduced a general ‘local region’ at master side to store the procedure data, etc.
The hfile of this region will be stored on the root fs while the wal will be stored on the wal fs. This issue supercedes part of the code for HBASE-23326, as now we store the data in ‘MasterData’ directory instead of ‘MasterProcs’.
The old hfiles will be moved to the global hfile archived directory with the suffix $-masterlocalhfile-$. The wal files will be moved to the global old wal directory with the suffix $masterlocalwal$. The TimeToLiveMasterLocalStoreHFileCleaner and TimeToLiveMasterLocalStoreWALCleaner are configured by default for cleaning the old hfiles and wal files, and the default TTLs are both 7 days.
Relocate test-only REST RemoteHTable and RemoteAdmin from src/ to test/. And mark them as InterfaceAudience.Private.
Config key: hbase.regionserver.slowlog.systable.enabled Default value: false
This config can be enabled if hbase.regionserver.slowlog.buffer.enabled is already enabled. While hbase.regionserver.slowlog.buffer.enabled ensures that any slow/large RPC logs with complete details are written to ring buffer available at each RegionServer, hbase.regionserver.slowlog.systable.enabled would ensure that all such logs are also persisted in new system table hbase:slowlog. Operator can scan hbase:slowlog with filters to retrieve specific attribute matching records and this table would be useful to capture historical performance of slowness of RPC calls with detailed analysis.
hbase:slowlog consists of single ColumnFamily info. info consists of multiple qualifiers similar to the attributes available to query as part of Admin API: get_slowlog_responses.
One example of a row from hbase:slowlog scan result (Attached a sample screenshot in the Jira) :
\x024\xC1\x06X\x81\xF6\xEC column=info:call_details, timestamp=2020-05-16T14:59:58.764Z, value=Scan(org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos$ScanRequest)
\x024\xC1\x06X\x81\xF6\xEC column=info:client_address, timestamp=2020-05-16T14:59:58.764Z, value=172.20.10.2:57348
\x024\xC1\x06X\x81\xF6\xEC column=info:method_name, timestamp=2020-05-16T14:59:58.764Z, value=Scan
\x024\xC1\x06X\x81\xF6\xEC column=info:param, timestamp=2020-05-16T14:59:58.764Z, value=region { type: REGION_NAME value: “cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.” } scan { a ttribute { name: “_isolationlevel_” value: “\x5C000” } start_row: “cccccccc” time_range { from: 0 to: 9223372036854775807 } max_versions: 1 cache_blocks: true max_result_size: 2 097152 caching: 2147483647 include_stop_row: false } number_of_rows: 2147483647 close_scanner: false client_handles_partials: true client_handles_heartbeats: true track_scan_met rics: false
\x024\xC1\x06X\x81\xF6\xEC column=info:processing_time, timestamp=2020-05-16T14:59:58.764Z, value=24
\x024\xC1\x06X\x81\xF6\xEC column=info:queue_time, timestamp=2020-05-16T14:59:58.764Z, value=0
\x024\xC1\x06X\x81\xF6\xEC column=info:region_name, timestamp=2020-05-16T14:59:58.764Z, value=cluster_test,cccccccc,1589635796466.aa45e1571d533f5ed0bb31cdccaaf9cf.
\x024\xC1\x06X\x81\xF6\xEC column=info:response_size, timestamp=2020-05-16T14:59:58.764Z, value=211227
\x024\xC1\x06X\x81\xF6\xEC column=info:server_class, timestamp=2020-05-16T14:59:58.764Z, value=HRegionServer
\x024\xC1\x06X\x81\xF6\xEC column=info:start_time, timestamp=2020-05-16T14:59:58.764Z, value=1589640743932
\x024\xC1\x06X\x81\xF6\xEC column=info:type, timestamp=2020-05-16T14:59:58.764Z, value=ALL
\x024\xC1\x06X\x81\xF6\xEC column=info:username, timestamp=2020-05-16T14:59:58.764Z, value=vjasani
HBASE-24271 makes changes the the default conf/hbase-site.xml such that bin/hbase will run directly out of the binary tarball or a compiled source tree without any configuration modifications vs. Hadoop 2.8+. This changes our long-standing history of shipping no configured values in conf/hbase-site.xml, so existing processes that assume this file is empty of configuration properties may require attention.
Use Slf4jRequestLog instead of the log4j HttpRequestLogAppender in HttpServer.
The request log is disabled by default in conf/log4j.properties by the following lines:
log4j.category.http.requests=INFO,NullAppender log4j.additivity.http.requests=false
Change the ‘NullAppender’ to what ever you want if you want to enable request log.
Notice that, the logger name for master status http server is ‘http.requests.master’, and for region server it is ‘http.requests.regionserver’
Use a empty string to represent no column specified for deleteall in shell mode. useage:
deleteall ‘test’,‘r1’,'',12345 deleteall ‘test’, {ROWPREFIXFILTER => ‘prefix’}, '', 12345
Added a new hbase-asyncfs module to hold the asynchronous dfs output stream implementation for implementing WAL.
Make the logic of the versions chosen more reasonable for raw scan, to avoid lose result when using filter.
Moved to hbase-thirdparty 3.3.0.
This feature enables the HBase Web UI‘s to accept a ‘proxyuser’ via the HTTP Request’s query string. When the parameter `hbase.security.authentication.spnego.kerberos.proxyuser.enable` is set to `true` in hbase-site.xml (default is `false`), the HBase UI will attempt to impersonate the user specified by the query parameter “doAs”. This query parameter is checked case-insensitively. When this option is not provided, the user who executed the request is the “real” user and there is no ability to execute impersonation against the WebUI.
For example, if the user “bob” with Kerberos credentials executes a request against the WebUI with this feature enabled and a query string which includes `doAs=alice`, the HBase UI will treat this request as executed as `alice`, not `bob`.
The standard Hadoop proxyuser configuration properties to limit users who may impersonate others apply to this change (e.g. to enable `bob` to impersonate `alice`). See the Hadoop documentation for more information on how to configure these proxyuser rules.
bin/hbase will now dynamically select a Garbage Collector implementation based on the detected JVM version. JDKs 8,9,10 use -XX:+UseConcMarkSweepGC, while JDK11+ use -XX:+UseG1GC.
Notice a slight compatibility change. Previously, the garbage collector choice would always be appended to a user-provided value for HBASE_OPTS. As of this change, this setting will only be applied when HBASE_OPTS is unset. That means that operators who provide a value for this variable will now need to also specify the collector. This is especially important for those on JDK8, where the vm default GC is not the recommended ConcMarkSweep.
Default value: false Description: If value is true, RegionServer will abort batch requests of Put/Delete with number of rows in a batch operation exceeding threshold defined by value of config: hbase.rpc.rows.warning.threshold.
StochasticLoadBalancer functional improvement:
StochasticLoadBalancer would rebalance the cluster if there are any idle RegionServers in the cluster (RegionServer having no region), while other RegionServers have at least 1 region available.
user or admin can now use hbase shell > rename_rsgroup ‘oldname’, ‘newname’ to rename rsgroup.
Add hadoop-3.2.0 and hadoop-3.2.1 in hadoop check and when ‘--quick-hadoopcheck’ we will only check hadoop-3.2.1.
Notice that, for aligning the personality scripts across all the active branches, we will commit the patch to all active branches, but the hadoop-3.2.x support in hadoopcheck is only applied to branch-2.2+.
`-PrunSmallTests` now pass on JDK11 when using `-Phadoop.profile=3.0`.
Tests that fail because a process -- RegionServer or Master -- called System.exit, will now instead throw an exception.
Hadoop hosts have had their ulimit -u raised from 10000 to 30000 (per user, by INFRA). The Docker build container has had its limit raised from 10000 to 12500.
Support RSGroup renaming in core codebase. New API Admin#renameRSGroup(String, String) is introduced in 3.0.0.
The Canary tool now offers a WebUI when run in region mode (the default mode). It is enabled by default, and by default, it binds to 0.0.0.0:16050. This can be overridden by setting hbase.canary.info.bindAddress and hbase.canary.info.port. To disable entirely, set the port to -1.
Pass --threads=2 building on jenkins. It shortens nightly build times by about ~25%.
It works by running module build/test in parallel when dependencies allow. Upping the forkcount beyond the pom default of 0.25C would have us broach our CPU budget on jenkins when two modules are running in parallel (2 modules at 0.25% of CPU each makes 0.5C and on jenkins, hadoop nodes run two jenkins executors per host). Higher forkcounts also seems to threaten build stability.
For running tests locally, to go faster, up fork count.
$ x=“0.5C” ; mvn --threads=2 -Dsurefire.firstPartForkCount=$x -Dsurefire.secondPartForkCount=$x test -PrunAllTests
You could up the x from 0.5C to 1.0C but YMMV (On overcommitted hardware, tests start bombing out pretty soon after startup). You could try upping thread count but on occasion are likely to overcommit hardware.
Start docker with upped ulimit for nproc passing ‘--ulimit nproc=12500’. It was 10000, the default, but made it 12500. Then, set PROC_LIMIT in hbase-personality so when yetus runs, it is w/ the new 12500 value.
Pass -T2 to mvn. Makes it so we do two modules-at-a-time dependencies willing. Helps speed build and testing. Doubles the resource usage when running modules in parallel.
Master & RegionService now support refresh policy authorization defined in hbase-policy.xml without restarting service. To refresh policy, please execute hbase shell command: update_config or update_config_all after policy file updated and synced on all nodes.
This change modifies the default acquisition policy for the region's close lock in order to prevent observed starvation of close requests. The new boolean configuration parameter ‘hbase.regionserver.fair.region.close.lock’ controls the lock acquisition policy: if true, the lock is created in fair mode (default); if false, the lock is created in nonfair mode (the old default).
The PrimaryRegionCountSkewCostFunction for the StochasticLoadBalancer is only needed when the read replicas feature is enabled. With this change, that function now properly indicates that it is not needed when the read replica feature is off.
If this improvement is not available, operators with clusters that are not using the read replica feature should manually disable it by setting hbase.master.balancer.stochastic.primaryRegionCountCost to 0.0 in hbase-site.xml for all HBase Masters.
Now AsyncFSWAL can also be used against the directory which has EC enabled. Need to make sure you also make use of the hadoop 3.x client as the option is only available in hadoop 3.x.
Branches-2.3+ use maven 3.5.3 building. Older branches use 3.5.4 still.
Our ‘Build Artifacts’ have a machine directory under which we emit vitals on the host the build was run on. We used to emit the result of ‘ulimit -l’ as a file named ‘ulimit-l’. This has been hijacked to instead emit result of running ‘ulimit -a’ which includes stat on ulimit -l.
ColumnFamilyDescriptor new builder API:
/\*\*
\* Retain all versions for a given TTL(retentionInterval), and then only a specific number
\* of versions(versionAfterInterval) after that interval elapses.
\*
\* @param retentionInterval Retain all versions for this interval
\* @param versionAfterInterval Retain no of versions to retain after retentionInterval
\*/
public ModifyableColumnFamilyDescriptor setVersionsWithTimeToLive(
final int retentionInterval, final int versionAfterInterval)
org.apache.hadoop.hbase.types.PBType is marked as deprecated without any replacement. It will be moved to hbase-example module and marked as IA.Private in 3.0.0. This is a mistake as it should not be part of our public API. Users who depend on this class should just copy the code your own code base.
Expose file system level read metrics for RegionServer.
If the HBase RS runs on top of HDFS, calculate the aggregation of ReadStatistics of each HdfsFileInputStream. These metrics include: (1) total number of bytes read from HDFS. (2) total number of bytes read from local DataNode. (3) total number of bytes read locally through short-circuit read. (4) total number of bytes read locally through zero-copy read.
Because HDFS ReadStatistics is calculated per input stream, it is not feasible to update the aggregated number in real time. Instead, the metrics are updated when an input stream is closed.
Admin can determine which tables go to which rsgroup by script (setting hbase.rsgroup.table.mapping.script with local filystem path) on Master side which aims to lighten the burden of admin operations. Note, since HBase 3+, rsgroup can be specified in TableDescriptor as well, if clients specify this, master will skip the determination from script.
Here is a simple example of script: {code}
#!/bin/bash namespace=$1 tablename=$2 if [[ $namespace == test ]]; then echo test elif [[ $tablename == *foo* ]]; then echo other else echo default fi {code}
MiniZKCluster now puts up its standalone node listening on loopback/127.0.0.1 rather than “localhost”.
Bumped hadoop-two.version to 2.10.0, which means we will drop the support for hadoop-2.8.x and hadoop-2.9.x.
Change timestamp display to be ISO8601 when toString on Cell and outputting in shell....
User used to see....
column=table:state, timestamp=1583967620343 .....
... but now sees:
column=table:state, timestamp=2020-03-11T23:00:20.343Z ....
merge_region shell command can now be used to merge more than 2 regions as well. It takes a list of regions as comma separated values or as an array of regions, and not just 2 regions. The full regionnames and encoded regionnames are continued to be accepted.
Rebuild our Dockerfile with support for multiple JDK versions. Use multiple stages in the Jenkinsfile instead of yetus's multijdk because of YETUS-953. Run those multiple stages in parallel to speed up results.
Note that multiple stages means multiple Yetus invocations means multiple comments on the PreCommit. This should become more obvious to users once we can make use of GitHub Checks API, HBASE-23902.
get_slowlog_responses and clear_slowlog_responses are used to retrieve and clear slow RPC logs from RingBuffer maintained by RegionServers.
New Admin APIs:
List<SlowLogRecord> getSlowLogResponses(final Set<ServerName> serverNames, final SlowLogQueryFilter slowLogQueryFilter) throws IOException;
List<Boolean> clearSlowLogResponses(final Set<ServerName> serverNames) throws IOException;
Configs:
Default 256
Default false
For more details, please look into “Get Slow Response Log from shell” section from HBase book.
Down the flakey re-rerun fork count from 1.0C -- i.e. a fork per CPU -- to 0.25C. On a recent run, the machine had 16 cores. 0.25 is 4 cores. We'd hardcoded fork count at 3 previous to changes made by parent.
Add a checkAndMutate(row, filter) method in the AsyncTable interface and the Table interface.
This method atomically checks if the row matches the specified filter. If it does, it adds the Put/Delete/RowMutations.
This is a fluent style API, the code is like:
For Table interface: {code} table.checkAndMutate(row, filter).thenPut(put); {code}
For AsyncTable interface: {code} table.checkAndMutate(row, filter).thenPut(put) .thenAccept(succ -> { if (succ) { System.out.println(“Check and put succeeded”); } else { System.out.println(“Check and put failed”); } }); {code}
The Jira Precommit job (https://builds.apache.org/job/PreCommit-HBASE-Build/) will now look for a file within the source tree (dev-support/jenkins_precommit_jira_yetus.sh) instead of depending on a script section embedded in the job.
Changed flakey list reporting to show 5 rather than 10 items. Also changed the second and first part fort counts to be 1C rather than hardcoded 3.
HBASE-23554 | Major | Encoded regionname to regionname utility
Adds shell command regioninfo:
hbase(main):001:0\> regioninfo '0e6aa5c19ae2b2627649dc7708ce27d0'
{ENCODED =\> 0e6aa5c19ae2b2627649dc7708ce27d0, NAME =\> 'TestTable,,1575941375972.0e6aa5c19ae2b2627649dc7708ce27d0.', STARTKEY =\> '', ENDKEY =\> '00000000000000000000299441'}
Took 0.4737 seconds
This JIRA adds a new configuration - `hbase.rs.cachecompactedblocksonwrite.threshold`. This configuration is the maximum total size (in bytes) of the compacted files below which the configuration `hbase.rs.cachecompactedblocksonwrite` is honoured. If the total size of the compacted fies exceeds this threshold, even when `hbase.rs.cachecompactedblocksonwrite` is enabled, the data blocks are not cached. Caching index and bloom blocks is not affected by this configuration (user configuration is always honoured).
Default value of this configuration is Long.MAX_VALUE. This means whatever the total size of the compacted files, it wil be cached.
Implements authorization for the HBase Web UI by limiting access to certain endpoints which could be used to extract sensitive information from HBase.
Access to these restricted endpoints can be limited to a group of administrators, identified either by a list of users (hbase.security.authentication.spnego.admin.users) or by a list of groups (hbase.security.authentication.spnego.admin.groups). By default, neither of these values are set which will preserve backwards compatibility (allowing all authenticated users to access all endpoints).
Further, users who have sensitive information in the HBase service configuration can set hbase.security.authentication.ui.config.protected to true which will treat the configuration endpoint as a protected, admin-only resource. By default, all authenticated users may access the configuration endpoint.
Enables master based registry as the default registry used by clients to fetch connection metadata. Refer to the section “Master Registry” in the client documentation for more details and advantages of this implementation over the default Zookeeper based registry.
Configuration parameter that controls the registry in use: hbase.client.registry.impl
Where to set this: HBase client configuration (hbase-site.xml)
Possible values:
org.apache.hadoop.hbase.client.ZKConnectionRegistry (For ZK based registry implementation)org.apache.hadoop.hbase.client.MasterRegistry (New, for master based registry implementation)Notes on defaults:
This feature has been back ported to 2.3.0 and later releases. MasterRegistry can be enabled by setting the following client configuration.
<property> <name>hbase.client.registry.impl</name> <value>org.apache.hadoop.hbase.client.MasterRegistry</value> </property>
caffeine: 2.6.2 => 2.8.1 commons-codec: 1.10 => 1.13 commons-io: 2.5 => 2.6 disrupter: 3.3.6 => 3.4.2 httpcore: 4.4.6 => 4.4.13 jackson: 2.9.10 => 2.10.1 jackson.databind: 2.9.10.1 => 2.10.1 jetty: 9.3.27.v20190418 => 9.3.28.v20191105 protobuf.plugin: 0.5.0 => 0.6.1 zookeeper: 3.4.10 => 3.4.14 slf4j: 1.7.25 => 1.7.30 rat: 0.12 => 0.13 asciidoctor: 1.5.5 => 1.5.8 asciidoctor.pdf: 1.5.0-alpha.15 => 1.5.0-rc.2 error-prone: 2.3.3 => 2.3.4
Adds being able to edit hbase:meta table schema. For example,
hbase(main):006:0> alter ‘hbase:meta’, {NAME => ‘info’, DATA_BLOCK_ENCODING => ‘ROW_INDEX_V1’} Updating all regions with the new schema... All regions updated. Done. Took 1.2138 seconds
You can even add columnfamilies. Howevert, you cannot delete any of the core hbase:meta column families such as ‘info’ and ‘table’.
This change introduces an internal abstraction layer which allows for new SASL-based authentication mechanisms to be used inside HBase services. All existing SASL-based authentication mechanism were ported to the new abstraction, making no external change in runtime semantics, client API, or RPC serialization format.
Developers familiar with extending HBase can implement authentication mechanism beyond simple Kerberos and DelegationTokens which authenticate HBase users against some other user database. HBase service authentication (Master to/from RegionServer) continue to operate solely over Kerberos.
Introduce a new hbase-assembly/src/main/assembly/hadoop-three-compat.xml for build with hadoop 3.x.
Add a new config to hbase-default.xml
<property> <name>hbase.procedure.store.region.hfilecleaner.plugins</name> <value>org.apache.hadoop.hbase.master.cleaner.TimeToLiveHFileCleaner</value> <description>A comma-separated list of BaseHFileCleanerDelegate invoked by the RegionProcedureStore HFileCleaner service. These HFiles cleaners are called in order, so put the cleaner that prunes the most files in front. To implement your own BaseHFileCleanerDelegate, just put it in HBase's classpath and add the fully qualified class name here. Always add the above default hfile cleaners in the list as they will be overwritten in hbase-site.xml.</description> </property>
It will share the same TTL with other HFileCleaners. And you can also implement your own cleaner and change this property to enable it.
Updated parent pom to Apache version 22.
This issues fixes an issue with Bulk Loading on installations with Kerberos enabled and more than a single RegionServer. When multiple tables are involved in hosting a table‘s regions which are being bulk-loaded into, all but the RegionServer hosting the table’s first Region will “leak” one DistributedFileSystem object onto the heap, never freeing that memory. Eventually, with enough bulk loads, this will create a situation for RegionServers where they have no free heap space and will either spend all time in JVM GC, lose their ZK session, or crash with an OutOfMemoryError.
The only mitigation for this issue is to periodically restart RegionServers. All earlier versions of HBase 2.x are subject to this issue (2.0.x, <=2.1.8, <=2.2.3)
Add a new feature to improve MTTR which have 3 steps to failover:
Compared to DLS(distributed log split), this feature will reduce region open time significantly.
Config hbase.wal.split.to.hfile to true to enable this featue.
Changed the logging in hbase-zookeeper to use built-in formatting
From the PR:
"Yes. The two create the same output... I just wrote a small test suite to increase my confidence on that. I generated many tens of millions of random byte patterns and compared the output of the two algorithms. They came back identical every time.
“Just in case any inquiring minds would like to know, there is no longer an encoding required when generating the strings. The JDK implementation specifically specifies that strings returned are StandardCharsets.ISO_8859_1. This does not change anything because UTF8 and ISO_8859 overlap for the limited character set (64 characters) the encoding uses.”
Set hbase.balancer.max.balancing to a int value which <=0 will disable region balance throttling.
If cacheOnWrite is enabled during flush or compaction, index and bloom blocks(with data blocks) would be automatically cached during write.
If a RegionServer reports a Region as OPEN in disagreement with Master's status on the Region, the Master now tells the RegionServer to silently close the Region.
Makes it so the recently added HBCKServerCrashProcedure -- the SCP that gets invoked when an operator schedules an SCP via hbck2 scheduleRecoveries command -- now works the same as SCP EXCEPT if master knows nothing of the scheduled servername. In this latter case, HBCKSCP will do a full scan of hbase:meta looking for instances of the passed servername. If any found it will attempt cleanup of hbase:meta references by reassigning any found OPEN or OPENING and by closing any in CLOSING state.
Used to fix instances of what the ‘HBCK Report’ page shows as ‘Unknown Servers’.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.HFileProcedurePrettyPrinter to run the tool.
RegionsRecoveryChore introduced as part of HBASE-22460 tries to reopen regions based on config: hbase.regions.recovery.store.file.ref.count. Region reopen needs to take into consideration all compacted away store files that belong to the region and not store files(non-compacted).
Fixed this bug as part of this Jira. Updated description for corresponding configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
Very large number of ref count on a compacted store file indicates that it is a ref leak on that object(compacted store file). Such files can not be removed after it is invalidated via compaction. Only way to recover in such scenario is to reopen the region which can release all resources, like the refcount, leases, etc. This config represents Store files Ref Count threshold value considered for reopening regions. Any region with compacted store files ref count > this value would be eligible for reopening by master. Here, we get the max refCount among all refCounts on all compacted away store files that belong to a particular region. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.WALProcedurePrettyPrinter to run the tool.
Use ./hbase org.apache.hadoop.hbase.procedure2.store.region.RegionProcedureStorePerformanceEvaluation to run the tool.
Use a region based procedure store to replace the old customized WAL based procedure store. The procedure data migration is done automatically during upgrading. After upgrading, the MasterProcWALs directory will be deleted and a new MasterProc directory will be created. And notice that a region will still write WAL so we still have WAL files and they will be moved to the oldWALs directory. The file name is mostly like a normal WAL file, and the only difference is that it is ended with “$masterproc$”.
Bumped surefire plugin to 3.0.0-M4
Now AsyncFSWAL also supports Durability.FSYNC_WAL.
The configuration ‘hbase.rs.cacheblocksonwrite’ was used to enable caching the blocks on write. But purposefully we were not caching the blocks when we do compaction (since it may be very aggressive) as the caching happens as and when the writer completes a block. In cloud environments since they have bigger sized caches - though they try to enable ‘hbase.rs.prefetchblocksonopen’ (non - aggressive way of caching the blocks proactively on reader creation) it does not help them because it takes time to cache the compacted blocks. This feature creates a new configuration ‘hbase.rs.cachecompactedblocksonwrite’ which when set to ‘true’ will enable the blocks created out of compaction. Remember that since it is aggressive caching the user should be having enough cache space - if not it may lead to other active blocks getting evicted. From the shell this can be enabled by using the option per Column Family also by using the below format {code} create ‘t1’, ‘f1’, {NUMREGIONS => 15, SPLITALGO => ‘HexStringSplit’, CONFIGURATION => {‘hbase.rs.cachecompactedblocksonwrite’ => ‘true’}} {code}
Users of the MOB feature can now use the mobrefs utility to get statistics about data in the MOB system and verify the health of backing files on HDFS.
HADOOP_CLASSPATH=/etc/hbase/conf:$(hbase mapredcp) yarn jar \
/some/path/to/hbase-shaded-mapreduce.jar mobrefs mobrefs-report-output some_table foo
See javadocs of the class MobRefReporter for more details.
the reference guide has added some information about MOB internals and troubleshooting.
The reference guide now includes a walk through of disabling the MOB feature if needed while maintaining availability.
Fixed unbalanced braces in string representation within HBase shell
The default rpc timeout for ReplicationSourceShipper#shipEdits is 60s, when bulkload replication enabled, timeout exception may be occurred. Now we can conf the timeout value through replication.source.shipedits.timeout, and it’s adaptive.
The newer HBase Thrift SPNEGO configs should not be required. The hbase.thrift.spnego.keytab.file and hbase.thrift.spnego.principal configs will fall back to the hbase.thrift.keytab.file and hbase.thrift.kerberos.principal original configs. The older configs will log a deprecation warning. It is preferred to new the newer SPNEGO configurations.
With BinaryComponentCompartor applications will be able to design diverse and powerful set of filters for rows and columns. See https://issues.apache.org/jira/browse/HBASE-22969 for example. In general, the comparator can be used with any filter taking ByteArrayComparable. As of now, following filters take ByteArrayComparable:
Adds a .editorconfig file with configurations populated by IntelliJ, based on our checkstyle configuration. There's lots of IntelliJ-specific configs in here that I assume are not replicated to Eclipse or Netbeans users. Any devs using those tools should push whatever updates they see fit, but please start with the checkstyle configs as the origin of truth.
An hbck2 scheduleRecoveries will run a subclass of ServerCrashProcedure which asks Master what Regions were on the dead Server but it will also do a hbase:meta table scan to see if any vestiges of the old Server remain (for the case where an SCP failed mid-point leaving references in place or where Master and hbase:meta deviated in accounting).
If holes in hbase:meta, hbck2 fixMeta now will update Master in-memory state so you do not need to restart master just so you can assign the new hole-bridging regions.
hbck2 scheduleRecoveries will now run a SCP that also looks in hbase:meta for any references to the scheduled server -- not just consult Master in-memory state -- just in case vestiges of the server are leftover in hbase:meta
HBase internal chores now log a moving average of how long execution of each chore takes at INFO level for the logger org.apache.hadoop.hbase.ScheduledChore.
Such messages will happen at most once per five minutes.
CleanerChore delegate initialization is now logged at INFO level instead of DEBUG
The ‘Procedures & Locks’ tab in Master UI only displays problematic Procedures now (RUNNABLE, WAITING-TIMEOUT, etc.). It no longer notes procedures whose state is SUCCESS.
the Apache HBase REST Proxy now uses Jackson Databind version 2.9.10.1 to address the following CVEs
Users of prior releases with Jackson Databind 2.9.10 are advised to either upgrade to this release or to upgrade their local Jackson Databind jar directly.
The MOB compaction process in the HBase Master now logs more about its activity.
In the event that you run into the problems described in HBASE-22075, there is a new HFileCleanerDelegate that will stop all removal of MOB hfiles from the archive area. It can be configured by adding org.apache.hadoop.hbase.mob.ManualMobMaintHFileCleaner to the list configured for hbase.master.hfilecleaner.plugins. This new cleaner delegate will cause your archive area to grow unbounded; you will have to manually prune files which may be prohibitively complex. Consider if your use case will allow you to mitigate by disabling mob compactions instead.
Caveats:
%hbase_root%/archive that accumulates files over time. You will have to determine which of these files are safe or not to remove.TRACE logging, the CleanerChore logger will include archive retention decision justifications.Adds per-user metrics for reads/writes to each RegionServer. These metrics are exported by default. hbase.regionserver.user.metrics.enabled can be used to disable the feature if desired for any reason.
Leaked store files can not be removed even after it is invalidated via compaction. A reasonable mitigation for a reader reference leak would be a fast reopen of the region on the same server.
Configs:
Regions Recovery Chore interval in milliseconds. This chore keeps running at this interval to find all regions with configurable max store file ref count and reopens them. Defaults to 20 mins
This config represents Store files Ref Count threshold value considered for reopening regions. Any region with store files ref count > this value would be eligible for reopening by master. Default value -1 indicates this feature is turned off. Only positive integer value should be provided to enable this feature.
Added a comment to make clear that read/write success counts are tallying column family success counts, not region success counts.
Additionally, the region read and write latencies previously only stored the latencies of the last column family of the region reads/writes. This has been fixed by using a map of each region to a list of read and write latency values.
Changes the message on the FNFE exception thrown when the file a Reference points to is missing; the message now includes detail on Reference as well as pointed-to file so can connect how FNFE relates to region open.
Use ‘totalRowActionRequestCount’ to calculate QPS on web UI.
Downstream users who wish to programmatically check the health of their HBase cluster may now rely on a public interface derived from the previously private implementation of the canary cli tool. The interface is named Canary and can be found in the user facing javadocs.
Downstream users who previously relied on the invoking the canary via the Java classname (either on the command line or programmatically) will need to change how they do so because the non-public implementation has moved.
Since 2.0.0，when one regionserver crashed and back online again, AssignmentManager will retain the region locations and try assign the regions to this regionserver(same host:port with the crashed one) again. But for 1.x.x, the behavior is round-robin assignment for the regions belong to the crashed regionserver. This jira change the “retain” assignment to round-robin assignment, which is same with 1.x.x version. This change will make the failover faster and improve availability.
Remove backward compatibility from `truncate` and `truncate_preserve` shell commands. This means that these commands from HBase Clients are not compatible with pre-0.99 HBase clusters.
giving the region mover “unload” command a region server name that isn‘t recognized by the cluster results in a "I don’t know about that host" message instead of a NPE.
set log level to DEBUG if you'd like the region mover to log the set of region server names it got back from the cluster.
Added a new IOEngine type for Bucket cache ie Persistent memory. In order to use BC over pmem configure IOEngine as <property> <name>hbase.bucketcache.ioengine</name> <value> pmem:///path in persistent memory </value> </property>
By default, snapshot auto cleanup based on TTL would be enabled for any new cluster. At any point in time, if snapshot cleanup is supposed to be stopped due to some snapshot restore activity or any other reason, it is advisable to disable it using shell command: hbase> snapshot_cleanup_switch false
We can re-enable it using: hbase> snapshot_cleanup_switch true
We can query whether snapshot auto cleanup is enabled for cluster using: hbase> snapshot_cleanup_enabled
Adds fix of overlaps to the fixMeta hbck service method. Uses the bulk-merge facility. Merges a max of 10 at a time. Set hbase.master.metafixer.max.merge.count to higher if you want to do more than 10 in the one go.
This issue adds via its subtasks:
* An ‘HBCK Report’ page to the Master UI added by HBASE-22527+HBASE-22709+HBASE-22723+ (since 2.1.6, 2.2.1, 2.3.0). Lists consistency or anomalies found via new hbase:meta consistency checking extensions added to CatalogJanitor (holes, overlaps, bad servers) and by a new ‘HBCK chore’ that runs at a lesser periodicity that will note filesystem orphans and overlaps as well as the following conditions: ** Master thought this region opened, but no regionserver reported it. ** Master thought this region opened on Server1, but regionserver reported Server2 ** More than one regionservers reported opened this region Both chores can be triggered from the shell to regenerate ‘new’ reports. * Means of scheduling a ServerCrashProcedure (HBASE-21393). * An ‘offline’ hbase:meta rebuild (HBASE-22680). * Offline replace of hbase.version and hbase.id * Documentation on how to use completebulkload tool to ‘adopt’ orphaned data found by new HBCK2 ‘filesystem’ check (see below) and ‘HBCK chore’ (HBASE-22859) * A ‘holes’ and ‘overlaps’ fix that runs in the master that uses new bulk-merge facility to collapse many overlaps in the one go. * hbase-operator-tools HBCK2 client tool got a bunch of additions: ** A specialized ‘fix’ for the case where operators ran old hbck ‘offlinemeta’ repair and destroyed their hbase:meta; it ties together holes in meta with orphaned data in the fs (HBASE-22567) ** A ‘filesystem’ command that reports on orphan data as well as bad references and hlinks with a ‘fix’ for the latter two options (based on hbck1 facility updated). ** Adds back the ‘replication’ fix facility from hbck1 (HBASE-22717)
The compound result is that hbck2 is now in excess of hbck1 abilities. The provided functionality is disaggregated as per the hbck2 philosophy of providing ‘plumbing’ rather than ‘porcelain’ so there is work to do still adding fix-it playbooks, scripting across outages, and automation.
HBASE-21879 introduces a utility class (org.apache.hadoop.hbase.io.ByteBuffAllocator) used for allocating/freeing ByteBuffers from/to NIO ByteBuffer pool, when BucketCache enabled with file or mmap engine, we will use this ByteBuffer pool to avoid temp ByteBuffer allocation a lot.
Introduces hbtop that‘s a real-time monitoring tool for HBase like Unix’s top command. See the ref guide for the details: https://hbase.apache.org/book.html#hbtop
Before this issue, we've made the read path 100% offheap when block hit the BucketCache 100%, but if the cache missed then RS need to read the block by on-heap API, which would cause high young GC pressure. This issue will read the block by offheap even if reading the block from filesystem directly, it have some requirement for hadoop version(>=2.9.3) but can also works with older hadoop version(means still works fine but will read block onheap). We have written a careful doc about the implementation, performance and practice here: https://docs.google.com/document/d/1xSy9axGxafoH-Qc17zbD2Bd--rWjjI00xTWQZ8ZwI_E/edit#heading=h.nch5d72p27ex, for more details please read it.
Extends StochasticLoadBalancer to support user-provided cost function. These are loaded in addition to the default set of cost functions. Custom function implementations must extend StochasticLoadBalancer$CostFunction. Enable any additional functions by placing them on the master class path and configuring hbase.master.balancer.stochastic.additionalCostFunctions with a comma-separated list of fully-qualified class names.
Replace the ForkJoinPool in CleanerChore by ThreadPoolExecutor which can limit the spawn thread size and avoid the master GC frequently. The replacement is an internal implementation in CleanerChore, so no config key change, the upstream users can just upgrade the hbase master without any other change.
Introduced a new config key for the snapshot taking/restoring operations at master side: hbase.master.executor.snapshot.threads, its default value is 3. means we can have 3 snapshot operations running at the same time.
Add serveral API in TimeRange class for avoiding using the deprecated TimeRange constructor: * TimeRange#from: Represents the time interval [minStamp, Long.MAX_VALUE) * TimeRange#until: Represents the time interval [0, maxStamp) * TimeRange#between: Represents the time interval [minStamp, maxStamp)
Provide a public method in MultiRowRangeFilter class to speed the requirement of filtering with multiple row prefixes, it will expand the row prefixes as multiple rowkey ranges by MultiRowRangeFilter, it's more efficient. {code} public MultiRowRangeFilter(byte[][] rowKeyPrefixes); {code}
Update the base docker image to ubuntu 18.04 for the find flaky tests jenkins job.
Adds a fixMeta method to hbck Service. Fixes holes in hbase:meta. Follow-up to fix overlaps. See HBASE-22567 also.
Follow-on is adding a client-side to hbase-operator-tools that can exploit this new addition (HBASE-22825)
Changes merge so you can merge more than two regions at a time. Currently only available inside HBase. HBASE-22827, a follow-on, is about exposing the facility in the Admin API (and then via the shell).
New shaded artifact for testing: hbase-shaded-testing-util.
After HBASE-22776, the steps to config user scan snapshot feature is as followings:
We found a critical bug which can lead to WAL corruption when Durability.ASYNC_WAL is used. The reason is that we release a ByteBuffer before actually persist the content into WAL file.
The problem maybe lead to several errors, for example, ArrayIndexOfOutBounds when replaying WAL. This is because that the ByteBuffer is reused by others.
ERROR org.apache.hadoop.hbase.executor.EventHandler: Caught throwable while processing event RS_LOG_REPLAY java.lang.ArrayIndexOutOfBoundsException: 18056 at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1365) at org.apache.hadoop.hbase.KeyValue.getFamilyLength(KeyValue.java:1358) at org.apache.hadoop.hbase.PrivateCellUtil.matchingFamily(PrivateCellUtil.java:735) at org.apache.hadoop.hbase.CellUtil.matchingFamily(CellUtil.java:816) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEditFamily(WALEdit.java:143) at org.apache.hadoop.hbase.wal.WALEdit.isMetaEdit(WALEdit.java:148) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:297) at org.apache.hadoop.hbase.wal.WALSplitter.splitLogFile(WALSplitter.java:195) at org.apache.hadoop.hbase.regionserver.SplitLogWorker$1.exec(SplitLogWorker.java:100)
And may even cause segmentation fault and crash the JVM directly. You will see a hs_err_pidXXX.log file and usually the problem is SIGSEGV. This is usually because that the ByteBuffer has already been returned to the OS and used for other purpose.
The problem has been reported several times in the past and this time Wellington Ramos Chevreuil provided the full logs and deeply analyzed the logs so we can find the root cause. And Lijin Bin figured out that the problem may only happen when Durability.ASYNC_WAL is used. Thanks to them.
The problem only effects the 2.x releases, all users are highly recommand to upgrade to a release which has this fix in, especially that if you use Durability.ASYNC_WAL.
Add a new method runHbckChore in Hbck interface and a new shell cmd hbck_chore_run to request HBCK chore to run at master side.
Adds a “CatalogJanitor hbase:meta Consistency Issues” section to the new ‘HBCK Report’ page added by HBASE-22709. This section is empty unless the most recent CatalogJanitor scan turned up problems. If so, will show table of issues found.
When CatalogJanitor runs, it now checks for holes, overlaps, empty info:regioninfo columns and bad servers. Dumps findings into log. Follow-up adds report to new ‘HBCK Report’ linked off the Master UI.
NOTE: All features but the badserver check made it into branch-2.1 and branch-2.0 backports.
The misspelled BufferedMutatorParams.opertationTimeout method has been marked as deprecated, and will be removed in 4.0.0. Please use the BufferedMutatorParams.operationTimeout method instead.
If a table user scan snapshots of the table, please config the following table scheme attribute to make granted users' ACLs are added to hfiles: alter ‘t1’, CONFIGURATION => {‘hbase.user.scan.snapshot.enable’ => ‘true’}
This feature is enabled by default. And the hbck chore run per 60 minutes by default. You can config “hbase.master.hbck.checker.interval” to a value lesser than or equal to 0 for disabling the chore.
Notice: the config “hbase.master.hbck.checker.interval” was renamed to “hbase.master.hbck.chore.interval” in HBASE-22737.
This adds [-h|-help] options to rowcounter. Passing either -h or -help will print rowcounter guide as below:
$hbase rowcounter -h
usage: hbase rowcounter <tablename> [options] [<column1> <column2>...] Options: --starttime=<arg> starting time filter to start counting rows from. --endtime=<arg> end time filter limit, to only count rows up to this timestamp. --range=<arg> [startKey],[endKey][;[startKey],[endKey]...]] --expectedCount=<arg> expected number of rows to be count. For performance, consider the following configuration properties: -Dhbase.client.scanner.caching=100 -Dmapreduce.map.speculative=false
The HFileCleaner will clean the empty directories under archive, but if enable user scan snaphot feature, the user ACLs are set at there directories, so please config the following cleaner to make the directories with user ACLs not be cleaned: hbase.master.hfilecleaner.plugins=org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclCleaner
Upgrade jackson databind dependency to 2.9.9.1 due to CVEs
https://nvd.nist.gov/vuln/detail/CVE-2019-12814
https://nvd.nist.gov/vuln/detail/CVE-2019-12384
Add a new master web UI to show the potentially problematic opened regions. There are three case:
Feature: Take a Snapshot With TTL for auto-cleanup
Attribute:
Configs:
Default Snapshot TTL:
If Snapshot cleanup is supposed to be stopped due to some snapshot restore activity, disable it with config:
For more details, see the section “Take a Snapshot With TTL” in the HBase Reference Guide.
The config point “hbase.offheapcache.minblocksize” was wrong and is now deprecated. The new config point is “hbase.blockcache.minblocksize”.
OfflineMetaRepair is no longer supported in HBase-2+. Please refer to https://hbase.apache.org/book.html#HBCK2
This tool is deprecated in 2.x and will be removed in 3.0.
Mark the Hbck#scheduleServerCrashProcedure(List<HBaseProtos.ServerName> serverNames) as deprecated. Use Hbck#scheduleServerCrashProcedures(List<ServerName> serverNames) instead.
In HBASE-20734 we moved the recovered.edits onto the wal file system but when constructing the directory we missed the BASE_NAMESPACE_DIR(‘data’). So when using the default config, you will find that there are lots of new directories at the same level with the ‘data’ directory.
In this issue, we add the BASE_NAMESPACE_DIR back, and also try our best to clean up the wrong directories. But we can only clean up the region level directories, so if you want a clean fs layout on HDFS you still need to manually delete the empty directories at the same level with ‘data’.
The effect versions are 2.2.0, 2.1.[1-5], 1.4.[8-10], 1.3.[3-5].
Add a coprocessor to set HDFS acls to make hbase granted users with READ permission have the access to scan snapshots. To use this feature, please make sure the HDFS config is set: dfs.namenode.acls.enabled=true fs.permissions.umask-mode=027
and set the HBase config: hbase.coprocessor.master.classes=“org.apache.hadoop.hbase.security.access.AccessController,org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclController” hbase.user.scan.snapshot.enable=true
hbase.regionserver.compaction.check.period is used for controlling how often the compaction checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
hbase.regionserver.flush.check.period is used for controlling how ofter the flush checker runs. If unset, will use hbase.server.thread.wakefrequency as default value.
When run with JDK11 HBase now uses more recent version of the jaxws reference implementation (v2.3.2).
Added completebulkload short name for BulkLoadHFilesTool to bin/hbase.
Change the default hadoop-3 version to 3.1.2. Drop the support for the releases which are effected by CVE-2018-8029, see this email https://lists.apache.org/thread.html/3d6831c3893cd27b6850aea2feff7d536888286d588e703c6ffd2e82@%3Cuser.hadoop.apache.org%3E
This change exposes the aggregate count of store reader references for a given store as ‘storeRefCount’ in region metrics and ClusterStatus.
The HBase “source checksum” now uses SHA512 instead of MD5.
The CellUtil.setTimestamp method changes to be an API with audience LimitedPrivate(COPROC) in HBase 3.0. With that designation the API should remain stable within a given minor release line, but may change between minor releases.
Previously, this method was deprecated in HBase 2.0 for removal in HBase 3.0. Deprecation messages in HBase 2.y releases have been updated to indicate the expected API audience change.
The access method was used to the HttpServerFunctionalTest class as a common place.
The class LossyCounting was unintentionally marked Public but was never intended to be part of our public API. This oversight has been corrected and LossyCounting is now marked as Private and going forward may be subject to additional breaking changes or removal without notice. If you have taken a dependency on this class we recommend cloning it locally into your project before upgrading to this release.
Warnings for level headings are corrected in the book for the HBase Incompatibilities section.
Add hadoop 3.0.3, 3.1.1 3.1.2 in our hadoop check jobs.
The DumpReplicationQueues tool will now list replication queues sorted in chronological order.
Support get|set LogLevel in secure(kerberized) environment.
Fixes a formatting issue in the administration section of the book, where listing indentation were a little bit off.
This change adds the new method listNamespaces to the Admin interface, which can be used to retrieve a list of the namespaces present in the schema as an unprivileged operation. Formerly the only available method for accomplishing this was listNamespaceDescriptors, which requires GLOBAL CREATE or ADMIN permissions.
Now the default hadoop-two.version has been changed to 2.8.5, and all hadoop versions before 2.8.2(exclude) will not be supported any more.
Removed extra + in HRegion, HStore and LoadIncrementalHFiles for branch-2 and HRegion and HStore for branch-1.
Updated metrics core from 3.2.1 to 3.2.6.
The rubocop definition for the maximum method length was set to 75.
Fixes the formatting of the “Voting on Release Candidates” to actually show the quote and code formatting of the RAT check.
The rubocop configuration in the hbase-shell module now allows a line length with 100 characters, instead of 80 as before. For everything before 2.1.5 this change introduces rubocop itself.
This change adds new conditions for rolling the WAL for when syncs on the HDFS writer pipeline are perceived to be slow.
As before the configuration parameter hbase.regionserver.wal.slowsync.ms sets the slow sync warning threshold.
If we encounter hbase.regionserver.wal.slowsync.roll.threshold number of slow syncs (default 100) within the interval defined by hbase.regionserver.wal.slowsync.roll.interval.ms (default 1 minute), we will request a WAL roll.
Or, if the time for any sync exceeds the threshold set by hbase.regionserver.wal.roll.on.sync.ms (default 10 seconds) we will request a WAL roll immediately.
Operators can monitor how often these new thresholds result in a WAL roll by looking at newly added metrics to the WAL related metric group: * slowSyncRollRequest - How many times a roll was requested due to sync too slow on the write pipeline.
Additionally, as a part of this change there are also additional metrics for existing reasons for a WAL roll: * errorRollRequest - How many times a roll was requested due to I/O or other errors. * sizeRollRequest - How many times a roll was requested due to file size roll threshold.
MajorCompactorTTL Tool allows to compact all regions in a table that have been TTLed out. This saves space on DFS and is useful for tables which are similar to time series data. This is typically scheduled to run frequently (say via cron) to cleanup old data on an ongoing basis.
RSGroupMajorCompactionTTL tool is similar to MajorCompactorTTL but runs at a region server group level. If multiple tables in an rsgroup are similar to time-series data, then it runs a single command to clean them up. As more tables are added/removed from rsgroup, it's easy to have a single command to take care of all of them.
This change allows the system and superusers to initiate compactions, even when a space quota violation policy disallows compactions from happening. The original intent behind disallowing of compactions was to prevent end-user compactions from creating undue I/O load, not disallowing *any* compaction in the system.
Maven project integration for Eclipse has been isolated into a maven profile to ensure it only is active when in an Eclipse project.
Things should continue to behave the same for Eclipse users. If something should go wrong folks should manually activate the eclipse-specific profile.
Deprecated Preemptive Fail Fast related constants in HConstants, the support of this feature will be removed in 3.0.0 so use these constants will have no effect for 3.0.0+ releases. And the constants will be kept till 4.0.0.
Users can use ‘hbase.client.perserver.requests.threshold’ to control the number of concurrent requests to the same region server. Please see the release note of HBASE-16388 for more details.
Adds new configuration hbase.client.failure.map.cleanup.interval which defaults to ten minutes.
The default version of JRuby shipped with HBase has been updated to the JRuby 9.1.17.0 release.
For details on changes see the release notes for JRuby 9.1.17.0
Add below method in Table interface:
RegionLocator getRegionLocator() throws IOException;
Add below methods in AsyncTable interface:
AsyncTableRegionLocator getRegionLocator(); CompletableFuture<TableDescriptor> getDescriptor();
LruBlockCache uses the Segmented LRU (SLRU) policy to capture frequency and recency of the working set. It achieves concurrency by using an O(n) background thread to prioritize the entries and evict. Accessing an entry is O(1) by a hash table lookup, recording its logical access time, and setting a frequency flag. A write is performed in O(1) time by updating the hash table and triggering an async eviction thread. This provides ideal concurrency and minimizes the latencies by penalizing the thread instead of the caller. However the policy does not age the frequencies and may not be resilient to various workload patterns.
This change introduces a new L1 policy, TinyLfuBlockCache, which records the frequency in a counting sketch, ages periodically by halving the counters, and orders entries by SLRU. An entry is discarded by comparing the frequency of the new arrival to the SLRU's victim, and keeping the one with the highest frequency. This allows the operations to be performed in O(1) time and, though the use of a compact sketch, a much larger history is retained beyond the current working set. In a variety of real world traces the policy had near optimal hit rates.
New configuration variable hfile.block.cache.policy sets the eviction policy for the L1 block cache. The default is “LRU” (LruBlockCache). Set to “TinyLFU” to use TinyLfuBlockCache instead.
Introduced
Future<Void> createTableAsync(TableDescriptor);
Introduced these methods: void move(byte[]); void move(byte[], ServerName); Future<Void> splitRegionAsync(byte[]);
These methods are deprecated: void move(byte[], byte[])
Add a new jenkins file for running pre commit check for GitHub PR.
Add cloneSnapshot/restoreSnapshot with acl methods in AsyncAdmin.
When insufficient permissions, you now get:
HTTP/1.1 403 Forbidden
on the HTTP side, and in the message
Forbidden org.apache.hadoop.hbase.security.AccessDeniedException: org.apache.hadoop.hbase.security.AccessDeniedException: Insufficient permissions for user ‘myuser',action: get, tableName:mytable, family:cf. at org.apache.ranger.authorization.hbase.RangerAuthorizationCoprocessor.authorizeAccess(RangerAuthorizationCoprocessor.java:547) and the rest of the ADE stack
Now we will sort the javac WARNING/ERROR before generating diff in pre-commit so we can get a stable output for the error prone. The downside is that we just sort the output lexicographically so the line number will also be sorted lexicographically, which is a bit strange to human.
Exposes a new configuration property “zookeeper.multi.max.size” which dictates the maximum size of deletes that HBase will make to ZooKeeper in a single RPC. This property defaults to 1MB, which should fall beneath the default ZooKeeper limit of 2MB, controlled by “jute.maxbuffer”.
Fixed awkward dependency issue that prevented site building.
HBase 2.1.4 shipped with an early version of this fix that incorrectly altered the libraries included in our binary assembly for using Apache Hadoop 2.7 (the current build default Hadoop version for 2.1.z). For folks running out of the box against a Hadoop 2.7 cluster (or folks who skip the installation step of replacing the bundled Hadoop libraries) this will result in a failure at Region Server startup due to a missing class definition. e.g.:
2019-03-27 09:02:05,779 ERROR [main] regionserver.HRegionServer: Failed construction RegionServer java.lang.NoClassDefFoundError: org/apache/htrace/SamplerBuilder at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:644) at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:628) at org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:149) at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2667) at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:93) at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2701) at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2683) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:372) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:171) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:356) at org.apache.hadoop.fs.Path.getFileSystem(Path.java:295) at org.apache.hadoop.hbase.util.CommonFSUtils.getRootDir(CommonFSUtils.java:362) at org.apache.hadoop.hbase.util.CommonFSUtils.isValidWALRootDir(CommonFSUtils.java:411) at org.apache.hadoop.hbase.util.CommonFSUtils.getWALRootDir(CommonFSUtils.java:387) at org.apache.hadoop.hbase.regionserver.HRegionServer.initializeFileSystem(HRegionServer.java:704) at org.apache.hadoop.hbase.regionserver.HRegionServer.<init>(HRegionServer.java:613) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.hadoop.hbase.regionserver.HRegionServer.constructRegionServer(HRegionServer.java:3029) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.start(HRegionServerCommandLine.java:63) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.run(HRegionServerCommandLine.java:87) at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) at org.apache.hadoop.hbase.util.ServerCommandLine.doMain(ServerCommandLine.java:149) at org.apache.hadoop.hbase.regionserver.HRegionServer.main(HRegionServer.java:3047) Caused by: java.lang.ClassNotFoundException: org.apache.htrace.SamplerBuilder at java.net.URLClassLoader.findClass(URLClassLoader.java:381) at java.lang.ClassLoader.loadClass(ClassLoader.java:424) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349) at java.lang.ClassLoader.loadClass(ClassLoader.java:357) ... 26 more
Workaround via any one of the following:
hadoop executable is in the PATH seen at Region Server startup and that you are not using the HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP bypass.htrace-core-3.1.0-incubating.jar available to all Region Servers via the HBASE_CLASSPATH environment variable.htrace-core-3.1.0-incubating.jar available to all Region Servers by copying it into the directory ${HBASE_HOME}/lib/client-facing-thirdparty/.Add a listTableDescriptors(List<TableName>) method in the AsyncAdmin interface, to align with the Admin interface.
Deprecate Admin.deleteSnapshot(byte[]), please use the String version instead.
Add a mergeRegionsAsync(byte[][], boolean) method in the AsyncAdmin interface.
Instead of using assert, now we will throw IllegalArgumentException when you want to merge less than 2 regions at client side. And also, at master side, instead of using assert, now we will throw DoNotRetryIOException if you want merge more than 2 regions, since we only support merging two regions at once for now.
Add drainXXX parameter for balancerSwitch/splitSwitch/mergeSwitch methods in the AsyncAdmin interface, which has the same meaning with the synchronous parameter for these methods in the Admin interface.
As of HBase 3.0, the ByteBufferUtils class is now marked as a Private API for internal project use only. Downstream users are advised that it no longer has any compatibility promises across releases.
As of earlier HBase release lines the class is now marked as deprecated to call attention to this planned transition.
bulkload (HFileOutputFormat2) support config the compression on client ,you can set the job configuration “hbase.mapreduce.hfileoutputformat.compression” override the auto-detection of the target table's compression
Add a cloneSnapshotAsync method with restoreAcl parameter. Deprecated restoreSnapshotAsync method as it just ignores the failsafe configuration. Make snapshotAsync method returns a Future<Void>. Deprecated the snapshot related methods which take a ‘byte[]’ as the snapshot name. Use default methods to reduce the code base for implementation classes.
Deprecated AsyncTable.isTableAvailable(TableName, byte[][]).
After HBASE-21871, we can specify a peer table name with --peerTableName in VerifyReplication tool like the following: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable 5 TestTable
In addition, we can compare any 2 tables in any remote clusters with specifying both peerId and --peerTableName.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable zk1,zk2,zk3:2181/hbase TestTable
Adds below flush, split, and compaction metrics
HBASE-21481 improves the quality of access control, by strengthening the protection of super users's privileges.
Now we have four types of RIT procedure metrics, assign, unassign, move, reopen. The meaning of assign/unassign is changed, as we will not increase the unassign metric and then the assign metric when moving a region. Also introduced two new procedure metrics, open and close, which are used to track the open/close region calls to region server. We may send open/close multiple times to finish a RIT since we may retry multiple times.
Problem: This is an old problem since HBASE-2231. The compaction event marker was only writed to WAL. But after flush, the WAL may be archived, which means an useful compaction event marker be deleted, too. So the compacted store files cannot be archived when region open and replay WAL.
Solution: After this jira, the compaction event tracker will be writed to HFile. When region open and load store files, read the compaction evnet tracker from HFile and archive the compacted store files which still exist.
HBase contains two quota scopes: MACHINE and CLUSTER. Before this patch, set quota operations did not expose scope option to client api and use MACHINE as default, CLUSTER scope can not be set and used. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’
This issue implements CLUSTER scope in a simple way: For user, namespace, user over namespace quota, use [ClusterLimit / RSNum] as machine limit. For table and user over table quota, use [ClusterLimit / TotalTableRegionNum * MachineTableRegionNum] as machine limit. After this patch, user can set CLUSTER scope quota, but MACHINE is still default if user ignore scope. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’ set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => MACHINE set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => CLUSTER
Change spotbugs version to 3.1.11.
This modifies “status ‘replication’” output, fixing inconsistencies on the reporting times and ages of last shipped edits, as well as wrong calculation of replication lags.
It also introduces additional info for each recovery queue, which was not accounted by this command before.
The new output for “status ‘replication’” command is explained in details below: a) Source started, target stopped, no edits arrived on source yet: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ... b) Source started, target stopped, add edit on source: ... Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:21:00 GMT 2018, Replication Lag=2459 ... c) Source started, target stopped, edit added on source, restart source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 Recovered Queue: 1-hbase01.home,16020,1542784524057 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:23:00 GMT 2018, Replication Lag=201495 ... d) Source started, target stopped, add edit on source, restart source, add another edit on source: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=6349 Recovered Queue: 1-hbase01.home,16020,1542782758742 No Ops shipped since last restart, SizeOfLogQueue=0, TimeStampOfLastArrivedInSource=Wed Nov 21 06:53:05 GMT 2018, Replication Lag=569394 ... e) Source started, target stopped, add edit on source, restart source, add another edit on source, start target: ... SOURCE: PeerID=1 Normal Queue: 1 AgeOfLastShippedOp=30000, TimeStampOfLastShippedOp=Wed Nov 21 07:07:58 GMT 2018, SizeOfLogQueue=1, TimeStampOfLastArrivedInSource=Wed Nov 21 07:02:28 GMT 2018, Replication Lag=0 ... f) Source started, target stopped, add edit on source, restart source, restart target: ... SOURCE: PeerID=1 Normal Queue: 1 No Ops shipped since last restart, SizeOfLogQueue=1, No edits for this source since it started, Replication Lag=0 ...
Remove bloom filter type ROWPREFIX_DELIMITED. May add it back when find a better solution.
Support enable or disable exceed throttle quota. Exceed throttle quota means, user can over consume user/namespace/table quota if region server has additional available quota because other users don't consume at the same time. Use the following shell commands to enable/disable exceed throttle quota: enable_exceed_throttle_quota disable_exceed_throttle_quota There are two limits when enable exceed throttle quota:
Remove jackson dependencies from most hbase modules except hbase-rest, use shaded gson instead. The output json will be a bit different since jackson can use getter/setter, but gson will always use the fields.
Mark HConstants.META_QOS as deprecated. It is for internal use only, which is the highest priority. You should not try to set a priority greater than or equal to this value, although it is no harm but also useless.
This patch adds the ability to disable split and/or merge for a table (By default, split and merge are enabled for a table).
Allows shell to set Scan options previously not exposed. See additions as part of the scan help by typing following hbase shell:
hbase> help ‘scan’
We can specify peerQuorumAddress instead of peerId in VerifyReplication tool. So it no longer requires peerId to be setup when using this tool.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication zk1,zk2,zk3:2181/hbase testTable
Introduce a VerifyWALEntriesReplicationEndpoint which replicates nothing but only verifies if all the cells are valid. It can be used to capture bugs for writing WAL, as most times we will not read the WALs again after writing it if there are no region server crashes.
Introduced an new config key in this issue: hbase.regionserver.inmemory.compaction.pool.size. the default value would be 10. you can configure this to set the pool size of in-memory compaction pool. Note that all memstores in one region server will share the same pool, so if you have many regions in one region server, you need to set this larger to compact faster for better read performance.
Make StoppedRpcClientException extend DoNotRetryIOException.
To implement user permission control in Precedure V2, move grant and revoke method from AccessController to master firstly. Mark AccessController#grant and AccessController#revoke as deprecated and please use Admin#grant and Admin#revoke instead.
IMPORTANT: Due to security issues, all users who use hbase thrift should avoid using releases which do not have this fix.
The effect releases are: 2.1.x: 2.1.2 and below 2.0.x: 2.0.4 and below 1.x: 1.4.x and below
If you are using the effect releases above, please consider upgrading to a newer release ASAP.
For users who have configured hbase.bucketcache.ioengine with either the file:, files:, or mmap: prefix, and configured it to be persistent via the hbase.bucketcache.persistent.path property, the serialization format of the bucket cache has changed between versions. The old state will not be read during startup, and there is currently no migration path. The impact is expected to be minimal, however, since the cache will rebuild over time as access patterns dictate.
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
See the document http://hbase.apache.org/book.html#upgrade2.2 about how to upgrade from 2.0 or 2.1 to 2.2+.
HBase 2.2+ uses a new Procedure form assiging/unassigning/moving Regions. It does not process HBase 2.1 and 2.0's Unassign/Assign Procedure types. Upgrade requires that we first drain the Master Procedure Store of old style Procedures before starting the new 2.2 Master. So you need to make sure that before you kill the old version (2.0 or 2.1) Master, there is no region in transition. And once the new version (2.2+) Master is up, you can rolling upgrade RegionServers one by one.
And there is a more safer way if you are running 2.1.1+ or 2.0.3+ cluster. It need four steps to upgrade Master.
Then you can rolling upgrade RegionServers one by one. See HBASE-21075 for more details.
Added completebulkload short name for BulkLoadHFilesTool to bin/hbase.
Change the default hadoop-3 version to 3.1.2. Drop the support for the releases which are effected by CVE-2018-8029, see this email https://lists.apache.org/thread.html/3d6831c3893cd27b6850aea2feff7d536888286d588e703c6ffd2e82@%3Cuser.hadoop.apache.org%3E
The CellUtil.setTimestamp method changes to be an API with audience LimitedPrivate(COPROC) in HBase 3.0. With that designation the API should remain stable within a given minor release line, but may change between minor releases.
Previously, this method was deprecated in HBase 2.0 for removal in HBase 3.0. Deprecation messages in HBase 2.y releases have been updated to indicate the expected API audience change.
The class LossyCounting was unintentionally marked Public but was never intended to be part of our public API. This oversight has been corrected and LossyCounting is now marked as Private and going forward may be subject to additional breaking changes or removal without notice. If you have taken a dependency on this class we recommend cloning it locally into your project before upgrading to this release.
Warnings for level headings are corrected in the book for the HBase Incompatibilities section.
Add hadoop 3.0.3, 3.1.1 3.1.2 in our hadoop check jobs.
The DumpReplicationQueues tool will now list replication queues sorted in chronological order.
Fixes a formatting issue in the administration section of the book, where listing indentation were a little bit off.
Now the default hadoop-two.version has been changed to 2.8.5, and all hadoop versions before 2.8.2(exclude) will not be supported any more.
Removed extra + in HRegion, HStore and LoadIncrementalHFiles for branch-2 and HRegion and HStore for branch-1.
Updated metrics core from 3.2.1 to 3.2.6.
The rubocop definition for the maximum method length was set to 75.
Fixes the formatting of the “Voting on Release Candidates” to actually show the quote and code formatting of the RAT check.
The rubocop configuration in the hbase-shell module now allows a line length with 100 characters, instead of 80 as before. For everything before 2.1.5 this change introduces rubocop itself.
This change allows the system and superusers to initiate compactions, even when a space quota violation policy disallows compactions from happening. The original intent behind disallowing of compactions was to prevent end-user compactions from creating undue I/O load, not disallowing *any* compaction in the system.
Adds new configuration hbase.client.failure.map.cleanup.interval which defaults to ten minutes.
Updates libs used internally by hbase via hbase-thirdparty as follows:
gson 2.8.1 -\> 2.8.5 guava 22.0 -\> 27.1-jre pb 3.5.1 -\> 3.7.0 netty 4.1.17 -\> 4.1.34 commons-collections4 4.1 -\> 4.3
Introduced
Future<Void> createTableAsync(TableDescriptor);
Introduced these methods: void move(byte[]); void move(byte[], ServerName); Future<Void> splitRegionAsync(byte[]);
These methods are deprecated: void move(byte[], byte[])
Add a new jenkins file for running pre commit check for GitHub PR.
Add cloneSnapshot/restoreSnapshot with acl methods in AsyncAdmin.
When insufficient permissions, you now get:
HTTP/1.1 403 Forbidden
on the HTTP side, and in the message
Forbidden org.apache.hadoop.hbase.security.AccessDeniedException: org.apache.hadoop.hbase.security.AccessDeniedException: Insufficient permissions for user ‘myuser',action: get, tableName:mytable, family:cf. at org.apache.ranger.authorization.hbase.RangerAuthorizationCoprocessor.authorizeAccess(RangerAuthorizationCoprocessor.java:547) and the rest of the ADE stack
Now we will sort the javac WARNING/ERROR before generating diff in pre-commit so we can get a stable output for the error prone. The downside is that we just sort the output lexicographically so the line number will also be sorted lexicographically, which is a bit strange to human.
Exposes a new configuration property “zookeeper.multi.max.size” which dictates the maximum size of deletes that HBase will make to ZooKeeper in a single RPC. This property defaults to 1MB, which should fall beneath the default ZooKeeper limit of 2MB, controlled by “jute.maxbuffer”.
Fixed awkward dependency issue that prevented site building.
HBase 2.1.4 shipped with an early version of this fix that incorrectly altered the libraries included in our binary assembly for using Apache Hadoop 2.7 (the current build default Hadoop version for 2.1.z). For folks running out of the box against a Hadoop 2.7 cluster (or folks who skip the installation step of replacing the bundled Hadoop libraries) this will result in a failure at Region Server startup due to a missing class definition. e.g.:
2019-03-27 09:02:05,779 ERROR [main] regionserver.HRegionServer: Failed construction RegionServer java.lang.NoClassDefFoundError: org/apache/htrace/SamplerBuilder at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:644) at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:628) at org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:149) at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2667) at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:93) at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2701) at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2683) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:372) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:171) at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:356) at org.apache.hadoop.fs.Path.getFileSystem(Path.java:295) at org.apache.hadoop.hbase.util.CommonFSUtils.getRootDir(CommonFSUtils.java:362) at org.apache.hadoop.hbase.util.CommonFSUtils.isValidWALRootDir(CommonFSUtils.java:411) at org.apache.hadoop.hbase.util.CommonFSUtils.getWALRootDir(CommonFSUtils.java:387) at org.apache.hadoop.hbase.regionserver.HRegionServer.initializeFileSystem(HRegionServer.java:704) at org.apache.hadoop.hbase.regionserver.HRegionServer.<init>(HRegionServer.java:613) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.hadoop.hbase.regionserver.HRegionServer.constructRegionServer(HRegionServer.java:3029) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.start(HRegionServerCommandLine.java:63) at org.apache.hadoop.hbase.regionserver.HRegionServerCommandLine.run(HRegionServerCommandLine.java:87) at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) at org.apache.hadoop.hbase.util.ServerCommandLine.doMain(ServerCommandLine.java:149) at org.apache.hadoop.hbase.regionserver.HRegionServer.main(HRegionServer.java:3047) Caused by: java.lang.ClassNotFoundException: org.apache.htrace.SamplerBuilder at java.net.URLClassLoader.findClass(URLClassLoader.java:381) at java.lang.ClassLoader.loadClass(ClassLoader.java:424) at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349) at java.lang.ClassLoader.loadClass(ClassLoader.java:357) ... 26 more
Workaround via any one of the following:
hadoop executable is in the PATH seen at Region Server startup and that you are not using the HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP bypass.htrace-core-3.1.0-incubating.jar available to all Region Servers via the HBASE_CLASSPATH environment variable.htrace-core-3.1.0-incubating.jar available to all Region Servers by copying it into the directory ${HBASE_HOME}/lib/client-facing-thirdparty/.Add a listTableDescriptors(List<TableName>) method in the AsyncAdmin interface, to align with the Admin interface.
Add a mergeRegionsAsync(byte[][], boolean) method in the AsyncAdmin interface.
Instead of using assert, now we will throw IllegalArgumentException when you want to merge less than 2 regions at client side. And also, at master side, instead of using assert, now we will throw DoNotRetryIOException if you want merge more than 2 regions, since we only support merging two regions at once for now.
Add drainXXX parameter for balancerSwitch/splitSwitch/mergeSwitch methods in the AsyncAdmin interface, which has the same meaning with the synchronous parameter for these methods in the Admin interface.
bulkload (HFileOutputFormat2) support config the compression on client ,you can set the job configuration “hbase.mapreduce.hfileoutputformat.compression” override the auto-detection of the target table's compression
Deprecated AsyncTable.isTableAvailable(TableName, byte[][]).
After HBASE-21871, we can specify a peer table name with --peerTableName in VerifyReplication tool like the following: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable 5 TestTable
In addition, we can compare any 2 tables in any remote clusters with specifying both peerId and --peerTableName.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication --peerTableName=peerTable zk1,zk2,zk3:2181/hbase TestTable
Adds below flush, split, and compaction metrics
From 2.2.0, hbase supports client login via keytab. To use this feature, client should specify `hbase.client.keytab.file` and `hbase.client.keytab.principal` in hbase-site.xml, then the connection will contain the needed credentials which be renewed periodically to communicate with kerberized hbase cluster.
After HBASE-21410, we add a helper page to Master UI. This helper page is mainly to help HBase operator quickly found all regions and pids that are get stuck. There are 2 entries to get in this page. One is showing in the Regions in Transition section, it made “num region(s) in transition” a link that you can click and check all regions in transition and their related procedure IDs. The other one is showing in the table details section, it made the number of CLOSING or OPENING regions a link, which you can click and check regions and related procedure IDs of CLOSING or OPENING regions of a certain table. In this helper page, not only you can see all regions and related procedures, there are 2 buttons at the top which will show these regions or procedure IDs in text format. This is mainly aim to help operator to easily copy and paste all problematic procedure IDs and encoded region names to HBCK2's command line, by which we HBase operator can bypass these procedures or assign these regions.
After HBASE-21588, we introduce a new way to do WAL splitting coordination by procedure framework. This can simplify the process of WAL splitting and no need to connect zookeeper any more. During ServerCrashProcedure, it will create a SplitWALProcedure for each WAL that need to split. Then each SplitWALProcedure will spawn a SplitWALRemoteProcedure to send the request to regionserver. At the RegionServer side, whole process is handled by SplitWALCallable. It split the WAL and return the result to master. According to my test, this patch has a better performance as the number of WALs that need to split increase. And it can relieve the pressure on zookeeper.
Previously the recovered.edits directory was under the root directory. This JIRA moves the recovered.edits directory to be under the hbase.wal.dir if set. It also adds a check for any recovered.edits found under the root directory for backwards compatibility. This gives improvements when a faster media(like SSD) or more local FileSystem is used for the hbase.wal.dir than the root dir.
When oldwals (and hfile) cleaner cleans stale wals (and hfiles), it will periodically check and wait the clean results from filesystem, the total wait time will be no more than a max time.
The periodically wait and check configurations are hbase.oldwals.cleaner.thread.check.interval.msec (default is 500 ms) and hbase.regionserver.hfilecleaner.thread.check.interval.msec (default is 1000 ms).
Meanwhile, The max time configurations are hbase.oldwals.cleaner.thread.timeout.msec and hbase.regionserver.hfilecleaner.thread.timeout.msec, they are set to 60 seconds by default.
All support dynamic configuration.
e.g. in the oldwals cleaning scenario, one may consider tuning hbase.oldwals.cleaner.thread.timeout.msec and hbase.oldwals.cleaner.thread.check.interval.msec
HBASE-21481 improves the quality of access control, by strengthening the protection of super users's privileges.
Now we have four types of RIT procedure metrics, assign, unassign, move, reopen. The meaning of assign/unassign is changed, as we will not increase the unassign metric and then the assign metric when moving a region. Also introduced two new procedure metrics, open and close, which are used to track the open/close region calls to region server. We may send open/close multiple times to finish a RIT since we may retry multiple times.
Problem: This is an old problem since HBASE-2231. The compaction event marker was only writed to WAL. But after flush, the WAL may be archived, which means an useful compaction event marker be deleted, too. So the compacted store files cannot be archived when region open and replay WAL.
Solution: After this jira, the compaction event tracker will be writed to HFile. When region open and load store files, read the compaction evnet tracker from HFile and archive the compacted store files which still exist.
HBase contains two quota scopes: MACHINE and CLUSTER. Before this patch, set quota operations did not expose scope option to client api and use MACHINE as default, CLUSTER scope can not be set and used. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’
This issue implements CLUSTER scope in a simple way: For user, namespace, user over namespace quota, use [ClusterLimit / RSNum] as machine limit. For table and user over table quota, use [ClusterLimit / TotalTableRegionNum * MachineTableRegionNum] as machine limit. After this patch, user can set CLUSTER scope quota, but MACHINE is still default if user ignore scope. Shell commands are as follows: set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’ set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => MACHINE set_quota, TYPE => THROTTLE, TABLE => ‘t1’, LIMIT => ‘10req/sec’, SCOPE => CLUSTER
Change spotbugs version to 3.1.11.
Remove bloom filter type ROWPREFIX_DELIMITED. May add it back when find a better solution.
Support enable or disable exceed throttle quota. Exceed throttle quota means, user can over consume user/namespace/table quota if region server has additional available quota because other users don't consume at the same time. Use the following shell commands to enable/disable exceed throttle quota: enable_exceed_throttle_quota disable_exceed_throttle_quota There are two limits when enable exceed throttle quota:
Remove jackson dependencies from most hbase modules except hbase-rest, use shaded gson instead. The output json will be a bit different since jackson can use getter/setter, but gson will always use the fields.
Mark HConstants.META_QOS as deprecated. It is for internal use only, which is the highest priority. You should not try to set a priority greater than or equal to this value, although it is no harm but also useless.
This patch adds the ability to disable split and/or merge for a table (By default, split and merge are enabled for a table).
Allows shell to set Scan options previously not exposed. See additions as part of the scan help by typing following hbase shell:
hbase> help ‘scan’
We can specify peerQuorumAddress instead of peerId in VerifyReplication tool. So it no longer requires peerId to be setup when using this tool.
For example: hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication zk1,zk2,zk3:2181/hbase testTable
Introduce a VerifyWALEntriesReplicationEndpoint which replicates nothing but only verifies if all the cells are valid. It can be used to capture bugs for writing WAL, as most times we will not read the WALs again after writing it if there are no region server crashes.
Deprecated HBaseConfiguration#getInt(Configuration, String, String, int) method and removed it from 3.0.0 version.
Introduced an new config key in this issue: hbase.regionserver.inmemory.compaction.pool.size. the default value would be 10. you can configure this to set the pool size of in-memory compaction pool. Note that all memstores in one region server will share the same pool, so if you have many regions in one region server, you need to set this larger to compact faster for better read performance.
Make StoppedRpcClientException extend DoNotRetryIOException.
To implement user permission control in Precedure V2, move grant and revoke method from AccessController to master firstly. Mark AccessController#grant and AccessController#revoke as deprecated and please use Admin#grant and Admin#revoke instead.
IMPORTANT: Due to security issues, all users who use hbase thrift should avoid using releases which do not have this fix.
The effect releases are: 2.1.x: 2.1.2 and below 2.0.x: 2.0.4 and below 1.x: 1.4.x and below
If you are using the effect releases above, please consider upgrading to a newer release ASAP.
HTableMultiplexer exposes the implementation class, and it is incomplete, so we mark it as deprecated and remove it in 3.0.0 release.
There is no direct replacement for HTableMultiplexer, please use BufferedMutator if you want to batch mutations to a table.
Introduce a BulkLoadHFiles interface which is marked as IA.Public, for doing bulk load programmatically. Introduce a BulkLoadHFilesTool which extends BulkLoadHFiles, and is marked as IA.LimitedPrivate(TOOLS), for using from command line. The old LoadIncrementalHFiles is deprecated and will be removed in 3.0.0.
Move the two getHbck method from ClusterConnection to Connection, and mark the methods as IA.LimitedPrivate(HBCK), as ClusterConnection is IA.Private and should not be depended by HBCK2.
Add a clearRegionLocationCache method in Connection to clear the region location cache for all the tables. As in RegionLocator, most of the methods have a ‘reload’ parameter, which implicitly tells user that we have a region location cache, so adding a method to clear the cache is fine.
Support set region server rpc throttle quota which represents the read/write ability of region servers and throttles when region server's total requests exceeding the limit.
Use the following shell command to set RS quota: set_quota TYPE => THROTTLE, REGIONSERVER => ‘all’, THROTTLE_TYPE => WRITE, LIMIT => ‘20000req/sec’ set_quota TYPE => THROTTLE, REGIONSERVER => ‘all’, LIMIT => NONE “all” represents the throttle quota of all region servers and setting specified region server quota isn't supported currently.
In shell commands “describe_namespace” and “describe”, which are used to see the descriptors of the namespaces and tables respectively, quotas set on that particular namespace/table will also be printed along.
Adds shell support for the following:
After HBASE-21620, the filterListWithOR has been a bit slow because we need to merge each sub-filter‘s RC , while before HBASE-21620, we will skip many RC merging, but the logic was wrong. So here we choose another way to optimaze the performance: removing the KeyValueUtil#toNewKeyCell. Anoop Sam John suggested that the KeyValueUtil#toNewKeyCell can save some GC before because if we copy key part of cell into a single byte[], then the block the cell refering won’t be refered by the filter list any more, the upper layer can GC the data block quickly. while after HBASE-21620, we will update the prevCellList for every encountered cell now, so the lifecycle of cell in prevCellList for FilterList will be quite shorter. so just use the cell ref for saving cpu. BTW, we removed all the arrays streams usage in filter list, because it's also quite time-consuming in our test.
We found the memstore snapshotting would cost much time because of calling the time-consuming ConcurrentSkipListMap#Size, it would make the p999 latency spike happen. So in this issue, we remove all ConcurrentSkipListMap#size in memstore by counting the cellsCount in MemstoreSizeing. As the issue described, the p999 latency spike was mitigated.
Provides a new throttle type: capacity unit. One read/write/request capacity unit represents that read/write/read+write up to 1K data. If data size is more than 1K, then consume additional capacity units.
Use shell command to set capacity unit(CU): set_quota TYPE => THROTTLE, THROTTLE_TYPE => WRITE, USER => ‘u1’, LIMIT => ‘10CU/sec’
Use the “hbase.quota.read.capacity.unit” property to set the data size of one read capacity unit in bytes, the default value is 1K. Use the “hbase.quota.write.capacity.unit” property to set the data size of one write capacity unit in bytes, the default value is 1K.
Does thread dump on stdout on abort.
Now all the Enum configs in ColumnFamilyDescriptor can accept lower case config value.
Python3 support was added to dev-support/submit-patch.py. To install newly required dependencies run `pip install -r dev-support/python-requirements.txt` command.
In HBASE-21657, I simplified the path of estimatedSerialiedSize() & estimatedSerialiedSizeOfCell() by moving the general getSerializedSize() and heapSize() from ExtendedCell to Cell interface. The patch also included some other improvments:
We gain almost 40% throughput improvement in 100% scan case for branch-2 (cacheHitRatio100%)[1], it‘s a good thing. While it’s a incompatible change in some case, such as if the upstream user implemented their own Cells, although it's rare but can happen, then their compile will be error.
Adds task monitor that shows ServerCrashProcedure progress in UI.
Before this issue, thrift1 server and thrift2 server are totally different servers. If a new feature is added to thrift1 server, thrfit2 server have to make the same change to support it(e.g. authorization). After this issue, thrift2 server is inherited from thrift1, thrift2 server now have all the features thrift1 server has(e.g http support, which thrift2 server doesn't have before). The way to start thrift1 or thrift2 server remain the same after this issue.
ThriftAdmin/ThriftTable are implemented based on Thrift2. With ThriftAdmin/ThriftTable, People can use thrift2 protocol just like HTable/HBaseAdmin. Example of using ThriftConnection Configuration conf = HBaseConfiguration.create(); conf.set(ClusterConnection.HBASE_CLIENT_CONNECTION_IMPL,ThriftConnection.class.getName()); Connection conn = ConnectionFactory.createConnection(conf); Table table = conn.getTable(tablename) It is just like a normal Connection, similar use experience with the default ConnectionImplementation
There was a bug when scan with the same startRow(inclusive=true) and stopRow(inclusive=false). The old incorrect behavior is return one result. After this fix, the new correct behavior is return nothing.
Support enable or disable rpc throttle when hbase quota is enabled. If hbase quota is enabled, rpc throttle is enabled by default. When disable rpc throttle, HBase will not throttle any request. Use the following commands to switch rpc throttle : enable_rpc_throttle / disable_rpc_throttle.
Add a new configuration “hbase.skip.load.duplicate.table.coprocessor”. The default value is false to keep compatible with the old behavior. Config it true to skip load duplicate table coprocessor.
Added DDL operations and some other structure definition to thrift2. Methods added: create/modify/addColumnFamily/deleteColumnFamily/modifyColumnFamily/enable/disable/truncate/delete table create/modify/delete namespace get(list)TableDescriptor(s)/get(list)NamespaceDescirptor(s) tableExists/isTableEnabled/isTableDisabled/isTableAvailabe And some class definitions along with those methods
Deprecated region coprocessor postMutationBeforeWAL and introduce two new region coprocessor postIncrementBeforeWAL and postAppendBeforeWAL instead.
Use de.skuzzle.enforcer.restrict-imports-enforcer-rule extension for maven enforcer plugin to ban illegal imports at compile time. Now if you use illegal imports, for example, import com.google.common.*, there will be a compile error, instead of a checkstyle warning.
Add a sanity check when constructing KeyValue from a byte[]. we use the constructor when we‘re reading kv from socket or HFIle or WAL(replication). the santiy check isn’t designed for discovering the bits corruption in network transferring or disk IO. It is designed to detect bugs inside HBase in advance. and HBASE-21459 indicated that there's extremely small performance loss for diff kinds of keyvalue.
The replication UI on master will show the replication endpoint classname.
Add a SERIAL flag for add_peer command to identifiy whether or not the replication peer is a serial replication peer. The default serial flag is false.
Log level of ReadOnlyZKClient moved to debug.
The HBase shell now includes a command to list regions currently in transition.
HBase Shell Use "help" to get list of supported commands. Use "exit" to quit this interactive shell. Version 1.5.0-SNAPSHOT, r9bb6d2fa8b760f16cd046657240ebd4ad91cb6de, Mon Oct 8 21:05:50 UTC 2018 hbase(main):001:0> help 'rit' List all regions in transition. Examples: hbase> rit hbase(main):002:0> create ... 0 row(s) in 2.5150 seconds => Hbase::Table - IntegrationTestBigLinkedList hbase(main):003:0> rit 0 row(s) in 0.0340 seconds hbase(main):004:0> unassign '56f0c38c81ae453d19906ce156a2d6a1' 0 row(s) in 0.0540 seconds hbase(main):005:0> rit IntegrationTestBigLinkedList,L\xCC\xCC\xCC\xCC\xCC\xCC\xCB,1539117183224.56f0c38c81ae453d19906ce156a2d6a1. state=PENDING_CLOSE, ts=Tue Oct 09 20:33:34 UTC 2018 (0s ago), server=null 1 row(s) in 0.0170 seconds
Allow passing of -Dkey=value option to shell to override hbase-* configuration: e.g.:
$ ./bin/hbase shell -Dhbase.zookeeper.quorum=ZK0.remote.cluster.example.org,ZK1.remote.cluster.example.org,ZK2.remote.cluster.example.org -Draining=false ... hbase(main):001:0> @shell.hbase.configuration.get(“hbase.zookeeper.quorum”) => “ZK0.remote.cluster.example.org,ZK1.remote.cluster.example.org,ZK2.remote.cluster.example.org” hbase(main):002:0> @shell.hbase.configuration.get(“raining”) => “false”
Incompatible change. Allow MasterObserver#preModifyTable to return a new TableDescriptor. And master will use this returned TableDescriptor to modify table.
HBase clusters will experience Region Server failures due to out of memory errors due to a leak given any of the following:
When there are long running scans the Region Server process attempts to optimize access by using a different API geared towards sequential access. Due to an error in HBASE-20704 for HBase 2.0+ the Region Server fails to release related resources when those scans finish. That same optimization path is always used for the HBase internal file compaction process.
Impact for this error can be minimized by setting the config value “hbase.storescanner.pread.max.bytes” to MAX_INT to avoid the optimization for default user scans. Clients should also be checked to ensure they do not pass the STREAM read type to the Scan API. This will have a severe impact on performance for long scans.
Compactions always use this sequential optimized reading mechanism so downstream users will need to periodically restart Region Server roles after compactions have happened.
Add a new method preCreateTableRegionInfos for MasterObserver, which will be called before creating region infos for the given table, before the preCreateTable method. It allows you to return a new TableDescritor to override the original one. Returns null or throws exception will stop the creation.
After HBASE-21492 the return type of WALCellCodec#getWALCellCodecClass has been changed from String to Class
To prevent race condition between in progress snapshot (performed by TakeSnapshotHandler) and HFileCleaner which results in data loss, this JIRA introduced mutual exclusion between taking snapshot and running HFileCleaner. That is, at any given moment, either some snapshot can be taken or, HFileCleaner checks hfiles which are not referenced, but not both can be running.
Changes group name of hbase metrics from “HBase Counters” to “HBaseCounters”.
Parent issue moved hbase-spark* modules to hbase-connectors. This issue removes hbase-spark* modules from hbase core repo.
hbase-spark* modules have been cloned to https://github.com/apache/hbase-connectors All spark connector dev is to happen in that repo from here on out.
Let me file a subtask to remove hbase-spark* modules from hbase core.
Add -Djdk.net.URLClassPath.disableClassPathURLCheck=true when executing surefire plugin.
Puts master startup into holding pattern if meta is not assigned (previous it would exit). To make progress again, operator needs to inject an assign (Caveats and instruction can be found in HBASE-21035).
Adds scheduleServerCrashProcedure to the HbckService.
Add two new config hbase.regionserver.abort.timeout and hbase.regionserver.abort.timeout.task. If regionserver abort timeout, it will schedule an abort timeout task to run. The default abort task is SystemExitWhenAbortTimeout, which will force to terminate region server when abort timeout. And you can config a special abort timeout task by hbase.regionserver.abort.timeout.task.
Adds to bin/hbase means of invoking hbck2. Pass the new ‘-j’ option on the ‘hbck’ command with a value of the full path to the HBCK2.jar.
E.g:
$ ./bin/hbase hbck -j ~/checkouts/hbase-operator-tools/hbase-hbck2/target/hbase-hbck2-1.0.0-SNAPSHOT.jar setTableState x ENABLED
Retry assigns ‘forever’ (or until an intervention such as a ServerCrashProcedure).
Previous retry was a maximum of ten times but on failure, handling was an indeterminate.
The description claims the balancer not dynamically configurable but this is an error; it is http://hbase.apache.org/book.html#dyn_config
Also, if balancer is seen to be cutting out too soon, try setting “hbase.master.balancer.stochastic.runMaxSteps” to true.
Adds cleaner logging around balancer start.
HBASE-21073 | Major | “Maintenance mode” master
Instead of being an ephemeral state set by hbck, maintenance mode is now an explicit toggle set by either configuration property or environment variable. In maintenance mode, master will host system tables and not assign any user-space tables to RSs. This gives operators the ability to affect repairs to meta table with fewer moving parts.
Changed waitTime parameter to lockWait on bypass. Changed default waitTime from 0 -- i.e. wait for ever -- to 1ms so if lock is held, we'll go past it and if override enforce bypass.
bypass will now throw an Exception if passed a lockWait <= 0; i.e bypass will prevent an operator getting stuck on an entity lock waiting forever (lockWait == 0)
Cleans up usage and docs around Canary. Does not change command-line args (though we should -- smile).
For the sub procedures which are successfully finished, do not do rollback. This is a change in rollback behavior.
State changes which are done by sub procedures should be handled by parent procedures when rolling back. For example, when rolling back a MergeTableProcedure, we will schedule new procedures to bring the offline regions online instead of rolling back the original procedures which off-lined the regions (in fact these procedures can not be rolled back...).
Scans that make use of QualifierFilter previously would erroneously return both columns with an empty qualifier along with those that matched. After this change that behavior has changed to only return those columns that match.
It is recommended to place the working directory on-cluster on HDFS as doing so has shown a strong performance increase due to data locality. It is important to note that the working directory should not overlap with any existing directories as the working directory will be cleaned out during the snapshot process. Beyond that, any well-named directory on HDFS should be sufficient.
This adds two extra features to WALPrettyPrinter tool:
Output for each cell combined size of cell descriptors, plus the cell value itself, in a given WAL edit. This is printed on the results as “cell total size sum:” info by default;
An optional -g/--goto argument, that allows to seek straight to that specific WAL file position, then sequentially reading the WAL from that point towards its end;
Local HBase cluster (as used by unit tests) wait times on startup and initialization can be configured via `hbase.master.start.timeout.localHBaseCluster` and `hbase.master.init.timeout.localHBaseCluster`
Adds anchors #tables, #tasks, etc.
Add table state column to the tables panel
Removed the abort_procedure command from shell -- dangerous -- and deprecated abortProcedure in Admin API.
Add two bloom filter type : ROWPREFIX_FIXED_LENGTH and ROWPREFIX_DELIMITED
Adds ‘raw’ assigns/unassigns to the Hbck Service. Takes a list of encoded region names and bulk assigns/unassigns. Skirts Master ‘state’ check and does not invoke Coprocessors. For repair only.
Here is what HBCK2 usage looks like now:
{code} $ java -cp hbase-hbck2-1.0.0-SNAPSHOT.jar org.apache.hbase.HBCK2 usage: HBCK2 <OPTIONS> COMMAND [<ARGS>]
Options: -d,--debug run with debug output -h,--help output this help message --hbase.zookeeper.peerport peerport of target hbase ensemble --hbase.zookeeper.quorum ensemble of target hbase --zookeeper.znode.parent parent znode of target hbase
Commands: setTableState <TABLENAME> <STATE> Possible table states: ENABLED, DISABLED, DISABLING, ENABLING To read current table state, in the hbase shell run: hbase> get ‘hbase:meta’, ‘<TABLENAME>’, ‘table:state’ A value of \x08\x00 == ENABLED, \x08\x01 == DISABLED, etc. An example making table name ‘user’ ENABLED: $ HBCK2 setTableState users ENABLED Returns whatever the previous table state was.
assign <ENCODED_REGIONNAME> ... A ‘raw’ assign that can be used even during Master initialization. Skirts Coprocessors. Pass one or more encoded RegionNames: e.g. 1588230740 is hard-coded encoding for hbase:meta region and de00010733901a05f5a2a3a382e27dd4 is an example of what a random user-space encoded Region name looks like. For example: $ HBCK2 assign 1588230740 de00010733901a05f5a2a3a382e27dd4 Returns the pid of the created AssignProcedure or -1 if none.
unassign <ENCODED_REGIONNAME> ... A ‘raw’ unassign that can be used even during Master initialization. Skirts Coprocessors. Pass one or more encoded RegionNames: Skirts Coprocessors. Pass one or more encoded RegionNames: de00010733901a05f5a2a3a382e27dd4 is an example of what a random user-space encoded Region name looks like. For example: $ HBCK2 unassign 1588230740 de00010733901a05f5a2a3a382e27dd4 Returns the pid of the created UnassignProcedure or -1 if none. {code}
This change ensures Append operations are assembled into the expected order.
Make it so can run the WAL parse and load system in isolation. Here is an example:
{code}$ HBASE_OPTS=" -XX:+UnlockDiagnosticVMOptions -XX:+UnlockCommercialFeatures -XX:+FlightRecorder -XX:+DebugNonSafepoints" ./bin/hbase org.apache.hadoop.hbase.procedure2.store.wal.WALProcedureStore ~/big_set_of_masterprocwals/ {code}
Add a new nettyDirectMemoryUsage under server's ipc metrics to show direct memory usage for netty rpc server.
Client facing artifacts are now built whenever Maven is run through the “package” goal. Previously, the client facing artifacts would create placeholder jars that skipped repackaging HBase and third-party dependencies unless the “release” profile was active.
Build times may be noticeably longer depending on your build hardware. For example, the Jenkins worker nodes maintained by ASF Infra take ~14% longer to do a full packaging build. An example portability-focused personal laptop took ~25% longer.
Allows configuration of the length of RPC messages printed to the log at TRACE level via “hbase.ipc.trace.param.size” in RpcServer.
Users who have previously made use of prefix tree encoding can now check that their existing HFiles no longer contain data that uses it with an additional preupgrade check command.
hbase pre-upgrade validate-hfile
Please see the “HFile Content validation” section of the ref guide's coverage of the pre-upgrade validator tool for usage details.
Adds an HBCK Service and a first method to force-change-in-table-state for use by an HBCK client effecting ‘repair’ to a malfunctioning HBase.
Cleanup all the cluster start override combos in HBaseTestingUtility by adding a StartMiniClusterOption and Builder.
Fence out hbase-1.x hbck1 instances. Stop them making state changes on an hbase-2.x cluster; they could do damage. We do this by writing the hbck1 lock file into place on hbase-2.x Master start-up.
To disable this new behavior, set hbase.write.hbck1.lock.file to false
Introduced a new TransitRegionStateProcedure to replace the old AssignProcedure/UnassignProcedure/MoveRegionProcedure. In the old code, MRP will not be attached to RegionStateNode, so it can not be interrupted by ServerCrashProcedure, which introduces lots of tricky code to deal with races, and also causes lots of other difficulties on how to prevent scheduling redundant or even conflict procedures for a region.
And now TRSP is the only one procedure which can bring region online or offline. When you want to schedule one, you need to check whether there is already one attached to the RegionStateNode, under the lock of the RegionStateNode. If not just go ahead, and if there is one, then you should do something, for example, give up and fail directly, or tell the TRSP to give up(This is what SCP does). Since the check and attach are both under the lock of RSN, it will greatly reduce the possible races, and make the code much simpler.
HFiles generated by 2.0.0, 2.0.1, 2.1.0 are not forward compatible to 1.4.6-, 1.3.2.1-, 1.2.6.1-, and other inactive releases. Why HFile lose compatability is hbase in new versions (2.0.0, 2.0.1, 2.1.0) use protobuf to serialize/deserialize TimeRangeTracker (TRT) while old versions use DataInput/DataOutput. To solve this, We have to put HBASE-21012 to 2.x and put HBASE-21013 in 1.x. For more information, please check HBASE-21008.
After HBASE-20965, we can use MasterFifoRpcScheduler in master to separate RegionServerReport requests to indenpedent handler. To use this feature, please set “hbase.master.rpc.scheduler.factory.class” to “org.apache.hadoop.hbase.ipc.MasterFifoRpcScheduler”. Use “hbase.master.server.report.handler.count” to set RegionServerReport handlers count, the default value is half of “hbase.regionserver.handler.count” value, but at least 1, and the other handlers count in master is “hbase.regionserver.handler.count” value minus RegionServerReport handlers count, but at least 1 too.
In previous releases, when a Space Quota was configured on a table or namespace and that table or namespace was deleted, the Space Quota was also deleted. This change improves the implementation so that the same is also done for RPC Quotas.
After HBASE-20986, we can set different value to block size of WAL and recovered edits. Both of their default value is 2 * default HDFS blocksize. And hbase.regionserver.recoverededits.blocksize is for block size of recovered edits while hbase.regionserver.hlog.blocksize is for block size of WAL.
With this change if a WAL's meta provider (hbase.wal.meta_provider) is not explicitly set, it now defaults to whatever hbase.wal.provider is set to. Previous, the two settings operated independently, each with its own default.
This change is operationally incompatible with previous HBase versions because the default WAL meta provider no longer defaults to AsyncFSWALProvider but to hbase.wal.provider.
The thought is that this is more in line with an operator's expectation, that a change in hbase.wal.provider is sufficient to change how WALs are written, especially given hbase.wal.meta_provider is an obscure configuration and that the very idea that meta regions would have their own wal provider would likely come as a surprise.
Update hadoop-two.version to 2.7.7 and hadoop-three.version to 3.0.3 due to a JDK issue which is solved by HADOOP-15473.
Make hasLock method final, and add a locked field in Procedure to record whether we have the lock. We will set it to true in doAcquireLock and to false in doReleaseLock. The sub procedures do not need to manage it any more.
Also added a locked field in the proto message. When storing, the field will be set according to the return value of hasLock. And when loading, there is a new field in Procedure called lockedWhenLoading. We will set it to true if the locked field in proto message is true.
The reason why we can not set the locked field directly to true by calling doAcquireLock is that, during initialization, most procedures need to wait until master is initialized. So the solution here is that, we introduced a new method called waitInitialized in Procedure, and move the wait master initialized related code from acquireLock to this method. And we added a restoreLock method to Procedure, if lockedWhenLoading is true, we will call the acquireLock to get the lock, but do not set locked to true. And later when we call doAcquireLock and pass the waitInitialized check, we will test lockedWhenLoading, if it is true, when we just set the locked field to true and return, without actually calling the acquireLock method since we have already called it once.
Exposing 2 new metrics in HBase to provide ReadRequestRate and WriteRequestRate at region server level. These metrics give the rate of request handled by the region server and are reset after every monitoring interval.
Added a new command to the shell to switch on/off compactions called “compaction_switch”. Disabling compactions will interrupt any currently ongoing compactions. This setting will be lost on restart of the server. Added the configuration hbase.regionserver.compaction.enabled so user can enable/disable compactions via hbase-site.xml.
Class org.apache.hadoop.hbase.util.Base64 has been removed in it's entirety from HBase 2+. In HBase 1, unused methods have been removed from the class and the audience was changed from Public to Private. This class was originally intended as an internal utility class that could be used externally but thinking since changed; these classes should not have been advertised as public to end-users.
This represents an incompatible change for users who relied on this implementation. An alternative implementation for affected clients is available at java.util.Base64 when using Java 8 or newer; be aware, it may encode/decode differently. For clients seeking to restore this specific implementation, it is available in the public domain for download at http://iharder.sourceforge.net/current/java/base64/
This enhances the AccessControlClient APIs to retrieve the permissions based on namespace, table name, family and qualifier for specific user. AccessControlClient can also validate a user whether allowed to perform specified operations on a particular table. Following APIs have been added,
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
After HBASE-20691 we have changed the default setting of hbase.wal.storage.policy from “HOT” back to “NONE” which means we defer the policy to HDFS. This fixes the problem of release 2.0.0 that the storage policy of WAL directory will defer to HDFS and may not be “HOT” even if you explicitly set hbase.wal.storage.policy to “HOT”
As we hack into the internal of DFSClient when implementing AsyncFSWAL to get better performance, a patch release of hadoop can make it broken.
So now, if user does not specify a wal provider, then we will first try to use ‘asyncfs’, i.e, the AsyncFSWALProvider. If we fail due to some compatible issues, we will fallback to ‘filesystem’, i.e, FSHLog.
After HBASE-20193, we add a section to web ui to show the replication status of each wal group. There are 2 parts of this section, they both show the peerId, wal group and current replicating log of each replication source. And one is showing the information of replication log queue, i.e. size of current log, log queue size and replicating offset. The other one is showing the delay of replication, i.e. last shipped age and replication delay. If the offset shows -1 and replication delay is UNKNOWN, that means replication is not started. This may be caused by this peer is disabled or the replicationEndpoint is sleeping due to some reason.
Now we have a ‘basically work’ solution for rolling upgrade from 1.4.x to 2.x. Please see the “Rolling Upgrade from 1.x to 2.x” section in ref guide for more details.
The command help that followed all errors, before, is now no longer available. Erroneous command inputs would now just show error-texts followed by the shell command to try for seeing the help message. It looks like: For usage try 'help “create”’. Operators can copy-paste the command to get the help message.
After HBASE-20194, we added 2 parts to master‘s web page. One is Peers that shows all replication peers and some of their configurations, like peer id, cluster key, state, bandwidth, and which namespace or table it will replicate. The other one is replication status of all regionservers, we added a tab to region servers division, then we can check the replication delay of all region servers for any peer. This table shows AgeOfLastShippedOp, SizeOfLogQueue and ReplicationLag for each regionserver and the table is sort by ReplicationLag in descending order. By this way we can easily find the problematic region server. If the replication delay is UNKNOWN, that means this walGroup doesn’t start replicate yet and it may get disabled. ReplicationLag will update once this peer start replicate.
Add prefetch support for async region locator. The default value is 10. Set ‘hbase.client.locate.prefetch.limit’ in hbase-site.xml if you want to use another value for it.
This changes client-side nonce generation to use the same nonce for re-submissions of client RPC DDL operations.
Introduce an InitMetaProcedure to initialize meta table for a new HBase deploy. Marked RecoverMetaProcedure deprecated and remove the usage of it in the current code base. We still need to keep it in place for compatibility. The code in RecoverMetaProcedure has been moved to ServerCrashProcedure, and SCP will always be enabled and we will rely on it to bring meta region online.
For more on the issue addressed by this commit, see the design doc for overview and plan: https://docs.google.com/document/d/1_872oHzrhJq4ck7f6zmp1J--zMhsIFvXSZyX1Mxg5MA/edit#heading=h.xy1z4alsq7uy
HBase now includes a helper script that can be used to run a basic functionality test for a given HBase installation at in dev_support. The test can optionally be given an HBase client artifact to rely on and can optionally be given specific Hadoop client artifacts to use.
For usage information see ./dev-support/hbase_nightly_pseudo-distributed-test.sh --help.
The project nightly tests now make use of this test to check running on top of Hadoop 2, Hadoop 3, and Hadoop 3 with shaded client artifacts.
The HBase convenience binary artifacts now includes a client focused tarball that a) includes more docs and b) does not include scripts or jars only needed for running HBase cluster services.
The new artifact is made as a normal part of the assembly:single maven command.
HBase's built in scripts now rely on the downstream facing shaded artifacts where possible. In particular interest to downstream users, the hbase classpath and hbase mapredcp commands now return the relevant shaded client artifact and only those third paty jars needed to make use of them (e.g. slf4j-api, commons-logging, htrace, etc).
Downstream users should note that by default the hbase classpath command will treat having hadoop on the shell's PATH as an implicit request to include the output of the hadoop classpath command in the returned classpath. This long-existing behavior can be opted out of by setting the environment variable HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP to the value “true”. For example: HBASE_DISABLE_HADOOP_CLASSPATH_LOOKUP="true" bin/hbase classpath.
Downstream users who need to use both HBase and Hadoop APIs should switch to relying on the new hbase-shaded-client-byo-hadoop artifact rather than the existing hbase-shaded-client artifact. The new artifact no longer includes and Hadoop classes.
It should work in combination with either the output of hadoop classpath or the Hadoop provided client-facing shaded artifacts in Hadoop 3+.
The hbase-shaded-mapreduce artifact no longer include its own copy of Hadoop classes. Users who make use of the artifact via YARN should be able to get these classes from YARN's classpath without having to make any changes.
Users of our integration tests on Hadoop 3 can now add all needed dependencies by pointing at jars included in our binary convenience artifact.
Prior to this fix, downstream users on Hadoop 3 would need to get a copy of the Hamcrest v1.3 jar from elsewhere.
Adds two new properties for hbase-site.xml for THRIFT SPNEGO when in HTTP mode: * hbase.thrift.spnego.keytab.file * hbase.thrift.spnego.principal
Adds a negotiation logic between a secure java REST client and server. After this jira the Java REST client will start responding to the Negotiate challenge sent by the server. Adds RESTDemoClient which can be used to verify whether the secure Java REST client works against secure REST server or not.
A second attempt at fixing HBASE-20173. Fixes unfinished keeping of server state inside AM (ONLINE=>SPLITTING=>OFFLINE=>null). Concurrent unassigns look at server state to figure if they should wait on SCP to wake them up or not.
This patch adds an FSUtil.copyFilesParallel() to help copy files in parallel, and it will return all the paths of directories and files traversed. Thus when we copy manifest in ExportSnapshot, we can copy reference files concurrently and use the paths it returns to help setOwner and setPermission. The size of thread pool is determined by the configuration snapshot.export.copy.references.threads, and its default value is the number of runtime available processors.
Before this change we would incorrectly include the size of enqueued store files for bulk replication in the calculation for determining whether or not to rate limit the transfer of WAL edits. Because bulk replication uses a separate and asynchronous mechanism for file transfer this could incorrectly limit the batch sizes for WAL replication if bulk replication in progress, with negative impact on latency and throughput.
PreUpgradeValidator tool with DataBlockEncoding validator was added to verify cluster is upgradable to HBase 2.
HBase is no longer able to maintain compatibility with Apache Hadoop versions that are no longer receiving updates. This release raises the minimum supported version to Hadoop 2.7.1. Downstream users are strongly advised to upgrade to the latest Hadoop 2.7 maintenance release.
Downstream users of earlier HBase versions are similarly advised to upgrade to Hadoop 2.7.1+. When doing so, it is especially important to follow the guidance from the HBase Reference Guide's Hadoop section on replacing the Hadoop artifacts bundled with HBase.
Add multiPut support Set --multiPut=number to enable batchput(meanwhile, --autoflush need be set to false)
Add Connection Count support Added a new parameter connCount to PE. set --connCount=2 means all threads will share 2 connections. oneCon option and connCount option shouldn't be set at the same time.
Add avg RT and avg TPS/QPS statstic for all threads
Delete some redundant code Now RandomWriteTest is inherited from SequentialWrite.
HBase now relies on an internal mechanism to determine when it is running a local hbase cluster meant for external interaction vs an encapsulated test. When created via the HBaseTestingUtility, ports for Master and RegionServer services and UIs will be set to random ports to allow for multiple parallel uses on a single machine. Normally when running a Standalone HBase Deployment (as described in the HBase Reference Guide) the ports will be picked according to the same defaults used in a full cluster set up. If you wish to instead use the random port assignment set hbase.localcluster.assign.random.ports to true.
Added ‘hbase.rest.http.allow.options.method’ configuration property to allow user to decide whether Rest Server HTTP should allow OPTIONS method or not. By default it is enabled in HBase 2.1.0+ versions and in other versions it is disabled. Similarly ‘hbase.thrift.http.allow.options.method’ is added HBase 1.5, 2.1.0 and 3.0.0 versions. It is disabled by default.
This change will enable users to perform append and increment operation with null qualifier via hbase-shell.
When attempting to clone a snapshot but using a namespace that does not exist, the HBase shell will now correctly report the exception as caused by the passed namespace. Previously, the shell would report that the problem was an unknown namespace but it would claim the user provided table name was not found as a namespace. Both before and after this change the shell properly used the passed namespace to attempt to handle the request.
When configured to do thrift-over-http, the HBase Thrift API Server no longer accepts the HTTP methods TRACE nor OPTIONS.
Now in replication we can make sure the order of pushing logs is same as the order of requests from client. Set the serial flag to true for a replication peer to enable this feature.
After HBASE-20159 we allow client to use different ZK quorums by introducing three new properties: hbase.client.zookeeper.quorum and hbase.client.zookeeper.property.clientPort to specify client zookeeper properties (note that the combination of these two properties should be different from the server ZK quorums), and hbase.client.zookeeper.observer.mode to indicate whether the client ZK nodes are in observer mode (false by default)
HConstants.DEFAULT_ZOOKEPER_CLIENT_PORT has been removed in HBase 3.0 and replaced by the correctly spelled DEFAULT_ZOOKEEPER_CLIENT_PORT.
Now when opening a region, we will store the current max sequence id of the region to its max sequence id file instead of the ‘next sequence id’. This could avoid the sequence id bumping when we fail to open a region, and also align to the behavior when we close a region.
The default durability setting for the synchronous WAL is Durability.SYNC_WAL, which triggers HDFS hflush() to flush edits to the datanodes. We also support Durability.FSYNC_WAL, which instead triggers HDFS hsync() to flush _and_ fsync edits. This change introduces the new configuration setting “hbase.wal.hsync”, defaulting to FALSE, that if set to TRUE changes the default durability setting for the synchronous WAL to FSYNC_WAL.
After HBASE-19389 we introduced a RegionServer self-protection mechanism to prevent write handler getting exhausted by high concurrency put with dense columns, mainly through two new properties: hbase.region.store.parallel.put.limit.min.column.count to decide what kind of put (with how many columns within a single column family) to limit (100 by default) and hbase.region.store.parallel.put.limit to limit the concurrency (10 by default). There's another property for advanced user and please check source and javadoc of StoreHotnessProtector for more details.
A new method setSerial has been added to the interface ReplicationPeerConfigBuilder which is marked as IA.Public. This interface is not supposed to be implemented by client code, but if you do, this will be an incompatible change as you need to add this method to your implementation too.
Introduce 5 procedures to do peer modifications: AddPeerProcedure RemovePeerProcedure UpdatePeerConfigProcedure EnablePeerProcedure DisablePeerProcedure
The procedures are all executed with the following stage:
The procedure will hold an exclusive lock on the peer id, so now there is no concurrent modifications on a single peer.
And now it is guaranteed that once the procedure is done, the peer modification has already taken effect on all RSes.
Abstracte a storage layer for replication peer/queue manangement, and refactored the upper layer to remove zk related naming/code/comment.
Add pre/postExecuteProcedures CP hooks to RegionServerObserver, and add permission check for executeProcedures method which requires the caller to be system user or super user.
On rolling upgrade: just do not do any replication peer modifications during the rolling upgrading. There is no pb/layout changes on the peer/queue storage on zk.
These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Change the default so that on creation of new tables, In-Memory Compaction BASIC is NOT enabled.
This change is in branch-2.0 only, not in branch-2.
The HBase shell now behaves as it did prior to the changes that started in HBASE-15965. Namely, some shell commands return values that may be further manipulated within the shell's IRB session.
The command line option --return-values is no longer acted on by the shell since it now always behaves as it did when passed this parameter. Passing the option results in a harmless warning about this change.
Users who wish to maintain the behavior seen in the 1.4.0-1.4.2 releases of the HBase shell should refer to the section irbrc in the reference guide for how to configure their IRB session to avoid echoing expression results to the console.
As of HBase version 2.0, the hbck tool is significantly changed. In general, all Read-Only options are supported and can be be used safely. Most -fix/ -repair options are NOT supported. Please see usage below for details on which options are not supported:
Usage: fsck [opts] {only tables} where [opts] are: -help Display help options (this) -details Display full report of all regions. -timelag <timeInSeconds> Process only regions that have not experienced any metadata updates in the last <timeInSeconds> seconds. -sleepBeforeRerun <timeInSeconds> Sleep this many seconds before checking if the fix worked if run with -fix -summary Print only summary of the tables and status. -metaonly Only check the state of the hbase:meta table. -sidelineDir <hdfs://> HDFS path to backup existing meta. -boundaries Verify that regions boundaries are the same between META and store files. -exclusive Abort if another hbck is exclusive or fixing.
Datafile Repair options: (expert features, use with caution!) -checkCorruptHFiles Check all Hfiles by opening them to make sure they are valid -sidelineCorruptHFiles Quarantine corrupted HFiles. implies -checkCorruptHFiles
Replication options -fixReplication Deletes replication queues for removed peers
Metadata Repair options supported as of version 2.0: (expert features, use with caution!) -fixVersionFile Try to fix missing hbase.version file in hdfs. -fixReferenceFiles Try to offline lingering reference store files -fixHFileLinks Try to offline lingering HFileLinks -noHdfsChecking Don‘t load/check region info from HDFS. Assumes hbase:meta region info is good. Won’t check/fix any HDFS issue, e.g. hole, orphan, or overlap -ignorePreCheckPermission ignore filesystem permission pre-check
NOTE: Following options are NOT supported as of HBase version 2.0+.
UNSUPPORTED Metadata Repair options: (expert features, use with caution!) -fix Try to fix region assignments. This is for backwards compatiblity -fixAssignments Try to fix region assignments. Replaces the old -fix -fixMeta Try to fix meta problems. This assumes HDFS region info is good. -fixHdfsHoles Try to fix region holes in hdfs. -fixHdfsOrphans Try to fix region dirs with no .regioninfo file in hdfs -fixTableOrphans Try to fix table dirs with no .tableinfo file in hdfs (online mode only) -fixHdfsOverlaps Try to fix region overlaps in hdfs. -maxMerge <n> When fixing region overlaps, allow at most <n> regions to merge. (n=5 by default) -sidelineBigOverlaps When fixing region overlaps, allow to sideline big overlaps -maxOverlapsToSideline <n> When fixing region overlaps, allow at most <n> regions to sideline per group. (n=2 by default) -fixSplitParents Try to force offline split parents to be online. -removeParents Try to offline and sideline lingering parents and keep daughter regions. -fixEmptyMetaCells Try to fix hbase:meta entries not referencing any region (empty REGIONINFO_QUALIFIER rows)
UNSUPPORTED Metadata Repair shortcuts -repair Shortcut for -fixAssignments -fixMeta -fixHdfsHoles -fixHdfsOrphans -fixHdfsOverlaps -fixVersionFile -sidelineBigOverlaps -fixReferenceFiles-fixHFileLinks -repairHoles Shortcut for -fixAssignments -fixMeta -fixHdfsHoles
A new RpcThrottlingException deprecates ThrottlingException. The new RpcThrottlingException is a retryable Exception that clients will retry when Rpc throttling quota is exceeded. The deprecated ThrottlingException is a nonretryable Exception.
Standalone webui was broken inadvertently by HBASE-20027.
If HBase is run on top of Apache Hadoop libraries that support the needed APIs it will verify that underlying Filesystem implementations provide the needed durability mechanisms to safely operate. The needed APIs should be present in Hadoop 3 release and Hadoop 2 releases starting in the Hadoop 2.9 series. If the APIs are not available, HBase behaves as it has in previous releases (that is, it moves forward assuming such a check would pass).
Where this check fails, it is unsafe to rely on HBase in a production setting. In the event of process or node failure, the HBase RegionServer process may fail to have access to all the data it previously wrote to its write ahead log, resulting in data loss. In the event of process or node failure, the HBase master process may lose all or part of the write ahead log that it relies on for cluster management operations, leaving the cluster in an inconsistent state that we aren't sure it could recover from.
Notably, the LocalFileSystem implementation provided by Hadoop reports (accurately) via these new APIs that it can not provide the durability HBase needs to operate. As such, the current instructions for single-node HBase operation have been updated both with a) how to bypass this safety check and b) a strong warning about the dire consequences of doing so outside of a dev/test environment.
Throws DoNotRetryIOException when you ask for a reverse scan loading adjacent column families on demand. Previous it threw IllegalStateException
Cleanup usage message and command-line processing (no functional change).
Now if we hit a split parent when locating a region, we will skip to the next row and try again until the region does not contain our row. So there will be no RegionOfflineException for a split parent any more, instead, if the split children have not been onlined yet, i.e, we finally arrive at a region which does not contain our row, an IOException will be thrown.
We no longer include dev or dev test javadocs in our binary bundle. We still build them; they are just not included because they were half the size of the resultant tarball.
Here is our story on javadoc as of this commit:
* apidocs - user facing main api javadocs. currently for a release line, published on website and linked from menu. included in the bin tarball * devapidocs - hbase internal javadocs. currently for a release line, published on the website but not linked from the menu. no longer included in the bin tarball. * testapidocs - user facing test scope api javadocs. currently for a release line, not published. included in the bin tarball. * testdevapidocs - hbase internal test scope javadocs. currently for a release line, not published. no longer included in the bin tarball
Moves us over to yetus releasedocmaker tooling generating CHANGES. CHANGES is not markdown (CHANGES.md) as opposed to CHANGES.txt. We've also added a new RELEASENOTES.md that lists JIRA release notes (courtesy of releasedocmaker).
CHANGES/RELEASENOTES are current as of now. Will need a ‘freshening’ when we cut the RC.
We will use yetus releasedocmaker to make our changes doc from here on out. A CHANGELOG.md will replace our current CHANGES.txt. Adjacent, we'll keep up a RELEASENOTES.md doc courtesy of releasedocmaker.
Over in HBASE-18828 is where we are working through steps for the RM integrating this new tooling.
Changed the default value for replication.source.ratio from 0.1 to 0.5. Which means now by default 50% of the total RegionServers in peer cluster(s) will participate in replication.
The HBase shell command no longer recognizes the option --format. Previously this option only recognized the default value of ‘console’. The default value is now always used.
Disables in-memory compaction as default.
Adds logging of in-memory compaction configuration on creation.
Adds a chapter to the refguide on this new feature.
`hbase regionsplitter` is a new short invocation for `hbase org.apache.hadoop.hbase.util.RegionSplitter`
Upgrade surefire plugin to 2.21.0.
When region servers bind to localhost (mostly in pseudo distributed mode), default ports (16020 & 16030) are used as base ports. This will support up to 9 instances of region servers by default with local-regionservers.sh script. If additional instances are needed, see the reference guide on how to deploy with a different range using the environment variables HBASE_RS_BASE_PORT and HBASE_RS_INFO_BASE_PORT.
When a split is requested on a Region, the RegionServer hosting that Region will now consult the configured SplitPolicy for that table when determining if a split of that Region is allowed. When a split is disallowed (due to the Region not being OPEN or the SplitPolicy denying the request), the operation will *not* be implicitly retried as it has previously done. Users will need to guard against and explicitly retry region split requests which are denied by the system.
Moves commons-cli and commons-collections4 into the HBase thirdparty shaded jar which means that these are no longer generally available for users on the classpath.
Removes Distributed Log Replay feature. Disable the feature before upgrading.
Throw UnknownProtocolException if a client connects and tries to invoke the old getClosestRowOrBefore method. Pre-hbase-1.0.0 or asynchbase do this instead of using its replacement, the reverse Scan.
getClosestRowOrBefore was implemented as a flag on Get. Before this patch though the flag was set, hbase2 were ignoring it. This made it look like a pre-1.0.0 client was ‘working’ but then it'd fail finding the appropriate Region for a client-specified row doing lookups into hbase:meta.
Set version as 2.0.0 on branch-2.0.
When there is concurrent region split, MemStoreFlusher may not find flushable region if the only candidate region left hasn‘t received writes (resulting in 0 data size). After this JIRA, such scenario wouldn’t trigger Precondition assertion (replaced by an if statement to see whether there is any flushable region). If there is no flushable region, a DEBUG log would appear in region server log, saying “Above memory mark but there is no flushable region”.
hbase-thirdparty libs have moved to o.a.h.thirdparty offset. Netty shading system property is no longer necessary.
Adds retry to load of incremental hfiles. Pertinent key is HConstants.HBASE_CLIENT_RETRIES_NUMBER. Default is HConstants.DEFAULT_HBASE_CLIENT_RETRIES_NUMBER.
This issue fixes a runtime dependency issues where JLine is not made available on the classpath which causes the ZooKeeper CLI to appear non-interactive. JLine was being made available unintentionally via the JRuby jar file on the classpath for the HBase shell. While the JRuby jar is not always present, the fix made here was to selectively include the JLine dependency on the zkcli command's classpath.
This behavior is available as a new feature. See HBASE-15968 release note.
This issue is just about adding to the refguide documentation on the HBASE_15968 feature.
Splits out most of ZooKeeper related code into a separate new module: hbase-zookeeper. Also, renames some ZooKeeper related classes to follow a common naming pattern - “ZK” prefix - as compared to many different styles earlier.
The result from server is changed from null to Result.EMPTY_RESULT when Append/Increment operation can't retrieve any data from server,
Committed to master and branch-1
Added ACL check to following Admin functions: enableCatalogJanitor, runCatalogJanitor, cleanerChoreSwitch, runCleanerChore, execProcedure, execProcedureWithReturn, normalize, normalizerSwitch, coprocessorService. When ACL is enabled, only those with ADMIN rights will be able to invoke these operations successfully.
Revert the serial replication feature from all branches. Plan to reimplement it soon and land onto 2.1 release line.
For backward compatibility, AsyncProtobufLogWriter uses “ProtobufLogWriter” as writer class name and SecureAsyncProtobufLogWriter uses “SecureProtobufLogWriter” as writer class name.
Replication between versions verified as basically working. 0.98.25-SNAPSHOT to beta-2 hbase2 and a 1.2-ish version tried.
This change fixes multithreading issues in the implementation of BufferedMutator. BufferedMutator should not be used with 1.4 releases prior to 1.4.2.
Add (latest) version elements missing from reporting plugins in top-level pom.
hadoop3 minidfscluster removes all shutdown handlers when the cluster goes down which made this test that does FS-stuff fail (Fix was to break up the test so each test method ran with an unadulterated FS).
Ups the overall test timeout from 10 minutes to 13minutes. 15minutes is the surefire timeout.
Add checkClosed to core Client methods. Avoid unnecessary retry.
Completes keep-alive logic and then enables it; ProcedureExecutor Workers will spin up more threads when need settling back to the core count after the burst in demand has passed. Default keep-alive is one minute. Default core-count is CPUs/4 or 16, which ever is greater. Maximum is an arbitrary core-count * 10 (a limit that should never be hit and if it is, there is something else very wrong).
ColumnValueFilter provides a way to fetch matched cells only by providing specified column, value and a comparator, which is different from SingleValueFilter, fetching an entire row as soon as a matched cell found.
A region is flushed if its memory component exceeds the region flush threshold. A flush policy decides which stores to flush by comparing the size of the store to a column-family-flush threshold. If the overall size of all memstores in the machine exceeds the bounds defined by the administrator (denoted global pressure) a region is selected and flushed. HBASE-18294 changes flush decisions to be based on heap-occupancy and not data (key-value) size, consistently across levels. This rolls back some of the changes by HBASE-16747. Specifically, (1) RSs, Regions and stores track their overall on-heap and off-heap occupancy, (2) A region is flushed when its on-heap+off-heap size exceeds the region flush threshold specified in hbase.hregion.memstore.flush.size, (3) The store to be flushed is chosen based on its on-heap+off-heap size (4) At the RS level, a flush is triggered when the overall on-heap exceeds the on-heap limit, or when the overall off-heap size exceeds the off-heap limit (low/high water marks).
Note that when the region flush size is set to XXmb a region flush may be triggered even before writing keys and values of size XX because the total heap occupancy of the region which includes additional metadata exceeded the threshold.
hbase-2.x sets KeyValue Comparators into the tail of hfiles rather than CellComparator, what it uses internally, just so hbase-1.x can continue to read hbase-2.x written hfiles.
In subtask, fixed doc and annotations to be more explicit that test timings are for the whole Test Fixture/Test Class/Test Suite NOT the test method only as we'd measuring up to this (tother subtasks untethered Categorization and test timeout such that all categories now have a ten minute timeout -- no test can run longer than ten minutes or it gets killed/timedout).
By default, we mirror table state to zookeeper so hbase-1.x clients will work against an hbase-2 cluster (With this patch, hbase-1.x clients can do most Admin functions including table create; hbase-1.x clients can do all Table/DML against hbase-2 cluster).
Flag to disable mirroring is hbase.mirror.table.state.to.zookeeper; set it to false in Configuration.
Related, Master on startup will look to see if there are table state znodes left over by an hbase-1 instance. If any found, it will migrate the table state to hbase-2 setting the state into the hbase:meta table where table state is now kept. We will do this check on every Master start. Notion is that this will be overall beneficial with low impediment. To disable the migration check, set hbase.migrate.table.state.from.zookeeper to false.
This fix makes the following changes to how client handle the both of action result and region exception.
HBaseTestingUtility now assumes that all clusters will use local storage until a MiniDFSCluster is started or assigned.
Tool allows you to compact a cluster with given concurrency of regionservers compacting at a given time. If tool completes successfully everything requested for compaction will be compacted, regardless of region moves, splits and merges.
(I thought this change innocuous but I made work for a co-worker when I upped interval between log cleaner runs -- meant a smoke test failed because we were slow doing an expected cleanup).
Edit of log lines removing redundancy. Shorten thread names shown in log. Made some log TRACE instead of DEBUG. Capitalizations.
Upped log cleaner interval from every minute to every ten minutes. hbase.master.cleaner.interval
Lowered default count of threads started by Procedure Executor from count of CPUs to 1/4 of count of CPUs.
Pass to yetus a dockermemlimit of 20G and a proclimit of 10000. Defaults are 4G and 1G respectively.
Remove useless ‘writeToWAL’ flag of Region#checkAndRowMutate & related class
Changed a few tests so they are medium sized rather than small size.
Also, upped the time we wait on small tests to 60seconds from 30seconds. Small tests are tests that run in 15seconds or less. What we changed was the timeout watcher. It is now more lax, more tolerant of dodgy infrastructure that might be running tests slowly.
Moved our internal yetus reference from 0.6.0 to 0.7.0. Concurrently, I changed hadoopqa to run with 0.7.0 (by editing the config in jenkins).
Along with @category -- small, medium, large -- all hbase tests must now carry a ClassRule as follows:
HBaseClassTestRule.forClass(TestInterfaceAudienceAnnotations.class);
where the class changes by test.
Currently the classrule enforces timeout for the whole test suite -- i.e. if a SmallTest Category then all the tests in the TestSuite must complete inside 60seconds, the timeout we set on SmallTest Category test suite -- but is meant to be a repository for general, runtime, hbase test facility.
Introduces a new option to the HBase shell: -r, --return-values. When the shell is in “interactive” mode (default), the return value of shell commands are not returned to the user as they dirty the console output. For those who desire this functionality, the “--return-values” option restores the old functionality of the commands passing their return value to the user.
HRegion.openReadOnlyFileSystemHRegion() provides the ability to open HRegion from a read-only hdfs snapshot. Because hdfs snapshots are read-only, no cleanup happens when using this API.
This change fixes an issue where users could have unintentionally configured the HBase Thrift1 server to run without wire-encryption, when they believed they had configured the Thrift1 server to do so.
Disables TestRegionsOnMasterOptions because Regions on Master does not work reliably; see HBASE-19831.
Modified HRegion.mutateRow() APIs to use batchMutate() instead of processRowsWithLocks() with MultiRowMutationProcessor. MultiRowMutationProcessor is removed to have single write path that uses batchMutate().
When there are many mutations against the same row in a batch, as each mutation will acquire a shared row lock, it will exceed the maximum shared lock count the java ReadWritelock supports (64k). Along with other optimization, the batch is divided into multiple possible minibatches. A new config is added to limit the maximum number of mutations in the minibatch.
<property> <name>hbase.regionserver.minibatch.size</name> <value>20000</value> </property> The default value is 20000.
Thrift IDLs are now shipped, bundled up in the respective hbase-*thrift.jars (look for files ending in .thrift).
Allows for users to bulk load entire tables from hdfs by specifying the parameter -loadTable. This allows you to pass in a table level directory and have all regions column families bulk loaded, if you do not specify the -loadTable parameter LoadIncrementalHFiles will work as before. Note: you must have a pre-created table to run with -loadTable it will not create one for you.
Client-side ZooKeeper metrics which were added to 2.0.0 alpha/beta releases cause issues when launching MapReduce jobs via {{yarn jar}} on the command line. This stems from ClassLoader separation issues that YARN implements. It was chosen that the easiest solution was to remove these ZooKeeper metrics entirely.
To reduce the confusing behavior, now when you call updatePeerConfig with empty ClusterKey or ReplicationEndpointImpl, but the value of field of the to-be-updated ReplicationPeerConfig is not null, we will throw exception instead of ignoring them.
This JIRA aims at refactoring AccessController, using ACL as core library in CPs.
After HBASE-19358 we introduced a new property hbase.split.writer.creation.bounded to limit the opening writers for each WALSplitter. If set to true, we won‘t open any writer for recovered.edits until the entries accumulated in memory reaching hbase.regionserver.hlog.splitlog.buffersize (which defaults at 128M) and will write and close the file in one go instead of keeping the writer open. It’s false by default and we recommend to set it to true if your cluster has a high region load (like more than 300 regions per RS), especially when you observed obvious NN/HDFS slow down during hbase (single RS or cluster) failover.
HBase had copied from guava the file LmiitedInputStream. This commit removes the copied file in favor of (our internal, shaded) guava‘s ByteStreams.limit. Guava 14.0’s LIS noted: “Use ByteStreams.limit(java.io.InputStream, long) instead. This class is scheduled to be removed in Guava release 15.0.”
This change reverts an unintentional requirement for global ADMIN permission to obtain cluster status from the active HMaster.
The BufferedMutator now supports two settings that are used to ensure records do not stay too long in the buffer of a BufferedMutator. For periodically flushing the BufferedMutator there is now a “Timeout”: “How old may the oldest record in the buffer be before we force a flush” and a “TimerTick”: How often do we check if the timeout has been exceeded. Using these settings you can make the BufferedMutator automatically flush the write buffer if after the specified number of milliseconds no flush has occurred.
This is mainly useful in streaming scenarios (i.e. writing data into HBase using Apache Flink/Beam/Storm) where it is common (especially in a test/development situation) to see small unpredictable bursts of data that need to be written into HBase. When using the BufferedMutator till now the effect was that records would remain in the write buffer until the buffer was full or an explicit flush was triggered. In practice this would mean that the ‘last few records’ of a burst would remain in the write buffer until the next burst arrives filling the buffer to capacity and thus triggering a flush.
Disable filtering of User API based off yetus annotation done in doclet. See parent issue for build failure currently being worked on but not done in time for a beta-1.
When MSLAB is in use (that is the default config) , we will always use the CellChunkMap indexing variant for in memory flushed Immutable segments. When MSLAB is turned off, we will use CellAraryMap. These can not be changed with any configs. The in memory flush threshold been made to be default to 10% of region flush size. This can be turned using ‘hbase.memstore.inmemoryflush.threshold.factor’.
ByteBufferCell → ByteBufferExtendedCell MapReduceCell → MapReduceExtendedCell ByteBufferChunkCell → ByteBufferChunkKeyValue NoTagByteBufferChunkCell → NoTagByteBufferChunkKeyValue KeyOnlyByteBufferCell → KeyOnlyByteBufferExtendedCell TagRewriteByteBufferCell → TagRewriteByteBufferExtendedCell ValueAndTagRewriteByteBufferCell → ValueAndTagRewriteByteBufferExtendedCell EmptyByteBufferCell → EmptyByteBufferExtendedCell FirstOnRowByteBufferCell → FirstOnRowByteBufferExtendedCell LastOnRowByteBufferCell → LastOnRowByteBufferExtendedCell FirstOnRowColByteBufferCell → FirstOnRowColByteBufferExtendedCell FirstOnRowColTSByteBufferCell → FirstOnRowColTSByteBufferExtendedCell LastOnRowColByteBufferCell → LastOnRowColByteBufferCell OffheapDecodedCell → OffheapDecodedExtendedCell
Add a ReplicationPeerConfigBuilder to create ReplicationPeerConfig and make ReplicationPeerConfig immutable. Meanwhile, deprecated set* methods in ReplicationPeerConfig.
We now have slf4j as our front-end. Be careful adding logging from here on out; make sure it slf4j.
From here on out, as us devs go, we need to convert log messages from being ‘guarded’ -- i.e. surrounded by if (LOG.isDebugEnabled...) -- to instead being parameterized log messages. e.g. the latter rather than the former in the below:
logger.debug(“The new entry is “+entry+”.”); logger.debug(“The new entry is {}.”, entry);
See [1] for background on perf benefits.
Note, FATAL log level is not present in slf4j. It is noted as a Marker but won't show in logs as a LEVEL.
Removed unused hbase.fs.tmp.dir from hbase-default.xml.
Upped hbase.master.fileSplitTimeout from 30s to 10minutes (suggested by production experience)
Added note that handler-count should be ~CPU count.
hbase.regionserver.logroll.multiplier has been changed from 0.95 to 0.5 AND the default block size has been doubled.
A few of the core configs are now dumped to the log on startup.
Add two new field: EXCLUDE_NAMESPACE and EXCLUDE_TABLECFS to replication peer config.
If replicate_all flag is true, it means all user tables will be replicated to peer cluster. Then allow config exclude namespaces or exclude table-cfs which can't be replicated to peer cluster.
If replicate_all flag is false, it means all user tables can't be replicated to peer cluster. Then allow to config namespaces or table-cfs which will be replicated to peer cluster.
Adds means of adding very basic filter on the sink side of replication. We already have a means of installing filter source-side, which is better place to filter edits before they are shipped over the network, but this facility is needed by hbase-indexer.
Set hbase.replication.sink.walentrysinkfilter with a no-param Constructor implementation. See test in patch for example.
Adds method Cell#getType which returns enum describing Cell Type.
Deprecates the following Cell methods:
getTypeByte getSequenceId getTagsArray getTagsOffset getTagsLength
CPs trying to build cells should use RawCellBuilderFactory that supports building cells with tags.
Implement a FanOutOneBlockAsyncDFSOutput for writing WAL only, the WAL provider which uses this class is AsyncFSWALProvider.
It is based on netty, and will write to 3 DNs at the same time concurrently(fan-out) so generally it will lead to a lower latency. And it is also fail-fast, the stream will become unwritable immediately after there are any read/write errors, no pipeline recovery. You need to call recoverLease to force close the output for this case. And it only supports to write a file with a single block. For WAL this is a good behavior as we can always open a new file when the old one is broken. The performance analysis in HBASE-16890 shows that it has a better performance.
Behavior changes:
Now the default WALProvider is AsyncFSWALProvider, i.e. ‘asyncfs’. If you want to change back to use FSHLog, please add this in hbase-site.xml {code} <property> <name>hbase.wal.provider</name> <value>filesystem</value> </property> {code} If you want to use FSHLog with multiwal, please add this in hbase-site.xml {code} <property> <name>hbase.wal.regiongrouping.delegate.provider</name> <value>filesystem</value> </property> {code}
This patch also sets hbase.wal.async.use-shared-event-loop to false so WAL has its own netty event group.
Deprecates Put#addImmutable as of release 2.0.0, this will be removed in HBase 3.0.0. Use {@link #add(Cell)} and {@link org.apache.hadoop.hbase.CellBuilder} instead
In Table interface deprecate checkAndPut, checkAndDelete and checkAndMutate methods. Similarly to AsyncTable a new method was added to replace the deprecated ones: CheckAndMutateBuilder checkAndMutate(byte[] row, byte[] family) with CheckAndMutateBuilder interface which can be used to construct the checkAnd*() operations.
Made WALKey an Interface and added a WALKeyImpl implementation. WALKey comes through to Coprocessors. WALKey is read-only.
Refactor of Coprocessor API for hbase2. Purged methods that exposed too much of our internals. Other hooks were recast so they no longer took or returned internal classes; instead we pass Interfaces or read-only versions of implementations.
Here is some overview doc on changes in hbase2 for Coprocessors including detail on why the change was made: https://github.com/apache/hbase/blob/branch-2.0/dev-support/design-docs/Coprocessor_Design_Improvements-Use_composition_instead_of_inheritance-HBASE-17732.adoc
Provided a way for the CP users to create a short circuitable connection with custom configs.
createConnection(Configuration) is added to MasterCoprocessorEnvironment, RegionServerCoprocessorEnvironment and RegionCoprocessorEnvironment.
The getConnection() method already available in these Env interfaces returns the cluster connection used by the server (which the server also uses) where as this new method will create a new connection on request. The difference from connection created using ConnectionFactory APIs is that this connection can short circuit the calls to same server avoiding the RPC paths. The connection will NOT be cached/maintained by server. That should be done the CPs.
Be careful creating Connections out of a Coprocessor. See the javadoc on these createConnection and getConnection.
Removed cacheDataInL1 option for HCD BucketCache is no longer the L2 for LRU on heap cache. When BC is used, data blocks will be strictly on BC only where as index/bloom blocks are on LRU L1 cache. Config ‘hbase.bucketcache.combinedcache.enabled’ is removed. There is no way set combined mode = false. Means make BC as victim handler for LRU cache. This will be one more noticeable change when one uses BucketCache in File mode. Then the system table's data block(Including the META table) will be cached in Bucket Cache files only. Plain scan from META files alone test reveal that the throughput of file mode BC is almost half only. But for META entries we have RegionLocation cache at client side connections. So this would not be a big concern in a real cluster usage. Will check more on this and probably fix even when we do tiered BucketCache.
All the cells which are used in server side are of ExtendedCell now.
CoprocessorEnvironment#getConfiguration returns a READ-ONLY Configuration. Attempts at altering the returned Configuration -- whether setting or adding resources -- will result in an IllegalStateException warning of the Read-only condition of the returned Configuration.
There is a new HBaseZKTestingUtility which can only start a mini zookeeper cluster. And we will publish sources for test-jar for all modules.
NettyRpcServer is now our default RPC server replacing SimpleRpcServer.
Moves #has and #setTimestamp back up to Mutation from the subclass Put so available to other Mutation implementations.
When a coprocessor sets ‘bypass’, we will skip calling subsequent Coprocessors that may be stacked-up on the method invocation; e.g. if a prePut has three coprocessors hooked up, if the first coprocessor decides to set ‘bypass’, we will not call the two subsequent coprocessors (this is similar to the ‘complete’ functionality that was in hbase1, removed in hbase2).
Enable TestAcidGuaranteesWithEagerPolicy and TestAcidGuaranteesWithAdaptivePolicy
Add two new shell cmd. move_namespaces_rsgroup is used to reassign tables of specified namespaces from one RegionServer group to another. move_servers_namespaces_rsgroup is used to reassign regionServers and tables of specified namespaces from one group to another.
Per-RegionServer table latency histograms have been returned to HBase (after being removed due to impacting performance). These metrics are exposed via a new JMX bean “TableLatencies” with the typical naming conventions: namespace, table, and histogram component.
The default value of hbase.client.retries.number was 35. It is now 10. And for server side, the default hbase.client.serverside.retries.multiplier was 10. So the server side retries number was 35 * 10 = 350. It is now 3.
In this task, we make it possible to run multiple mappers per region in the table snapshot. The following code is primary table snapshot mapper initializatio:
TableMapReduceUtil.initTableSnapshotMapperJob( snapshotName, // The name of the snapshot (of a table) to read from scan, // Scan instance to control CF and attribute selection mapper, // mapper outputKeyClass, // mapper output key outputValueClass, // mapper output value job, // The current job to adjust true, // upload HBase jars and jars for any of the configured job classes via the distributed cache (tmpjars) restoreDir, // a temporary directory to copy the snapshot files into );
The job only run one map task per region in the table snapshot. With this feature, client can specify the desired num of mappers when init table snapshot mapper job：
TableMapReduceUtil.initTableSnapshotMapperJob( snapshotName, // The name of the snapshot (of a table) to read from scan, // Scan instance to control CF and attribute selection mapper, // mapper outputKeyClass, // mapper output key outputValueClass, // mapper output value job, // The current job to adjust true, // upload HBase jars and jars for any of the configured job classes via the distributed cache (tmpjars) restoreDir, // a temporary directory to copy the snapshot files into splitAlgorithm, // splitAlgo algorithm to split, current split algorithms support RegionSplitter.UniformSplit() and RegionSplitter.HexStringSplit() n // how many input splits to generate per one region );
Fixes an issue with loading customer coprocessor endpoint implementations inside of the HBase Master which breaks Apache Ranger.
This JIRA aims at exposing Tags for Coprocessor usage. Tag interface is now exposed to Coprocessors and CPs can make use of this interface to create their own Tags. RawCell is a new interface that is a subtype of Cell and that is exposed to CPs. RawCell has the following APIs
List<Tag> getTags() Optional<Tag> getTag(byte type) byte[] cloneTags()
The above APIs helps to read tags from the Cell.
CellUtil#createCell(Cell cell, List<Tag> tags) CellUtil#createCell(Cell cell, byte[] tags) CellUtil#createCell(Cell cell, byte[] value, byte[] tags) are deprecated. If CPs want to create a cell with Tags they can use the RegionCoprocessorEnvironment#getCellBuilder() that returns an ExtendedCellBuilder. Using ExtendedCellBuilder the CP can create Cells with Tags. Other helper methods to work on Tags are available as static APIs in Tag interface.
separate the TestAcidGuarantees by the policy:
TestAcidGuaranteesWithEagerPolicy and TestAcidGuaranteesWithAdaptivePolicy are disabled by default as the eager policy may cause the negative size of memstore.
Add a replicate_all flag to replication peer config. The default value is true, which means all user tables (REPLICATION_SCOPE != 0 ) will be replicated to peer cluster.
How to config a peer from replicate all to only replicate special namespace/tablecfs? Step1. Add a new peer with no namespace/tablecfs config, the replicate_all flag will be true automatically. Step2. User want only replicate some namespaces or tables, so set replicate_all flag to false first. Step3. Add special namespaces or table-cfs config to the replication peer.
How to config a peer from replicate special namespace/tablecfs to replicate all? Step1. Add a new peer with special namespace/tablecfs config, the replicate_all flag will be false automatically. Step2. User want replicate all user tables, so remove the special namespace/tablecfs config first. Step3. Set replicate_all flag to true.
How to config replicate nothing? Set replicate_all flag to false and no namespace/tablecfs config, then all tables cannot be replicated to peer cluster.
Remove the ability to ‘bypass’ preFlush and preCompact by returning a null Scanner. Bypass is disallowed on these methods in hbase2.
ConnectionImplementation now uses asynchronous connections to zookeeper via ZKAsyncRegistry to get cluster id, master address, meta region location, etc. Since ZKAsyncRegistry uses curator framework, this change purges a lot of zookeeper dependencies in hbase-client. Now hbase-client only depends on only ZKAsyncRegistry, ZNodePaths and the newly introduced ZKMetadata.
Introduce a AcidGuaranteesTestTool and expose as tool instead of TestAcidGuarantees. Now TestAcidGuarantees is just a UT.
Add a boolean parameter which means the new replication peer‘s state is enabled or disabled for Admin/AsyncAdmin’s addReplicationPeer method. Meanwhile, you can use shell cmd to add a enabled/disabled replication peer. The STATE parameter is optional and the default state is enabled.
hbase> add_peer ‘1’, CLUSTER_KEY => “server1.cie.com:2181:/hbase”, STATE => “ENABLED” hbase> add_peer ‘1’, CLUSTER_KEY => “server1.cie.com:2181:/hbase”, STATE => “DISABLED”
This issue removes the ‘complete’ facility that was in ObserverContext. It is no longer possible for a Coprocessor to cut the chain-of-invocation and insist its response prevails.
Deprecated 4 methods for Admin interface. Deprecated compactRegionServer(ServerName, boolean). Use compactRegionServer(ServerName) and majorCompactcompactRegionServer(ServerName) instead. Deprecated getRegionLoad(ServerName) method. Use getRegionLoads(ServerName) instead. Deprecated getRegionLoad(ServerName, TableName) method. Use getRegionLoads(ServerName, TableName) instead. Deprecated getQuotaRetriever(QuotaFilter) instead. Use getQuota(QuotaFilter) instead.
Add 7 methods for Admin interface. ServerName getMaster(); Collection<ServerName> getBackupMasters(); Collection<ServerName> getRegionServers(); boolean splitSwitch(boolean enabled, boolean synchronous); boolean mergeSwitch(boolean enabled, boolean synchronous); boolean isSplitEnabled(); boolean isMergeEnabled();
Two write paths Region.batchMutate() and Region.mutateRows() are unified and inconsistencies are resolved.
RowProcessor and Region#processRowsWithLocks() methods that take RowProcessor as an argument are deprecated. Use Coprocessors if you want to customize handling.
Merge the RawAsyncTable and AsyncTable interfaces. Use generic to reflection the difference between the observer style scan API. For the implementation which does not have a user specified thread pool, the observer is AdvancedScanResultConsumer. For the implementation which needs a user specified thread pool, the observer is ScanResultConsumer.
Change the import order rule that now we should put the shaded import at bottom. Ignore the VisibilityModifier warnings for test code.
Removing the on heap Bucket cache feature. The config “hbase.bucketcache.ioengine” no longer support the ‘heap’ value. Its supported values now are ‘offheap’, ‘file:<path>’, ‘files:<path>’ and ‘mmap:<path>’
This change introduces compile time support for running the error-prone suite of static analyses. Enable with -PerrorProne on the Maven command line. Requires JDK 8 or higher. (Don't enable if building with JDK 7.)
(Incomplete)
= Incompatbiles
== Coprocessor Incompatibilities
Split/Merge have moved to the Master; it runs them now. Means hooks around Split/Merge are now noops. To intercept Split/Merge phases, CPs need to intercept on MasterObserver.
Folks can now test out tests on an arbitrary release branch. Head over to builds.a.o job “HBase-adhoc-run-tests”, then pick “Build with parameters”. Tests are specified as just names e.g. TestLogRollingNoCluster. can also be a glob. e.g. TestHFile*
Changed retries from 3 to 30 for zk initial connect for registry.
With the change in Coprocessor APIs, the hbase-examples module has been updated to provide additional examples that show how to write Coprocessors against the new API.
HRegion.batchMutate()/ doMiniBatchMutate() is refactored with aim to unify batchMutate() and mutateRows() code paths later. batchMutate() currently handles 2 types of batches: MutationBatchOperations and ReplayBatchOperations. Common base class BatchOperations is augmented with common methods which are overridden in derived classes as needed. doMiniBatchMutate() is implemented using common methods in base class BatchOperations.
If BigDecimal is stored as value, and you need to add a matched comparator to the value filter when scanning, a BigDecimalComparator can be used.
A new public API method was added to CellUtil “isPut(Cell)” for clients to use to determine if the Cell is for a Put operation.
Additionally, other CellUtil API calls which expose Cell-implementation were marked as deprecated and will be removed in a future version.
CellComparator is now InterfaceAudience.Public
Add 4 methods in RegionObserver: preMemStoreCompaction preMemStoreCompactionCompactScannerOpen preMemStoreCompactionCompact postMemStoreCompaction preMemStoreCompaction and postMemStoreCompaction will always be called for all in memory compactions. Under eager mode, preMemStoreCompactionCompactScannerOpen will be called before opening store scanner to allow you changing the max versions and TTL, and preMemStoreCompactionCompact will be called after the creation to let you do wrapping.
The make_rc.sh script can run an hbase2 build now generating tarballs and pushing up to maven repository. TODO: Sign and checksum, check tarball, push to apache dist.....
Purged the hbase-prefix-tree module and all references from the code base.
prefix-tree data block encoding was a super cool experimental feature that saw some usage initially but has since languished. If interested in carrying this sweet facility forward, write the dev list and we'll restore this module.
Removed the hbase-native-client module from branch-2 (it is still in Master). It is not complete. Look for a finished C++ client in the near future. Will restore native client to branch-2 at that point.
When regionserver placement groups (RSGroups) is active, as servers join the cluster the Master will attempt to reassign regions in FAILED_OPEN state.
Removes blanket bypass mechanism (Observer#bypass). Instead, a curated subset of methods are bypassable.
Changes Coprocessor ObserverContext 'bypass' semantic. We flip the default so bypass is NOT supported on Observer invocations; only a couple of preXXX methods in RegionObserver allow it: e.g. preGet and prePut but not preFlush, etc. Everywhere else, we throw a Exception if a Coprocessor Observer tries to invoke bypass. Master Observers can no longer stop or change move, split, assign, create table, etc. preBatchMutate can no longer be bypassed (bypass the finer-grained prePut, preDelete, etc. instead) Ditto on complete, the mechanism that allowed a Coprocessor rule that all subsequent Coprocessors are skipped in an invocation chain; now, complete is only available to bypassable methods (and Coprocessors will get an exception if they try to 'complete' when it is not allowed). See javadoc for whether a Coprocessor Observer method supports 'bypass'. If no mention, 'bypass' is NOT supported.
The below methods have been marked deprecated in hbase2. We would have liked to have removed them because they use IA.Private parameters but they are in use by CoreCoprocessors or are critical to downstreamers and we have no alternatives to provide currently.
@Deprecated public boolean prePrepareTimeStampForDeleteVersion(final Mutation mutation, final Cell kv, final byte[] byteNow, final Get get) throws IOException {
@Deprecated public boolean preWALRestore(final RegionInfo info, final WALKey logKey, final WALEdit logEdit) throws IOException {
@Deprecated public void postWALRestore(final RegionInfo info, final WALKey logKey, final WALEdit logEdit) throws IOException {
@Deprecated public DeleteTracker postInstantiateDeleteTracker(DeleteTracker result) throws IOException
Metrics are updated now even if the Coprocessor does a bypass; e.g. The put count is updated even if a Coprocessor bypasses the core put operation (We do it this way so no need for Coprocessors to have access to our core metrics system).
Add back the three methods without a return value: preFlushScannerOpen preCompactScannerOpen preStoreScannerOpen
Introduce a ScanOptions interface to let CP users change the max versions and TTL of a ScanInfo. It will be passed as a parameter in the three methods above.
Inntroduce a new example WriteHeavyIncrementObserver which convert increment to put and do aggregating when get. It uses the above three methods.
Made defaults for Server#isStopping and Server#getFileSystem. Should have done this when I added them (lesson learned, was actually mentioned in a review).
RegionObserver#preScannerOpen signature changed RegionScanner preScannerOpen( ObserverContext<RegionCoprocessorEnvironment> c, Scan scan, RegionScanner s) -> void preScannerOpen( ObserverContext<RegionCoprocessorEnvironment> c, Scan scan)
The pre hook can no longer return a RegionScanner instance.
Split CellUtil into public CellUtil and PrivateCellUtil for Internal use only.
Provided an API in Region (Exposed to CPs) boolean waitForFlushes(long timeout) This call will make the current thread to be waiting for all flushes in this region to be finished. (Upto the time out time being specified). The boolean return value specify whether the flushes are really over or the time out being elapsed. Return false when timeout elapsed but flushes are not over or true when flushes are over
Add a FlushLifeCycleTracker which is similiar to CompactionLifeCycleTracker for tracking flush. Add a requestFlush method in Region interface to let CP users request flush on a region. The operation is asynchronous, you need to use the FlushLifeCycleTracker to track the flush. The difference with CompactionLifeCycleTracker is that, flush is per region so we do not use Store as a parameter of the methods. And also, notExecuted means the whole flush has not been executed, and afterExecution means the whole flush has been finished, so we do not have a separated completed method. A flush will be ended either by notExecuted or afterExecution.
Purged InterfaceAudience.Private parameters from methods in MasterObserver.
preAbortProcedure no longer takes a ProcedureExecutor.
postGetProcedures no longer takes a list of Procedures.
postGetLocks no longer takes a list of locks.
preRequestLock and postRequestLock no longer take lock type.
preLockHeartbeat and postLockHeartbeat no longer takes a lock procedure.
The implication is that that the Coprocessors that depended on these params have had to coarsen so for example, the AccessController can not do access per Procedure or Lock but rather, makes a judgement on the general access (You'll need to be ADMIN to see list of procedures and locks).
Added a new config 'hbase.systemtables.compacting.memstore.type" for the system tables. By default all the system tables will have ‘NONE’ as the type and so it will be using the default memstore by default. {code} <property> <name>hbase.systemtables.compacting.memstore.type</name> <value>NONE</value> </property> {code}
Deprecate the following methods in Table:
Add the following methods to Table:
Add missing deprecation tag for long getRpcTimeout(TimeUnit unit) in AsyncTableBase
In this task, we fixed all existing bugs in FilterList, and did the code refactor which ensured interface compatibility .
The primary bug fixes are :
About code refactor, we divided the FilterList into two separated sub-classes: FilterListWithOR and FilterListWithAND, The FilterListWithOR has been optimised to choose the next minimal step to seek cell rather than SKIP cell one by one, and the FilterListWithAND has been optimised to choose the next maximal key to seek among sub-filters in filter list. All in all, The code in FilterList is clean and easier to follow now.
Note that ReturnCode NEXT_ROW has been redefined as skipping to next row in current family, not to next row in all family. it’s more reasonable, because ReturnCode is a concept in store level, not in region level.
Another bug that needs attention is: filterAllRemaining() in FilterList with MUST_PASS_ONE will now return false if the filter list is empty whereas earlier it used to return true for Operator.MUST_PASS_ONE. it's more reasonable now.
Adds getOnlineRegions to the RegionCoprocessorEnvironment (Context) and ditto to RegionServerCoprocessorEnvironment. Allows Coprocessor get list of Regions online on the currently hosting RegionServer.
Re-enabled ‘hbase.master.loadbalance.bytable’, default ‘false’. Draining servers are removed from consideration by blancer.balanceCluster() call.
HBase now relies on Kerby version 1.0.1 for its test environment. No downstream facing change is expected.
Patch to print summary of call queues by size and count. This is displayed on the debug dump page of region server UI
Makes it so hbase-indexer/lily can move off dependence on internal APIs and instead move to public APIs.
Adds being able to disable near-all HRegionServer services. This along with an existing plugin mechanism which allows configuring the RegionServer to host an alternate Connection implementation, makes it so we can put up a cluster of hollowed-out HRegionServers purposed to pose as a Replication Sink for a source HBase Cluster (Users do not need to figure our RPC, our PB encodings, build a distributed service, etc.). In the alternate supplied Connection implementation, hbase-indexer would install its own code to catch the Replication.
Below and attached are sample hbase-server.xml files and alternate Connection implementations. To start up an HRegionServer as a sink, first make sure there is a ZooKeeper ensemble we can talk to. If none, just start one: {code} ./bin/hbase-daemon.sh start zookeeper {code}
To start up a single RegionServer, put in place the below sample hbase-site.xml and a derviative of the below IndexerConnection on the CLASSPATH, and then start the RegionServer: {code} ./bin/hbase-daemon.sh start org.apache.hadoop.hbase.regionserver.HRegionServer {code} Stdout and Stderr will go into files under configured logs directory. Browse to localhost:16030 to find webui (unless disabled).
DETAILS
This patch adds configuration to disable RegionServer internal Services, Managers, Caches, etc., starting up.
By default a RegionServer starts up an Admin and Client Service. To disable either or both, use the below booleans: {code} hbase.regionserver.admin.service hbase.regionserver.client.service {code}
Both default true.
To make a HRegionServer startup and stay up without expecting to communicate with a master, set the below boolean to false:
{code} hbase.masterless {code] Default is false.
h3. Sample hbase-site.xml that disables internal HRegionServer Services Below is an example hbase-site.xml that turns off most Services and that then installs an alternate Connection implementation, one that is nulled out in all regards except in being able to return a “Table” that can catch a Replication Stream in its {code}batch(List<? extends Row> actions, Object[] results){code} method. i.e. what the hbase-indexer wants. I also add the example alternate Connection implementation below (both of these files are also attached to this issue). Expects there to be an up and running zookeeper ensemble.
{code} <configuration> <!-- This file is an example for hbase-indexer. It shuts down facility in the regionserver and interjects a special Connection implementation which is how hbase-indexer will receive the replication stream from source hbase cluster. See the class referenced in the config.
Most of the config in here is booleans set to off and setting values to zero so services doon't start. Some of the flags are new via this patch.
--> <!--Need this for the RegionServer to come up standalone--> <property> <name>hbase.cluster.distributed</name> <value>true</value> </property>
<!--This is what you implement, a Connection that returns a Table that overrides the batch call. It is at this point you do your indexer inserts. --> <property> <name>hbase.client.connection.impl</name> <value>org.apache.hadoop.hbase.client.IndexerConnection</value> <description>A customs connection implementation just so we can interject our own Table class, one that has an override for the batch call which receives the replication stream edits; i.e. it is called by the replication sink #replicateEntries method.</description> </property>
<!--Set hbase.regionserver.info.port to -1 for no webui-->
<!--Below are configs to shut down unused services in hregionserver--> <property> <name>hbase.regionserver.admin.service</name> <value>false</value> <description>Do NOT stand up an Admin Service Interface on RPC</description> </property> <property> <name>hbase.regionserver.client.service</name> <value>false</value> <description>Do NOT stand up a client-facing Service on RPC</description> </property> <property> <name>hbase.wal.provider</name> <value>org.apache.hadoop.hbase.wal.DisabledWALProvider</value> <description>Set WAL service to be the null WAL</description> </property> <property> <name>hbase.regionserver.workers</name> <value>false</value> <description>Turn off all background workers, log splitters, executors, etc.</description> </property> <property> <name>hfile.block.cache.size</name> <value>0.0001</value> <description>Turn off block cache completely</description> </property> <property> <name>hbase.mob.file.cache.size</name> <value>0</value> <description>Disable MOB cache.</description> </property> <property> <name>hbase.masterless</name> <value>true</value> <description>Do not expect Master in cluster.</description> </property> <property> <name>hbase.regionserver.metahandler.count</name> <value>1</value> <description>How many priority handlers to run; we probably need none. Default is 20 which is too much on a server like this.</description> </property> <property> <name>hbase.regionserver.replication.handler.count</name> <value>1</value> <description>How many replication handlers to run; we probably need none. Default is 3 which is too much on a server like this.</description> </property> <property> <name>hbase.regionserver.handler.count</name> <value>10</value> <description>How many default handlers to run; tie to # of CPUs. Default is 30 which is too much on a server like this.</description> </property> <property> <name>hbase.ipc.server.read.threadpool.size</name> <value>3</value> <description>How many Listener request reaaders to run; tie to a portion # of CPUs (1/4?). Default is 10 which is too much on a server like this.</description> </property> </configuration> {code}
h2. Sample Connection Implementation Has call-out for where an hbase-indexer would insert its capture code. {code} package org.apache.hadoop.hbase.client;
import com.google.protobuf.Descriptors; import com.google.protobuf.Message; import com.google.protobuf.Service; import com.google.protobuf.ServiceException; import org.apache.hadoop.conf.Configuration; import org.apache.hadoop.hbase.CompareOperator; import org.apache.hadoop.hbase.HTableDescriptor; import org.apache.hadoop.hbase.TableName; import org.apache.hadoop.hbase.client.coprocessor.Batch; import org.apache.hadoop.hbase.filter.CompareFilter; import org.apache.hadoop.hbase.ipc.CoprocessorRpcChannel; import org.apache.hadoop.hbase.security.User;
import java.io.IOException; import java.util.List; import java.util.Map; import java.util.concurrent.ExecutorService;
/** * Sample class for hbase-indexer. * DO NOT COMMIT TO HBASE CODEBASE!!! * Overrides Connection just so we can return a Table that has the * method that the replication sink calls, i.e. Table#batch. * It is at this point that the hbase-indexer catches the replication * stream so it can insert into the lucene index. */ public class IndexerConnection implements Connection { private final Configuration conf; private final User user; private final ExecutorService pool; private volatile boolean closed = false;
public IndexerConnection(Configuration conf, ExecutorService pool, User user) throws IOException { this.conf = conf; this.user = user; this.pool = pool; }
@Override public void abort(String why, Throwable e) {}
@Override public boolean isAborted() { return false; }
@Override public Configuration getConfiguration() { return this.conf; }
@Override public BufferedMutator getBufferedMutator(TableName tableName) throws IOException { return null; }
@Override public BufferedMutator getBufferedMutator(BufferedMutatorParams params) throws IOException { return null; }
@Override public RegionLocator getRegionLocator(TableName tableName) throws IOException { return null; }
@Override public Admin getAdmin() throws IOException { return null; }
@Override public void close() throws IOException { if (!this.closed) this.closed = true; }
@Override public boolean isClosed() { return this.closed; }
@Override public TableBuilder getTableBuilder(final TableName tn, ExecutorService pool) { if (isClosed()) { throw new RuntimeException(“IndexerConnection is closed.”); } final Configuration passedInConfiguration = getConfiguration(); return new TableBuilder() { @Override public TableBuilder setOperationTimeout(int timeout) { return null; }
@Override
public TableBuilder setRpcTimeout(int timeout) {
return null;
}
@Override
public TableBuilder setReadRpcTimeout(int timeout) {
return null;
}
@Override
public TableBuilder setWriteRpcTimeout(int timeout) {
return null;
}
@Override
public Table build() {
return new Table() {
private final Configuration conf = passedInConfiguration;
private final TableName tableName = tn;
@Override
public TableName getName() {
return this.tableName;
}
@Override
public Configuration getConfiguration() {
return this.conf;
}
@Override
public void batch(List\<? extends Row\> actions, Object[] results)
throws IOException, InterruptedException {
// Implementation goes here.
}
@Override
public HTableDescriptor getTableDescriptor() throws IOException {
return null;
}
@Override
public TableDescriptor getDescriptor() throws IOException {
return null;
}
@Override
public boolean exists(Get get) throws IOException {
return false;
}
@Override
public boolean[] existsAll(List\<Get\> gets) throws IOException {
return new boolean[0];
}
@Override
public \<R\> void batchCallback(List\<? extends Row\> actions, Object[] results, Batch.Callback\<R\> callback) throws IOException, InterruptedException {
}
@Override
public Result get(Get get) throws IOException {
return null;
}
@Override
public Result[] get(List\<Get\> gets) throws IOException {
return new Result[0];
}
@Override
public ResultScanner getScanner(Scan scan) throws IOException {
return null;
}
@Override
public ResultScanner getScanner(byte[] family) throws IOException {
return null;
}
@Override
public ResultScanner getScanner(byte[] family, byte[] qualifier) throws IOException {
return null;
}
@Override
public void put(Put put) throws IOException {
}
@Override
public void put(List\<Put\> puts) throws IOException {
}
@Override
public boolean checkAndPut(byte[] row, byte[] family, byte[] qualifier, byte[] value, Put put) throws IOException {
return false;
}
@Override
public boolean checkAndPut(byte[] row, byte[] family, byte[] qualifier, CompareFilter.CompareOp compareOp, byte[] value, Put put) throws IOException {
return false;
}
@Override
public boolean checkAndPut(byte[] row, byte[] family, byte[] qualifier, CompareOperator op, byte[] value, Put put) throws IOException {
return false;
}
@Override
public void delete(Delete delete) throws IOException {
}
@Override
public void delete(List\<Delete\> deletes) throws IOException {
}
@Override
public boolean checkAndDelete(byte[] row, byte[] family, byte[] qualifier, byte[] value, Delete delete) throws IOException {
return false;
}
@Override
public boolean checkAndDelete(byte[] row, byte[] family, byte[] qualifier, CompareFilter.CompareOp compareOp, byte[] value, Delete delete) throws IOException {
return false;
}
@Override
public boolean checkAndDelete(byte[] row, byte[] family, byte[] qualifier, CompareOperator op, byte[] value, Delete delete) throws IOException {
return false;
}
@Override
public void mutateRow(RowMutations rm) throws IOException {
}
@Override
public Result append(Append append) throws IOException {
return null;
}
@Override
public Result increment(Increment increment) throws IOException {
return null;
}
@Override
public long incrementColumnValue(byte[] row, byte[] family, byte[] qualifier, long amount) throws IOException {
return 0;
}
@Override
public long incrementColumnValue(byte[] row, byte[] family, byte[] qualifier, long amount, Durability durability) throws IOException {
return 0;
}
@Override
public void close() throws IOException {
}
@Override
public CoprocessorRpcChannel coprocessorService(byte[] row) {
return null;
}
@Override
public \<T extends Service, R\> Map\<byte[], R\> coprocessorService(Class\<T\> service, byte[] startKey, byte[] endKey, Batch.Call\<T, R\> callable) throws ServiceException, Throwable {
return null;
}
@Override
public \<T extends Service, R\> void coprocessorService(Class\<T\> service, byte[] startKey, byte[] endKey, Batch.Call\<T, R\> callable, Batch.Callback\<R\> callback) throws ServiceException, Throwable {
}
@Override
public \<R extends Message\> Map\<byte[], R\> batchCoprocessorService(Descriptors.MethodDescriptor methodDescriptor, Message request, byte[] startKey, byte[] endKey, R responsePrototype) throws ServiceException, Throwable {
return null;
}
@Override
public \<R extends Message\> void batchCoprocessorService(Descriptors.MethodDescriptor methodDescriptor, Message request, byte[] startKey, byte[] endKey, R responsePrototype, Batch.Callback\<R\> callback) throws ServiceException, Throwable {
}
@Override
public boolean checkAndMutate(byte[] row, byte[] family, byte[] qualifier, CompareFilter.CompareOp compareOp, byte[] value, RowMutations mutation) throws IOException {
return false;
}
@Override
public boolean checkAndMutate(byte[] row, byte[] family, byte[] qualifier, CompareOperator op, byte[] value, RowMutations mutation) throws IOException {
return false;
}
@Override
public void setOperationTimeout(int operationTimeout) {
}
@Override
public int getOperationTimeout() {
return 0;
}
@Override
public int getRpcTimeout() {
return 0;
}
@Override
public void setRpcTimeout(int rpcTimeout) {
}
@Override
public int getReadRpcTimeout() {
return 0;
}
@Override
public void setReadRpcTimeout(int readRpcTimeout) {
}
@Override
public int getWriteRpcTimeout() {
return 0;
}
@Override
public void setWriteRpcTimeout(int writeRpcTimeout) {
}
};
}
};
} } {code}
GlobalQuotaSettings was introduced to avoid protocol-specific Java classes from leaking into API which is users may leverage. This class has a number of methods which return plain-Java-objects instead of these protocol-specific classes in an effort to better provide stability in the future.
The RPC calls for Add/Modify/DeleteColumn have been removed and are now backed by ModifyTable functionality. The corresponding permissions in AccessController have been removed as well.
The shell already bypassed these RPCs and used ModifyTable directly, and thus would not be getting these permission checks, this change brings the rest of the RPC inline with that.
Coprocessor hooks for pre/post Add/Modify/DeleteColumn have likewise been removed. Coprocessors needing to take special actions on schema change should instead process ModifyTable events (which they should have been doing already, but it was easy for developers to miss this nuance).
HBase has upgraded from Jackson 1 to Jackson 2. JSON output should not have changed and this should not be user facing, but server classpaths should be adjusted accordingly.
Add new split algorithm DecimalStringSplit，row are decimal-encoded long values in the range “00000000” => “99999999” . create ‘t1’,‘f’, { NUMREGIONS => 10 , SPLITALGO => ‘DecimalStringSplit’ } The split point will be 10000000,20000000,...,90000000
Removed CP exposed StoreFile#getHDFSBlockDistribution
Add two new methods in CompactionLifeCycleTracker. The notExecuted method will be called if the selectCompaction failed or space quota limitation reached. The completed method will be called after all the requested compactions are finished. The compaction scheduling is pre Store so if you request compaction on a region it may lead to multiple compactions. Remove the User parameter in Region.requestCompaction methods as it is useless for CP users. Add a boolean parameter to indicate whether you want to do a major compaction. And so that the triggerMajorCompaction method is removed. Remove the getCompactionProgress method in Store interface. Add a UT to confirm that CompactionLifeCycleTracker works correctly, and it also shows how to use CompactionLifeCycleTracker to wait for the completion of a compaction.
RegionObserver#postCompactSelection signature is changed. Arg type org.apache.hadoop.hbase.shaded.com.google.common.collect.ImmutableList is replaced with java.util.List
Removes getTable from the CoprocessorEnvrionment Interface and from the BaseEnvironment implementation. Also removes TableWrapper and CoprocessorHConnection, two classes that were used by BaseEnvironment to keep a tag on Tables created by Coprocessors that BaseEnvironment might close them out on #shutdown.
Long after these classes and methods were added, in HBase 1.0.0, we moved to a mode where management of Tables was shifted from HBase to the Client; the Client is to manage lifecycle. Table also became a (relatively) lightweight construct so folks are used to getting a Table instance, using it, and then immediately closing it when done.
Coprocessors should do the same in hbase2.0.0.
CoprocessorHConnection short-circuited RPC. This feature has since been integrated into Server Connections; when they create a Connection, they get one that will short-circuit if the request is to a localhost so no need of CoprocessorHConnection any more.
Coprocessors get the Server Connection when they ask for a Connection from their *CoprocessorEnvironment.
Running tests with a wildcard selector, i.e.{{-Dtest=org.apache.hadoop.hbase.server.*}} no longer works.
Added three top level Admin APIs to help decommissioning and graceful stop of region servers.
/** * Mark region server(s) as decommissioned to prevent additional regions from getting * assigned to them. Optionally unload the regions on the servers. If there are multiple servers * to be decommissioned, decommissioning them at the same time can prevent wasteful region * movements. Region unloading is asynchronous. * @param servers The list of servers to decommission. * @param offload True to offload the regions from the decommissioned servers */ void decommissionRegionServers(List<ServerName> servers, boolean offload) throws IOException;
/** * List region servers marked as decommissioned, which can not be assigned regions. * @return List of decommissioned region servers. */ List<ServerName> listDecommissionedRegionServers() throws IOException;
/** * Remove decommission marker from a region server to allow regions assignments. * Load regions onto the server if a list of regions is given. Region loading is * asynchronous. * @param server The server to recommission. * @param encodedRegionNames Regions to load onto the server. */ void recommissionRegionServer(ServerName server, List<byte[]> encodedRegionNames) throws IOException;
Precommit switched from Oracle JDK 8 to OpenJDK-8.
CellCompartor has been added as an interface with IA.LimitedPrivate. It has the following methods #int compare(Cell leftCell, Cell rightCell); #int compareRows(Cell leftCell, Cell rightCell) #int compareRows(Cell cell, byte[] bytes, int offset, int length) #int compareWithoutRow(Cell leftCell, Cell rightCell) #int compareFamilies(Cell leftCell, Cell rightCell #int compareQualifiers(Cell leftCell, Cell rightCell) #int compareTimestamps(Cell leftCell, Cell rightCell) #int compareTimestamps(long leftCellts, long rightCellts)
This is exposed to CPs and CPs can make use of the above methods to do comparisons on the cells. For internal usage we have CellComparatorImpl and it has static references to COMPARATOR and META_CELL_COMPARATOR. So when a region or store is initialized we should use one of the above comparator. For META table we need the META_CELL_COMPARATOR and all other table's regions/stores will use the COMPARTOR. While writing the comparator name in FixedFileTrailer of the Hfile we have now ensured that this rename of CellComparator.COMPARATOR/CellComparator.META_CELL_COMPARATOR to CellComparatorImpl.COMPARATOR/CellComparatorImpl.META_CELL_COMPARATOR is handled.
CellUtils is an util method that provides lot of APIs that helps to do compare, matching functionalities between two cells, or with a cell and a corrpesponding byte[] etc. Some of the APIs are internally used which will be cleaned up in a follow on JIRA HBASE-18995.
These methods are removed: KeyValueScanner preStoreScannerOpen(ObserverContext<RegionCoprocessorEnvironment> c, Store store, Scan scan, NavigableSet<byte[]> targetCols, KeyValueScanner s, long readPt) throws IOException; InternalScanner preFlushScannerOpen(ObserverContext<RegionCoprocessorEnvironment> c, Store store, List<KeyValueScanner> scanners, InternalScanner s, long readPoint) throws IOException; InternalScanner preCompactScannerOpen(ObserverContext<RegionCoprocessorEnvironment> c, Store store, List<? extends KeyValueScanner> scanners, ScanType scanType, long earliestPutTs, InternalScanner s, CompactionLifeCycleTracker tracker, CompactionRequest request, long readPoint) throws IOException;
For flush and compaction, CP users are expected to wrap the InternalScanner in preFlush/preCompact. And for normal region operation, just use preGetOp/preScannerOpen to modify the Get/Scan object.
This method in Region interface is also removed as we do not need to use read point in CP hooks anymore: long getReadPoint(IsolationLevel isolationLevel);
Moves RSGroup on to AMv2. Reenables disabled RSGroups tests.
All operations for which further processing is skipped by preBatchMutate coprocessor hook are treated as SUCCESS instead of FAILED.
Add a new config hbase.separate.oldlogdir.by.regionserver. The default value is false. If this config is true, the old wal dir will be separated by regionservers. This will change the oldWALs layout. The oldWALs is used by replication. So if a cluster didn't use replication, it can be rolling upgrade (upgrade this config from false to true) directly. If a cluster use replication, the oldWALs will be not found when layout changed. So the cluster need rolling upgrade twice. Firstly, only rolling cluster to use new version code. Secondly rolling the config from false to true. Because the cluster already rolling to new version code, so it can find the oldWALs in the new dir layout.
This optimization, targeting SEEK_NEXT_USING_HINT return values, utilizes the max seek value and is transparent to Filters.
Modify ZooKeeperScanPolicyObserver in hbase-examples to show how to do filtering in the CP hooks of flush and compaction in hbase-2.0.
The archived Procedure WALs are moved to <hbase_root>/oldWALs/masterProcedureWALs directory. TimeToLiveProcedureWALCleaner class was added which regularly cleans the Procedure WAL files from there.
The TimeToLiveProcedureWALCleaner is added to hbase.master.logcleaner.plugins configuration value.
A new config parameter is added: hbase.master.procedurewalcleaner.ttl, which specifies how long a Procedure WAL should stay in the archive directory.
Below methods are removed from CP exposed Region interface getOpenSeqNum getOldestSeqIdOfStore isLoadingCfsOnDemandDefault getReadpoint updateReadRequestsCount updateWriteRequestsCount getRegionServicesForStores getMetrics getHDFSBlocksDistribution releaseRowLocks batchReplay get(Get get, boolean withCoprocessor, long nonceGroup, long nonce) bulkLoadHFiles execService registerService checkFamilies checkTimestamps prepareDelete prepareDeleteTimestamps updateCellTimestamps flush compact waitForFlushesAndCompactions waitForFlushes
Change signature of below methods by dropping params ‘nonceGroup’, ‘nonce’ append(Append append, long nonceGroup, long nonce) batchMutate(Mutation[] mutations, long nonceGroup, long nonce) increment(Increment increment, long nonceGroup, long nonce)
Remove the CompactionRequest parameter in preCompactSelection as we do not have a CompactionRequest at that time.
Pushed to master and branch-2. Thanks all for reviewing.
Changes ObserverContext from a class to an interface and hides away constructor, testing functions and other internal-only functions in the implementation class.
WARNING: No release note provided for this change.
All the mappers and reducers output type will be now of MapReduceCell type. No more KeyValue type. How ever in branch-2 for compatibility we have allowed the older interfaces/classes that work with KeyValue to stay in the code base but they have been marked as deprecated. The following interfaces/classes have been deprecated in branch-2 Import#KeyValueWritableComparablePartitioner Import#KeyValueWritableComparator Import#KeyValueWritableComparable Import#KeyValueReducer Import#KeyValueSortImporter Import#KeyValueImporter KeyValueSortReducer KeyValueSerialization WALPlayer#WALKeyValueMapper
So any existing MR jobs that is using the above public interfaces/classes will continue to work in branch-2 and the expected output value type of those mappers and reducers can continue to be KeyValue type.
In branch-3 the mappers and reducers output will only expect MapReduceCell as the type and will no longer work with KeyValue type. The new public classes/interfaces added for branch-3 and in branch-2 are CellSerialization CellSortReducer Import#CellWritableComparablePartitioner Import#CellWritableComparable Import#CellWritableComparator Import#CellReducer Import#CellSortImporter Import#CellImporter WALPlayer#WALCellMapper
The changes of IA.Public/IA.LimitedPrivate classes are shown below: HTableDescriptor class * boolean hasRegionMemstoreReplication()
RegionLoadStats class * int getMemstoreLoad()
ServerLoad class * int getMemstoreSizeInMB()
Region class
Store class
StoreFile class
The CellChunkMap is very dense index for Memstore ImmutableSegment and the only one that can be taken off-heap. However, CellChunkMap works on-heap as well. The coding of the entire flow of working with CellChunkMap is not yet finished, thus CellChunkMap is disabled for usage so far. The continuation is done under HBASE-18232.
Curator version has been updated from 2.x to 4.0 (running in ZK 3.4 compatibility mode).
Users who experience classpath issues due to version conflicts are recommended to use either the hbase-shaded-client or hbase-shaded-mapreduce artifacts.
Move KeyValue.parseColumn() to CellUtil
The incompatible changes of IA.Public/LimitedPrivate classes are shown below.
HRegionLocation class
AsyncAdmin class
HBaseTestingUtility class
WALEdit class
RegionScanner class
RegionPlan class
Region class
TableSnapshotInputFormat.TableSnapshotRegionSplit class * HRegionInfo getRegionInfo()
RawAsyncTable.CoprocessorCallback class
WARNING: No release note provided for this change.
We are moving from Inheritence
We used to pass the RegionServerServices (RSS) which gave Coprocesosrs (CP) all sort of access to internal Server machinery. We now only allows the CP a subset of the RSS in the form of the CPRSS Interface. Particulars:
Removed method getRegionServerServices from CP exposed RegionCoprocessorEnvironment and RegionServerCoprocessorEnvironment and replaced with getCoprocessorRegionServerServices. This returns a new interface CoprocessorRegionServerServices which is only a subset of RegionServerServices. With that below methods are no longer exposed for CPs WAL getWAL(HRegionInfo regionInfo) List<WAL> getWALs() FlushRequester getFlushRequester() RegionServerAccounting getRegionServerAccounting() RegionServerRpcQuotaManager getRegionServerRpcQuotaManager() SecureBulkLoadManager getSecureBulkLoadManager() RegionServerSpaceQuotaManager getRegionServerSpaceQuotaManager() void postOpenDeployTasks(final PostOpenDeployContext context) void postOpenDeployTasks(final Region r) boolean reportRegionStateTransition(final RegionStateTransitionContext context) boolean reportRegionStateTransition(TransitionCode code, long openSeqNum, HRegionInfo... hris) boolean reportRegionStateTransition(TransitionCode code, HRegionInfo... hris) RpcServerInterface getRpcServer() ConcurrentMap<byte[], Boolean> getRegionsInTransitionInRS() Leases getLeases() ExecutorService getExecutorService() Map<String, Region> getRecoveringRegions() public ServerNonceManager getNonceManager() boolean registerService(Service service) HeapMemoryManager getHeapMemoryManager() double getCompactionPressure() ThroughputController getFlushThroughputController() double getFlushPressure() MetricsRegionServer getMetrics() EntityLock regionLock(List<HRegionInfo> regionInfos, String description, Abortable abort) void unassign(byte[] regionName) Configuration getConfiguration() ZooKeeperWatcher getZooKeeper() ClusterConnection getClusterConnection() MetaTableLocator getMetaTableLocator() CoordinatedStateManager getCoordinatedStateManager() ChoreService getChoreService() void stop(String why) void abort(String why, Throwable e) boolean isAborted() void updateRegionFavoredNodesMapping(String encodedRegionName, List<ServerName> favoredNodes) InetSocketAddress[] getFavoredNodesForRegion(String encodedRegionName) void addToOnlineRegions(Region region) boolean removeFromOnlineRegions(final Region r, ServerName destination)
Also 3 methods name have been changed List<Region> getOnlineRegions(TableName tableName) -> List<Region> getRegions(TableName tableName) List<Region> getOnlineRegions() -> List<Region> getRegions() Region getFromOnlineRegions(final String encodedRegionName) -> Region getRegion(final String encodedRegionName)
Signature of below methods in MasterObserver changed and instead of org.apache.hadoop.hbase.shaded.protobuf.generated.SnapshotDescription param, we will be passing org.apache.hadoop.hbase.client.SnapshotDescription preListSnapshot postListSnapshot preSnapshot postSnapshot preCloneSnapshot postCloneSnapshot preRestoreSnapshot postRestoreSnapshot preDeleteSnapshot postDeleteSnapshot
Also changed signature of RegionServerObserver#preReplicateLogEntries and preReplicateLogEntries by removing params List<org.apache.hadoop.hbase.shaded.protobuf.generated.AdminProtos.WALEntry>, org.apache.hadoop.hbase.CellScanner
No longer pass the protobuf request to prePrepareBulkLoad and preCleanupBulkLoad in BulkLoadObserver as part of our effort to purge protobuf from our Coprocessor API Interface (if you need to read the Table and RegionInfo, pull it from the passed in RegionCoprocessorEnvironment ObserverContext).
The following methods in QuotaSettings were annotated InterfaceAudience.Private; they are for internal use only in hbase-2.0.0
buildSetQuotaRequestProto(final QuotaSettings settings) setupSetQuotaRequest(SetQuotaRequest.Builder builder)
Note that there were versions of these methods in HBase 1.y that used classes in the {{org.apache.hadoop.hbase.protobuf.generated}} package. That package no longer exists as a part of our cleanup of protobufs from our public facing API and the related methods have been removed.
Cleanup the StoreFile interface.
The metadata keys are moved to HStoreFile.
These methods are removed: CacheConfig getCacheConf(); byte[] getMetadataValue(byte[] key); boolean isCompactedAway(); boolean isReferencedInReads(); void initReader() throws IOException; StoreFileScanner getPreadScanner(boolean cacheBlocks, long readPt, long scannerOrder, boolean canOptimizeForNonNullColumn); StoreFileScanner getStreamScanner(boolean canUseDropBehind, boolean cacheBlocks, boolean isCompaction, long readPt, long scannerOrder, boolean canOptimizeForNonNullColumn) throws IOException; StoreFileReader getReader(); void closeReader(boolean evictOnClose) throws IOException; void markCompactedAway(); void deleteReader() throws IOException;
Notice that these methods are still available in HStoreFile.
And the return value of getFirstKey and getLastKey are changed from Cell to Optional<Cell> to better indicate that they may not be available.
FileNotFoundException opening a StoreFile in a primary replica now causes a RegionServer to crash out where before it would be ignored (or optionally handled via close/reopen).
Adds a new plugin point ReplicationEndpoint. ReplicationSource, internal to hbase, tails the WAL and calls registered ReplicationEndpoints. ReplicationEndpoint implementations are responsible for actually shipping the edits to the other (hbase or non-hbase) cluster. ReplicationEndpoint can be defined per peer. Default inter-cluster replication works without any changes (lily etc should still work). ReplicationEndpoints have various facility including means for filtering out WAL edits source-side before they can be shipped to remote peers.
Now, delete.rb won't delete all versions of the specified column. It only delete the specified version (if user assigns a timestamp) or the latest version (default behavior)
Mark StoreFileScanner and StoreFileReader as IA.LimitPrivate(Phoenix). Deprecated the preStoreFileReaderOpen and postStoreFileReaderOpen method in RegionObserver to indicate that these methods are only supposed to be used by Phoenix.
Remove the following APIs from RegionServerObserver:
Specify an explicit version for javax.el. Without it we rely on repository cached metadata of which a prevalent version seems to list all versions between b01 and b08 but finishes with a b08-jbossorg which is in the jboss repo, a repo most of us do not list in our poms.
Provide alternate user-facing API that takes a RegionInfo Interface instead of a HRegionInfo; the old HRegionInfo methods have been deprecated in 2.0.0 and will be removed in 3.0.0.
Now when replicating a wal file which is still opened for write, we will get its committed length from the WAL instance in the same RS to prevent replicating uncommit WALEdit.
This is very important if you use AsyncFSWAL, as we use fan-out in AsyncFSWAL. The data written to DN will be visible immediately as all DNs think it is the end of a pipeline, although the client has not received an ack, and also NN may truncate the file if the client crashes at the same time.
Set version on branch-2 to be 2.0.0-alpha3 as part of RC making.
Moved on to commons-math3. Removed commons-math2.
Introduce a CompactionLifeCycleTracker to let the CP users know when the compaction starts and ends. CompactionRequest is marked as IA.Private and should be used in CP implementation any more.
The removed APIs are shown below.
* Deprecates getAlterStatus. Everywhere else we talk of ‘modify’ rather ‘alter’ and should use Future returned from async instead. * isTableAvailable(TableName, byte [][]) has been deprecated to be removed; use the overrie instead. This is a weird method. * Changed listTableDescriptor to getDescriptor. * Renamed other like methods to have same pattern (deprecating the old): balancer => balance setBalancerRunning => balancerSwitch setNormalizerRunning => normalizerSwitch enableCatalogJanitor => catalogJanitorSwitch setCleanerChoreRunning => cleanerChoreSwitch setSplitOrMergeEnabled => splitOrMergeEnabledSwitch
* Renamed (with deprecation of old) runCatalogScan => runCatalogJanitor. * Reviewed generated javadoc and made some edits; purged reference to hbase issues from our API, fixed param names, etc. * Made all the enable services methods have same pattern. * Renamed takeSnapshotAsync as snapshotAsync (with deprecation of old) * Renamed execProcedureWithRet as execProcedureWithReturn (with deprecation)
Purged a bunch of dependencies included but unused. Added reference to dependencies we do use but did not list (transitively included). Purged all but junit from parent pom dependency set and did explicit include in modules instead; not all modules need mockito, etc. Still work to do: grey area around hadoop and its transitive includes need cleanup still to make the dependency:analyze runs clean. Also figure how to purge junit from parent dependency list.
HBase now uses stability and audience annotations sourced from Apache Yetus, instead of the custom annotations that were previously in place.
These deprecated methods are removed from RegionObserver: InternalScanner preFlushScannerOpen(ObserverContext, Store, List, InternalScanner) throws IOException; void preCompactSelection(ObserverContext, Store, List) throws IOException; void postCompactSelection(ObserverContext, Store, ImmutableList); InternalScanner preCompact(ObserverContext, Store, InternalScanner, ScanType) throws IOException; InternalScanner preCompactScannerOpen(ObserverContext, Store, List, ScanType, long, InternalScanner, CompactionRequest) throws IOException; InternalScanner preCompactScannerOpen( ObserverContext, Store store, List, ScanType, long, InternalScanner) throws IOException; void preSplit(ObserverContext) throws IOException; void preSplit(ObserverContext, byte[]) throws IOException; void postSplit(ObserverContext, Region, Region) throws IOException; void preSplitBeforePONR(ObserverContext, byte[], List) throws IOException; void preSplitAfterPONR(ObserverContext) throws IOException; void preRollBackSplit(ObserverContext) throws IOException; void postRollBackSplit(ObserverContext) throws IOException; void postCompleteSplit(ObserverContext) throws IOException; long preIncrementColumnValue(ObserverContext, byte[], byte[], byte[], long, boolean) throws IOException; long postIncrementColumnValue(ObserverContextc, byte[], byte[], byte[], long, boolean, long) throws IOException; KeyValueScanner preStoreScannerOpen(ObserverContext, Store, Scan, NavigableSet, KeyValueScanner) throws IOException; boolean postScannerFilterRow(ObserverContext, InternalScanner, byte[], int, short, boolean) throws IOException; boolean postBulkLoadHFile(ObserverContext, List, boolean) throws IOException;
And this method is also removed since we never call it in our code base: InternalScanner preFlushScannerOpen(ObserverContext, Store, KeyValueScanner, InternalScanner, long) throws IOException;
The deprecated annotation is removed for these two methods as they are still being used: void preFlush(ObserverContext) throws IOException; void postFlush(ObserverContextc) throws IOException;
WALKey, @InterfaceAudience.LimitedPrivate(HBaseInterfaceAudience.REPLICATION), changed a bunch for 2.0.0. See below. We figured it ok hiding it since it should be internals anyway -- only we should be making them.
Adds more spec on how Get, Delete, and Put work and how they differ to help the user.
Incompatible move of WALEdit class from regionserver.wal to wal. Effects @InterfaceAudience.LimitedPrivate({ HBaseInterfaceAudience.REPLICATION, HBaseInterfaceAudience.COPROC })
(
Purge 0.94=>0.96 deprecated, migration code. This means that if you are on 0.94 and wish to go to hbase 2.0, you must first migrate to a version of hbase that is >= 0.96.
WARNING: No release note provided for this change.
Admin.listProcedures and Admin.listLocks were renamed to getProcedures and getLocks (listProcedures was added to hbase 1.2). This change was done in an incompatible way -- we just yanked listProcedures (Because Admin Interface is not compatible with hbase1).
Main changes:
- ProcedureInfo and LockInfo were removed, we use JSON instead of them
- Procedure and LockedResource are their server side equivalent
- Procedure protobuf state\_data became obsolate, it is only kept for
reading previously written WAL
- Procedure protobuf contains a state\_message field, which stores the internal
state messages (Any type instead of bytes)
- Procedure.serializeStateData and deserializeStateData were changed slightly
- Procedures internal states are available on client side
- Procedures are displayed on web UI and in shell in the following jruby format:
{ ID =\> '1', PARENT\_ID = '-1', PARAMETERS =\> [ ..extra state information.. ] }
Provide a new way to get desired ClusterStatus with a set of ClusterStatus.Option, such that the response back to client can be limited. Note that, the constructor way to new a ClusterStatus will be no longer support after 2.0.0, and use ClusterStatus.Builder instead.
WARNING: No release note provided for this change.
Deprecate checkAnd* APIs that take the filter CompareOp. Added new overrides that take a generic CompareOperator instead. CompareOperator will be used by checkAnd* in Table API and by filters going forward.
Other nice improvements suggested by this issue have been moved out to HBASE-18774.
After this jira, mergePool will be permanently removed from CompactSplitThread.
WARNING: No release note provided for this change.
Replaces hbase-shaded-server-<version>.jar with hbase-shaded-mapreduce-<version>.jar.
All the references to Protos in Admin.java have been removed and replaced with respective POJO classes. The references to Protos that were removed are AdminProtos.GetRegionInfoResponse, HBaseProtos.SnapshotDescription, HBaseProtos.SnapshotDescription.Type, MasterProtos.SnapshotResponse. CompactionType, CompactionState and MasterSwitchType Enums have been moved out of Admin.java to standalone Enums.
Move to commons-lang3 from common-lang (check it out!... Nice lib...Some nice utility)
Changed the passed arguments from HTD/HCD to TD/CFD for Admin.
Introduce a new o.a.h.h.tool.LoadIncrementalHFiles. The old o.a.h.h.mapreduce.LoadIncrementalHFiles is deprecated and will be removed in 3.0.0.
All constructors have already been deprecated. This change makes them InterfaceAudience Private.
Standalone clusters and minicluster instances can now configure the session timeout for our embedded ZooKeeper quorum using hbase.zookeeper.property.minSessionTimeout and hbase.zookeeper.property.maxSessionTimeout.
org.apache.hadoop.hbase.coprocessor.Export Instructs HBase to dump the contents of table to HDFS in a sequence file
The HBase shaded artifacts (hbase-shaded-client and hbase-shaded-server) no longer contain several non-relocated third party dependency classes that were mistakenly included. Downstream users who relied on these classes being present will need to add a runtime dependency onto an appropriate third party artifact.
Previously, we erroneously packaged several third party libs without relocating them. In some cases these libraries have now been relocated; in some cases they are no longer included at all.
Includes:
There is now testing to ensure that the shaded artifacts only contain expected relocated content. It can be run via mvn -Dtest=noUnitTests -pl hbase-shaded/hbase-shaded-check-invariants -am -Prelease verify.
For version 2.0+ this patch removes hadoop-mapreduce-client-core from the set of dependencies included for the hbase-client and hbase-shaded-client artifacts.
For 2.0+, the slf4j-log4j12 dependency is now optional for both shaded artifacts.
Previously some dependencies in hbase-shaded-client were still leaking into the un-shaded namespace. This should now be fixed.
Additionally the rat checking on generated intermediate files from shading should be skipped.
Performing reverse scan on tables used the meta cache incorrectly and fetched data from meta table every time. This fix solves this issue and which results in performance improvement for reverse scans.
This patch removed the storefile_index_size_MB in protobuf. It will cause the value of storefile_index_size_MB is zero if user still use hbase-client 1.x.
Introduce the CellBuilder helper.
Adds a new RefreshHFiles Coprocessor Endpoint example. Includes client and serverside-endpoint that iterates region Stores to call #refreshStoreFiles.
Removed hbase Service class. It was not fully-formed. Now Guava is relocated, use its Service instead internally; it has nice implementation facility too in AbstractService.
HBASE-15982 | Blocker | Interface ReplicationEndpoint extends Guava's Service
Breaking change to our ReplicationEndpoint and BaseReplicationEndpoint.
ReplicationEndpoint implemented Guava 0.12 Service. An abstract subclass, BaseReplicationEndpoint, provided default implementations and facility, among other things, by extending Guava's AbstractService class.
Both of these HBase classes were marked LimitedPrivate for REPLICATION so these classes were semi-public and made it so Guava 0.12 was part of our API.
Having Guava in our API was a mistake. It anchors us and the implementation of the Interface to Guava 0.12. This is untenable given Guava changes and that the Service Interface in particular has had extensive revamp and improvement done. We can‘t hold to the Guava Interface. It changed. We can’t stay on Guava 0.12; implementors and others on our CLASSPATH won't abide being stuck on an old Guava.
So we make breaking changes. The unhitching of our Interface from Guava could only be done in a breaking manner. It undoes the LimitedPrivate on BaseReplicationEndpoint while keeping it for the RE Interface. It means consumers will have to copy/paste the AbstractService-based BRE into their own codebase also supplying their own Guava; HBase no longer ‘supplies’ this (our Guava usage has been internalized, relocated).
This patch then adds into RE the basic methods RE needs of the old Guava Service rather than return a Service to start/stop only to go back to the RE instance to do actual work. A few method names had to be changed so could make implementations with Guava Service internally and not have RE method names and types clash). Semantics remained the same otherwise. For example startAsync and stopAsync in Guava are start and stop in RE.
Introduce an AsyncBufferedMutator for batching requests to HBase for a single table.
Use AsyncConnection.getBufferedMutator method to get an AsyncBufferedMutator instance.
If there is no existing cell in submitting Append/Increment, the custom ts won‘t be overridden. By contrast, the cell’s ts will always be overridden by server.
Moved from Jetty 9.3.x to 9.4.x.
Jetty returns more correct HTTP code when Header is too long, 431 instead of 413, and it requires more threads to start up (made default 16 instead of 10).
Move replication implementation's classes from hbase-client to hbase-replication package.
Change the yetus profile for branch-2 so it no longer runs hadoop 2.4.x and 2.5.x build checks.
Removed doubled instances of javax.inject and commons-beanutils where the versions were close.
Other instances of ‘double’ includes have different groupids so wary pruning especially when transitive includes (hadoop or jetty et al.)
This change invalidates the need for a separate Java properties file to configure the ChaosMonkey included with HBase. These properties can be provided directly in hbase-site.xml. If configuration in provided in both locations, the Java properties file takes precendence.
Add a ‘cursor’ method which returns an ‘Optional<Cursor>’ in ‘RawScanResultConsumer.ScanController’. You can use this method to obtain the scan cursor if available.
Changes the configuration hbase.balancer.tablesOnMaster from list of table names that the can carry (with ‘none’ meaning no tables on the master) to instead be a boolean that is set to true if master carries tables/regions and false if it does not. If true, the master acts like any regionserver.
If false, then the master carries no tables. This is the default for hbase-2.0.0.
Another boolean configuration, hbase.balancer.tablesOnMaster.systemTablesOnly, when set to true, enables hbase.balancer.tablesOnMaster and makes it so the master hosts system tables exclusively (the long-time deploy mode of master branch and branch-2 up until this commit).
UPDATE: This is broke. See HBASE-19785. UPDATE2: Master carrying Regions does not work reliably, see HBASE-19828.
See HBASE-19831, the issue to fix regions on Master
The change of hbase.balancer.tablesOnMaster from String list to boolean and the addition of a simple boolean to enable system-tables on Master was done to constrain what operators might ask for via this master configuration. Stipulating what tables are bound to the Master server verges into regionserver grouping territory, a more robust means of specifying table and server combinations. Operators should use this latter if they want layouts more exotic than those supplied by the provided booleans.
The ResultScanner which is gotten from an AsyncTable will also return cursor results if Scan.isNeedCursorResult is true.
Introduce FIFO algorithm to get a candidate locate request for AsyncNonMetaRegionLocator.
This patch exposes configuration for Bucketcache. These configs are very similar to those for the LRU cache, but are described below:
“hbase.bucketcache.single.factor”; /** Single access bucket size */ “hbase.bucketcache.multi.factor”; /** Multiple access bucket size */ “hbase.bucketcache.memory.factor”; /** In-memory bucket size */ “hbase.bucketcache.extrafreefactor”; /** Free this floating point factor of extra blocks when evicting. For example free the number of blocks requested * (1 + extraFreeFactor) */ “hbase.bucketcache.acceptfactor”; /** Acceptable size of cache (no evictions if size < acceptable) */ “hbase.bucketcache.minfactor”; /** Minimum threshold of cache (when evicting, evict until size < min) */
WARNING: No release note provided for this change.
Depend on hbase-thirdparty for our netty instead of directly relying on netty-all. netty is relocated in hbase-thirdparty from io.netty to org.apache.hadoop.hbase.shaded.io.netty. One kink is that netty bundles an .so. Its files also are relocated. So netty can find the .so content, need to specify on command-line a system property telling netty about the shading.
The .so trick is from https://stackoverflow.com/questions/33825743/rename-files-inside-a-jar-using-some-maven-plugin
In essence we need the below defined whenever we run tests or deploy:
-Dorg.apache.hadoop.hbase.shaded.io.netty.packagePrefix=org.apache.hadoop.hbase.shaded.
(The trailing ‘.’ is required)
See toward the end of this issue for how to pass config: https://github.com/netty/netty/issues/6665
The system property has been added to bin/hbase. If starting hbase with other than bin/hbase, add this system property (at least on linux).
For devs, going forward, do not reference io.netty. Reference org.apache.hadoop.hbase.io.netty instead. Here is sample:
{code} -import io.netty.channel.Channel; -import io.netty.channel.EventLoop; +import org.apache.hadoop.hbase.shaded.io.netty.channel.Channel; +import org.apache.hadoop.hbase.shaded.io.netty.channel.EventLoop; {code}
Provide a new way to get desired ClusterStatus with a set of ClusterStatus.Option, such that the response back to client can be limited. Note that, the constructor way to new a ClusterStatus will be no longer support after 2.0.0, and use ClusterStatus.Builder instead.
Unassign will not proceed if it is unable to talk to the remote server. Now it will expire the server it is unable to communicate with and then wait until it is signaled by ServerCrashProcedure that the server's logs have been split. Only then will judge the unassign successful.
We do this because a subsequent assign lacking the crashed server context might open a region w/o first splitting logs.
In HBASE-18469 we introduced a new RegionServer metrics in name of “totalRowActionRequestCount” which counts in all row actions and equals to the sum of “readRequestCount” and “writeRequestCount”. Meantime, we have changed “totalRequestCount” to count only once for multi request, while previously we will count in action number of the request. As a result, existing monitoring system on totalRequestCount will still work but see a smaller value, and we strongly recommend to change to use the new metrics to monitor server load.
Remove the deprecated method get/setWriteBufferSize from Table and remove writeBufferSize from TableBuilder. Remove the BufferedMutatorImpl from HTable.
This change allows the demonstration Thrift client to customize the server principal used by the Thrift server for instances secured with Kerberos.
Marked Scan and Get‘s setMaxVersions() and setMaxVersions(int) as deprecated. They are easy to misunderstand with column family’s max versions, so use readAllVersions() and readVersions(int) instead.
Favors new servers over older versions when assigning system table regions (more to follow in this area; i.e. changes in the AM itself).
Sets a log length max of 1000 characters.
The methods which change to use TableDescriptor/ColumnFamilyDescriptor are shown below.
This JIRA adds a JMX value to track when the Master has finished initializing. The jmx config is ‘masterFinishedInitializationTime’ and details the time in millis that the Master is fully usable and ready to serve requests.
Purge all checked in generated protobuf files (30MB). Generate protobuf files inline with the build. Remove checked-in and patched protobuf. Get it from new hbase-thirdparty instead.
Side-effect: Our protobuf went from 3.1.0 to 3.3.1.
Build does not take noticeably longer (still about 2.5 minutes to do a mvn clean install -DskipTests).
IDEs will probably require a mvn build first else they'll complain about missing (generated) files.
This change adds the latency metrics checkAndPut, checkAndDelete, putBatch and deleteBatch . Also the previous regionserver “mutate” latency metrics are renamed to “put” metrics. Batch metrics capture the latency of the entire batch containing put/delete whereas put/delete metrics capture latency per operation. Note this change will break existing monitoring based on regionserver “mutate” latency metric.
HBASE-18023 introduces a warning message in the RegionServer log when an RPC is received from a client that has more than 5000 “actions” (where an “action” is a collection of mutations for a specific row) in a single RPC. Misbehaving clients who send large RPCs to RegionServers can be malicious, causing temporary pauses via garbage collection or denial of service via crashes. The threshold of 5000 actions per RPC is defined by the property “hbase.rpc.rows.warning.threshold” in hbase-site.xml.
This issue resolved two long-term issues in HBase: Puts may be masked by a delete before them. Major compactions change query results.
This issue offer a new behavior to fix this issue with a little performance reduction. Set NEW_VERSION_BEHAVIOR to true to enable this feature in CF level. See HBASE-15968 for details. Note if you enable this feature, the order of Mutations matters. But replication will disorder the entries by default. So you have to enable serial replication if you have slave clusters. See HBASE-9465 for details.
Removes merge region code added into branch-2 but that was not needed after all. Branch-2 replaced dispatchMergingRegions with MergeTableRegionsProcedure.
Removed:
Added setPriority(int priority) API to Put, Delete, Increment, Append, Get and Scan pojos. So for all these ops, the user can provide a custom priority level.
Pushed to {{master}}. Thanks, Coral! Congratulations on your first Apache HBase commit!
Use relocated guava 22.0 gotten from the new hbase-thirdparty ancillary project.
Incompatible change. ReplicationEndpoint and subclasses extend guava Service which changed pretty radically between 12.0 and 22.0. Change is kosher because implementations are marked audience private. Still, this will likely cause grief for the likes of the downstream lily indexer.
Any value for hbase.bucketcache.bucket.sizes configuration to be multiple of 256. If that is not the case, instantiation of L2 Bucket cache itself will fail throwing IllegalArgumentException.
pushed to 1.3 and 1.2. SyncTable was introduced in 1.2, so skipping 1.1.
Committed to master and branch-2. Thanks!
In order to use this feature, a user must
To register their tables (and configure their job for incremental load into multiple tables), a user must call the static MultiHFileOutputFormat.configureIncrementalLoad function to register the HBase tables that will be ingested into.
To create the composite key, a helper function MultiHFileOutputFormat2.createCompositeKey should be called with the destination tablename and rowkey as arguments, and the result should be output as the mapper key.
Before this JIRA, for HFileOutputFormat2 a configuration for the storage policy was set per Column Family. This was set manually by the user. In this JIRA, this is unchanged when using HFileOutputFormat2. However, when specifically using MultiHFileOutputFormat2, the user now has to manually set the prefix by creating a composite of the table name and the column family. The user can create the new composite value by calling MultiHFileOutputFormat2.createCompositeKey with the tablename and column family as arguments.
Changes added through this JIRA are backwards compatible with existing HFileOutputFormat2 apis and functionality.
The configuration parameter “hbase.mapreduce.hfileoutputformat.table.name” is now a REQUIRED parameter though it is normally set automatically when configureIncrementalLoad method is called within HFileOutputFormat2
A new splitRegionAsync() API is added in client. The existing splitRegion() and split() API will call the new API so client does not have to change its code.
Move HBaseAdmin.splitXXX() logic to master, client splitXXX() API now go to master directly instead of going to RegionServer first.
Also added splitSync() API
HBase now defaults to Apache Hadoop 3.0.0-alpha4 when the Hadoop 3 profile is active.
WARNING: No release note provided for this change.
There are two configuration name changes as the event loop configs will not only effect rpc server but be shared by different components in the same RS instance.
‘hbase.rpc.server.nativetransport’ -> ‘hbase.netty.nativetransport’
‘hbase.netty.rpc.server.worker.count’ -> ‘hbase.netty.worker.count’
Region class
Store class
Table class * HTableDescriptor getTableDescriptor()
*Admin class* * HTableDescriptor getTableDescriptor(TableName)
*HBaseTestingUtility class* * Table createTable(HTableDescriptor, byte[][], Configuration)
After HBASE-18083 we could configure HFileCleaner to use multiple threads for large/small (archived) hfile cleaning with hbase.regionserver.hfilecleaner.large.thread.count and hbase.regionserver.hfilecleaner.small.thread.count, both default to 1. These properties support online configuration change.
We usually keep compatibility between old client and new server so we can do rolling upgrade, HBase cluster first, then HBase client. But we don't guarantee new client can access old server. In an HBase cluster, we have system tables and region servers will access these tables so for servers they are also an HBase client. So if the system tables are in region servers with lower version we may get trouble because region servers with higher version may can not access them. After this patch, we will move all system regions to region servers with highest version. So when we do a rolling upgrade across two major or minor versions, we should ALWAYS UPGRADE MASTER FIRST and then upgrade region servers. The new master will handle system tables correctly.
Make us build against hadoop trunk (3.0)
Add unit tests for truncate_preserve
Adds a new project, hbase-thirdparty, at https://git-wip-us.apache.org/repos/asf/hbase-thirdparty used by core hbase. GroupID org.apache.hbase.thirdparty. Version 1.0.0.
This project packages relocated third-party libraries used by Apache HBase such as protobuf, guava, and netty among others. HBase core depends on it.
It has threre submodules, one to patch and then relocate (shade) protobuf, and one to do messy .so renaming (netty). The remainder module relocates a bundle of other (unpatched) libs used by hbase. This latter set includes protobuf-util, netty-all, gson, and guava.
All shading is done using the same relocation offset of org.apache.hadoop.hbase.shaded; we add this prefix to the relocated thirdparty library class names.
See the pom.xml in hbase-thirdparty for the explicit version of each third-party lib included (of note, we update out internal protobuf from 3.1.0 to 3.3.1).
Adds new "Process Metrics' tab along the top which leads to new page that dumps mbean -- mostly jvm -- metrics
Changes the checkstyle so that on a continuation line for javadoc, instead of default four spaces, instead now it is two spaces. Also one line statements as in if (true) x =1; now pass checkstyle.
After HBASE-17110 the bytable strategy for SimpleLoadBalancer will also take server level balance into account
Adds clear_compaction_queues to the hbase shell. {code} Clear compaction queues on a regionserver. The queue_name contains short and long. short is shortCompactions‘s queue,long is longCompactions’s queue.
Examples: hbase> clear_compaction_queues ‘host187.example.com,60020’ hbase> clear_compaction_queues ‘host187.example.com,60020’,‘long’ hbase> clear_compaction_queues ‘host187.example.com,60020’, [‘long’,‘short’] {code}
New locality cost function and candidate generator that use caching and incremental computation to allow the stochastic load balancer to consider ~20x more cluster configurations for big clusters.
The following config is added by this JIRA:
hbase.regionserver.hostname.disable.master.reversedns
This config is for experts: don't set its value unless you really know what you are doing. When set to true, regionserver will use the current node hostname for the servername and HMaster will skip reverse DNS lookup and use the hostname sent by regionserver instead. Note that this config and hbase.regionserver.hostname are mutually exclusive. See https://issues.apache.org/jira/browse/HBASE-18226 for more details.
Caution: please make sure rolling upgrade succeeds before turning on this feature.
Apache HBase now specifies that version 1.7.7 of the Apache Avro library should be pulled in by maven and included in the convenience binary tarball.
Add documentation for async client in section ‘66. Client’ in ref guide.
Add two examples for async client. AsyncClientExample is a simple example to show you how to use AsyncTable. HttpProxyExample is an example for advance user to show you how to use RawAsyncTable to write a fully asynchronous HTTP proxy server. There is no extra thread pool, all operations are executed inside netty's event loop.
Allow setting different hadoop check versions for branch-2 and branch-2.x when running pre commit check.
Pushed the release. For detail: http://apache-hbase.679495.n3.nabble.com/ANNOUNCE-Apache-HBase-2-0-0-alpha-1-is-now-available-for-download-td4088484.html
0-length WAL files can potentially cause the replication queue to get stuck. A new config “replication.source.eof.autorecovery” has been added: if set to true (default is false), the 0-length WAL file will be skipped after 1) the max number of retries has been hit, and 2) there are more WAL files in the queue. The risk of enabling this is that there is a chance the 0-length WAL file actually has some data (e.g. block went missing and will come back once a datanode is recovered).
If a region server that is processing recovered queue for another previously dead region server is gracefully shut down, it can drop the recovered queue under certain conditions. Running without this fix on a 1.2+ release means possibility of continuing data loss in replication, irrespective of which WALProvider is used. If a single WAL group (or DefaultWALProvider) is used, running without this fix will always cause dataloss in replication whenever a region server processing recovered queues is gracefully shutdown.
Adds a sort of procedures before submission so system tables are queued first (which will help ensure they go out first). This should be good enough along w/ existing scheduling mechanisms to ensure system/meta are assigned first (See reasoning below). Open new issue if insufficient.
If the table-scope attributes value is false, you need not to enclose ‘false’ in single quotation.Both COMPACTION_ENABLED => false and COMPACTION_ENABLED => ‘false’ will take effect
When randomRead and randomSeekScan is used with PE tool, now we allow using both --size and --rows. The --size specifies the total size of the data (the range) on which the reads should be performed and --rows specifies the number of rows to be read by each client with in that range.
If you don't like scanning being blocked too long because of heartbeat and partial result, you can use Scan#setNeedCursorResult(true) to get a special result within scanning timeout setting time which will tell you where row the server is scanning. See its javadoc for more details.
Following AMv2 procedures are modified to override onSubmit(), onFinish() hooks provided by HBASE-17888 to do metrics calculations when procedures are submitted and finshed: * AssignProcedure * UnassignProcedure * MergeTableRegionProcedure * SplitTableRegionProcedure * ServerCrashProcedure
Following metrics is collected for each of the above procedure during lifetime of a process: * Total number of requests submitted for a type of procedure * Histogram of runtime in milliseconds for successfully completed procedures * Total number of failed procedures
As we are moving away from Hadoop's metric2, hbase-metrics-api module is used for newly added metrics.
To handle this issue client need to have Hadoop client 2.6.4 or 2.7.0+ Hadoop version as CanUnBuffer interface which was added as part of HDFS-7694 is available in only those versions.
StoreFile is now changed to an interface. This is an incompatible change. The coprocessors which implement RegionObserver may need to modify their code.
The bundled JRuby 1.6.8 has been updated to version 9.1.9.0. The represents a change from Ruby 1.8 to Ruby 2.3.3, which introduces non-compatible language changes for user scripts.
This JRuby version update required an update to joni-2.1.11 and jcodings-1.0.18, used for regular expression matching, as well as several transitive dependency updates that should not be user-visible.
Replaces the AssignmentManager with a new procedurev2-based AssignmentManager
h1. AMv2 Puts AssignmentManager up on top of the ProcedureV2 state machine with persistence engine. Each assignment atom is now a Procedure implementation; e.g. an AssignProcedure and an UnassignProcedure. Molecules of aggregated Procedures are used to do more involved assignment steps: e.g. the move region procedure is made of an Unassign followed by an Assign subprocedure.
AMv2 is 1500 lines. Old AM was near 4000. Functionality has been moved out to Procedures. In-memory states of regions and servers has been cleaned up stored in new RegionStates implementation. RegionStateStore takes care of publishing final region state out to the hbase:meta table.
New RemoteProcedureDispatcher/RSProcedureDispatcher runs the Procedure-based assignments ‘remotely’. Knows about ‘servers’. Does aggregation of assignments by time on a time/count basis so can send procedures in batches rather than one per RPC. Procedure status comes back on the back of the RegionServer heartbeat reporting online regions. The response is passed to the AMv2 to ‘process’. It will check against the in-memory state. If there is a mismatch, it fences out the RegionServer on the assumption that something went wrong on the RS side.Timeouts trigger retries. The Procedure machine ensures only one operation at a time on any one region/table using locking and smarts about what is serial and what can be run concurrently.
New accounting of RegionServer version will be used running rolling restarts.
‘States’ -- OPENING, CLOSING, etc. -- are now in-memory in-the-master only serialized out to the ProcedureV2 WAL. They are no longer persisted to ZooKeeper.
h2. Assign Detail The Assign starts by pushing the “assign” operation to the AssignmentManager and then will go into a “waiting" state. The AM will batch the “assign” requests and ask the Balancer where to put the region (the various policies will be respected: retain, round-robin, random). Once the AM and the balancer have found a place for the region, the procedure will be resumed and an “open region” request will be placed in the Remote Dispatcher queue, and the procedure once again will go into a “waiting state”. The Remote Dispatcher will batch the various requests for that server and they will be sent to the RS for execution. The RS will complete the open operation by calling master.reportRegionStateTransition(). The AM will intercept the transition report, and notify the procedure. The procedure will finish the assignment by publishing to new state on hbase:meta or it will retry the assignment.
h3. Unassign Detail The Unassign starts by placing a “close region” request in the Remote Dispatcher queue, and the procedure will then go into a “waiting state”. The Remote Dispatcher will batch the various requests for that server and they will be sent to the RS for execution. The RS will complete the open operation by calling master.reportRegionStateTransition(). The AM will intercept the transition report, and notify the procedure. The procedure will finish the unassign by publishing its new state on meta or it will retry the unassign.
h1. New Configs * “hbase.procedure.remote.dispatcher.threadpool.size” defaults 128 * “hbase.procedure.remote.dispatcher.delay.msec” default 150ms * “hbase.procedure.remote.dispatcher.max.queue.size” with default 32 * “hbase.regionserver.rpc.startup.waittime” with default 60 seconds. h1. TODO As of this writing.
Put up a model diagram.
* Handle region migration * Handle meta assignment first * Handle sys table assignment first (e.g. acl, namespace) * Handle table priorities * Do we report same AM metrics as we used too? We do it all in here now.
INCOMPATIBLE A known incompatible is that because splits and merges are now run from the master, Coprocessors that used to watch for merge/split from a RegionObserver now no longer work; to watch split/merges, you need to have an observer on the Master instead.
UI pages for splitting/merging now operate by taking a row key prefix from the user rather than a full region name.
The command truncate_preserve will be fine when the truncate method doesn't exist on the master
The scanner id is not from 1 anymore. The first 32 bits are MurmurHash32 of ServerName string “host,port,ts”. The ServerName contains both host, port, and start timestamp so it can prevent collision. The lowest 32bit is generated by atomic int.
When JRUBY_HOME is specified, if the command is “hbase shell” or “hbase org.jruby.Main”, CLASSPATH and HBASE_OPTS will be updated according to JRUBY_HOME specified * Jar under JRUBY_HOME is added to CLASSPATH * The following will be added into HBASE_OPTS
-Djruby.home=$JRUBY_HOME -Djruby.lib=$JRUBY_HOME/lib
That is, as long as JRUBY_HOME is specified, JRUBY_HOME specified will take precedence. * In dev env, the jar recorded in cached_classpath_jruby.txt will be ignored * In non dev env, jruby-complete jar packaged with HBase will be ignored
After this issue, all table operations will support null qualifier, such as put/get/scan/increment/append/checkAndMutate/checkAndPut/checkAndDelete.
When a client is configured to use meta replica, it sends scan request to all meta replicas almost at the same time. Since meta replica contains stale data, if result from one of replica comes back first, the client may get wrong region locations. To fix this, “hbase.client.meta.replica.scan.timeout” is introduced, a client will always send to primary meta region first, wait the configured timeout for reply. If no result is received, it will send request to replica meta regions. The unit for “hbase.client.meta.replica.scan.timeout” is microsecond, the default value is 1000000 (1 second).
While creating a snapshot, it will save permissions of the original table into .snapshotinfo file(Backward compatibility) , which is in the snapshot root directory. For clone_snapshot/restore_snapshot command, we provide an additional option( RESTORE_ACL) to decide whether we will grant permissons of the origin table to the newly created table.
The default behavior for abort() method of StateMachineProcedure class is changed to support aborting all procedures irrespective of if procedure supports rollback or not.
Two blog posts on Apache HBase blog: user manual and programmer manual. Ref. guide draft published: https://docs.google.com/document/d/1Xi1jh_30NKnjE3wSR-XF5JQixtyT6H_CdFTaVi78LKw/edit
This JIRA changes the default MemStore to be CompactingMemStore instead of DefaultMemStore. In-memory compaction of CompactingMemStore demonstrated sizable improvement in HBase’s write amplification and read/write performance.
CompactingMemStore achieves these gains through smart use of RAM. The algorithm periodically re-organizes the in-memory data in efficient data structures and reduces redundancies. The HBase server’s memory footprint therefore periodically expands and contracts. The outcome is longer lifetime of data in memory, less I/O, and overall faster performance. More details about the algorithm and its use appear in the Apache HBase Blog: https://blogs.apache.org/hbase/
How To Use: The in-memory compaction level can be configured both globally and per column family. The supported levels are none (DefaultMemStore), basic, and eager.
By default, all tables apply basic in-memory compaction. This global configuration can be overridden in hbase-site.xml, as follows:
<property> <name>hbase.hregion.compacting.memstore.type</name> <value><none|basic|eager></value> </property>
The level can also be configured in the HBase shell per column family, as follows:
create ‘<tablename>’, {NAME => ‘<cfname>’, IN_MEMORY_COMPACTION => ‘<NONE|BASIC|EAGER>’}
$ bin/hbase org.apache.hadoop.hbase.master.balancer.LoadBalancerPerformanceEvaluation -help usage: hbase org.apache.hadoop.hbase.master.balancer.LoadBalancerPerformanceEvaluation <options> Options: -regions <arg> Number of regions to consider by load balancer. Default: 1000000 -servers <arg> Number of servers to consider by load balancer. Default: 1000 -load_balancer <arg> Type of Load Balancer to use. Default: org.apache.hadoop.hbase.master.balancer.StochasticLoadBalancer
Now we pass on list of memstoreScanners to the StoreScanner along with the new files to ensure that the StoreScanner sees the latest memstore after flush.
<!-- mardown --> Refactor that breaks out StoreFile Reader and Writer inner classes as StoreFileReader and StoreFileWriter.
NOTE! Changes RegionObserver Coprocessor Interface so incompatible change (Discussed on dev list in thread “Note breaking change on RegionObserver in hbase-2.0.0”
The JRuby jar is no longer automatically included in classpaths for HBase server processes nor clients. It is still included in the classpath for the HBase shell and for invocations of org.jruby.Main, which should cover HBase provided support scripts.
The return value of CallRunner.getCall is changed so this is an incompatible change as CallRunner is declared as IA.LimitedPrivate. CallRunner is declared as IS.Evolving so we do not break the rule. And we still keep the getCall method to reduce the impact to user code.
Added a shell command ‘list_regions’ for displaying the table's region info through command line.
List all regions for a particular table as an array and also filter them by server name (optional) as prefix
and maximum locality (optional). By default, it will return all the regions for the table with any locality.
The command displays server name, region name, start key, end key, size of the region in MB, number of requests
and the locality. The information can be projected out via an array as third parameter. By default all these information
is displayed. Possible array values are SERVER\_NAME, REGION\_NAME, START\_KEY, END\_KEY, SIZE, REQ and LOCALITY. Values
are not case sensitive. If you don't want to filter by server name, pass an empty hash / string as shown below.
Examples:
hbase\> list\_regions 'table\_name'
hbase\> list\_regions 'table\_name', 'server\_name'
hbase\> list\_regions 'table\_name', {SERVER\_NAME =\> 'server\_name', LOCALITY\_THRESHOLD =\> 0.8}
hbase\> list\_regions 'table\_name', {SERVER\_NAME =\> 'server\_name', LOCALITY\_THRESHOLD =\> 0.8}, ['SERVER\_NAME']
hbase\> list\_regions 'table\_name', {}, ['SERVER\_NAME', 'start\_key']
hbase\> list\_regions 'table\_name', '', ['SERVER\_NAME', 'start\_key']
MVCCPreAssign is added by HBASE-16698, but pre-assign mvcc is only used in put/delete path. Other write paths like increment/append still assign mvcc in ringbuffer's consumer thread. If put and increment are used parallel. Then seqid in WAL may not increase monotonically. Disorder in wals will lead to data loss.This patch bring all mvcc/seqid event in wal.append, and synchronize wal append and mvcc acquirement. No disorder in wal will happen. Performance test shows no regression with this patch.
Support for snapshots in VerifyReplication tool i.e. verifyrep can compare source table snapshot against peer table snapshot which reduces load on RS by reading data from HDFS directly using Snapshot scanners. Instead of comparing against live tables whose state changes due to writes and compactions its better to compare HBase snapshots which are immutable in nature.
A new RPC server based on Netty4 which can improve random read (get) performance. By default, it is off. To use this feature, please set “hbase.rpc.server.impl" to “org.apache.hadoop.hbase.ipc.NettyRpcServer”.
In one deploy, doubled the throughput and lowered the latency significantly: see https://www.slideshare.net/HBaseCon/lift-the-ceiling-of-hbase-throughputs?qid=597ee2fa-8125-4faa-bb3b-2bf1ba9ccafb&v=&b=&from_search=6
Global custom metrics names follow the “source.metricsName” format.
Blocksize is set in columnfamily‘s atrributes. It is used to control block sizes when generating blocks. But, it doesn’t take encoding into count. If you set encoding to blocks, after encoding, the block size varies. Since blocks will be cached in memory after encoding (default), it will cause memory fragment if using blockcache, or decrease the pool efficiency if using bucketCache. This issue introduced a new config named ‘hbase.writer.unified.encoded.blocksize.ratio’. The default value of this config is 1, meaning doing nothing. If this value is set to a smaller value like 0.5, and the blocksize is set to 64KB(default value of blocksize). It will unify the blocksize after encoding to 64KB * 0.5 = 32KB. Unified blocksize will releaf the memory problems mentioned above.
HBASE-14286 Correct typo in argument name for WALSplitter.writeRegionSequenceIdFile
Add table name to exception logging when a coprocessor is removed from a table by the region server
updated the lexicographic byte array comparator to use a slightly more optimized version similar to the one available in the guava library that compares only the first index where left[index] != right[index]. The comparator also returns the diff directly instead of mapping it to -1, 0, +1 range as was being done in the earlier version. We have seen significant performance gains, calculated in terms of throughput (ops/ms) with these changes ranging from approx 20% for smaller byte arrays upto 200 bytes and almost 100% for large byte array sizes that are in few KB's. We benchmarked with upto 16KB arrays and the general trend indicates that the performance improvement increases as the size of the byte array increases.
Non-idempotent operations (increment/append/checkAndPut/...) may throw OperationConflictException even though the increment/append succeeded. For example (client rpc retries number set to 3):
This patch will solve this problem by read the previous result when receive a duplicate rpc request.
Now raw scan can also read expired cells.
Adds a new Admin#listLocks, a panel on the procedures page to list procedure locks, and a list_locks command to the shell. Use it to see current state of procedure locking in Master process.
If users of the Thrift 1 Server enable proxy user support without enabling the prerequisite HTTP transport, we now log a WARN message about the mismatch.
StoreFile.createReader method is gone. Call initReader and then getReader instead.
Removes the Writables, and related code from WALEdit class. HBase-2.0 will not be able to read WAL files written with 0.94.x and before.
Updated refguide to tell users that IS annotation is only valid for IA.LimitedPrivate classes.
Now we do not have InterfaceStability annotations for IA,Public API. The stability of these classes will follow the rule of ‘Semantic Versioning’.
After HBASE-17215 we change to use two threads for (archived) hfile cleaning. The size throttling for large/small files could be set through “hbase.regionserver.thread.hfilecleaner.throttle” and default to 67108864 (64M). It supports online configuration change, just find the active master address through zookeeper dump and use it in update_config command, e.g. update_config ‘hbasem1.et2.tbsite.net,60100,1488038696741’
Upgrade internal pb to 3.2 from 3.1. 3.2 has fix for 64MB limit.
If filesystem is not available during log split, abort master server.
Reviving the merge of the compacting pipeline: making the limit on the number of the segments in the pipeline configurable and adding the merge test.
In order to customize the pipeline size limit change the value of the “hbase.hregion.compacting.pipeline.segments.limit” in the hbase-site.xml
Value 1 means to merge the segments on any flush-in-memory. Value higher than 16 means no merge.
HTableInterface was deprecated in 0.21.0 and is removed in 2.0.0. Use org.apache.hadoop.hbase.client.Table instead.
Now small scan and limited scan could also return partial results.
Add 2 constructors fot API Get
Now you can use ResultScanner.getScanMetrics to get the scan metrics at any time during the scan operation. The old Scan.getScanMetrics is deprecated and still work, but if you use ResultScanner.getScanMetrics to get the scan metrics and reset it, then the metrics published to the Scan instaince will be messed up.
Update our semver section to include a note on our allowing ourselves the right to add methods to an Interface over a minor version as agreed to up on the dev list: “If a Client implements an HBase Interface, a recompile MAY be required upgrading to a newer minor version (See release notes for warning about incompatible changes). All effort will be made to provide a default implementation so this case should not arise.”
In bin/hbase-config.sh, if value for HBASE_JMX_BASE is empty, keep current behavior. if HBASE_JMX_OPTS is not empty, keep current behavior. otherwise use the value of HBASE_JMX_BASE
Now async client has the same semantic with sync client for batch and partial. ''' Now setBatch doesn't mean setAllowPartialResult(true) If user setBatch(5) and rpc returns 3+5+5+5+3 cells, we should return 5+5+5+5+1 to user. '''
Also a minor API change: Result#createCompleteResult(List<Result>) is changed to Result#createCompleteResult(Iterable<Result>).
The executor for CoDel is changed to FastPathBalancedQueueRpcExecutor
Add a config named ‘hbase.hregion.unassign.for.fnfe’. It is used to control whether to reopen a region when hitting FileNotFoundException. The default value is true.
A new option -removeParents is now available that will remove an old parent when two valid daughters for that parent exist and -fixHdfsOverlaps is used. If there is an issue trying to remove the parent from META or sidelining the parent from HDFS we will fallback to do a regular merge. For now this option only works when the overlap group consists only of 3 regions (a parent, daughter A and daughter B)
Thrift2 proxy supports scan timeRange per column family
Fix our accidentally registering a RegionServer's ephermal znode BEFORE we checked in with the master.
In previous versions of HBase, the system intended to set a ZooKeeper ACL on all “sensitive” ZNodes for the user specified in the hbase.superuser configuration property. Unfortunately, the ACL was malformed which resulted in the hbase.superuser being unable to access the sensitive ZNodes that HBase creates. This JIRA issue fixes this bug. HBase will automatically correct the ACLs on start so users do not need to manually correct the ACLs.
HBASE-17716 breaks compatibility of ServerSideScanMetrics by changing public field names, and the issue is fixed through HBASE-17886
Now setBatch doesn't mean setAllowPartialResult(true) If user setBatch(5) and rpc returns 3+5+5+5+3 cells, we should return 5+5+5+5+1 to user. Scan#setBatch is helpful in paging queries, if you just want to prevent OOM at client, use setAllowPartialResults(true) is better. We deprecated isPartial and use mayHaveMoreCellsInRow. If it returns false, current Result must be the last one of this row.
Deletes BaseMasterAndRegionObserver, BaseMasterObserver, BaseRegionObserver, BaseRegionServerObserver and BaseWALObserver. Their corresponding interface classes now use JDK8's ‘default’ keyword to provide empty/no-op implementations so that:
Earlier, BaseXXXObserver classes provided these exact two benefits, but with ‘default’ keyword in JDK8, they are not needed anymore.
To fix the breakages because of this change, simply change “Foo extends BaseXXXObserver” to “Foo implements XXXObserver”.
WARNING: No release note provided for this change.
The table information page in the Master UI now includes a schema section that describes the column families defined for that table as well as any column family specific properties that are set.
Before this patch, later granted permissions will override previous granted permissions, and previous granted permissions LOST. this issue re-define grant semantic: for master branch, later granted permissions will merge with previous granted permissions. for branch-1.4, grant keep override behavior for compatibility purpose, and a grant with mergeExistingPermission flag provided.
Now you can include/exlude the startRow and stopRow for a scan. And the new methods to specify startRow and stopRow are withStartRow and withStopRow. The old methods to specify startRow and Row(include constructors) are marked as deprecated as in the old time if startRow and stopRow are equal then we will consider it as a get scan and include the stopRow implicitly. This is strange after we can set inclusiveness explicitly so we add new methods and depredate the old methods. The deprecated methods will be removed in the future.
Changes all tests to use the TestName JUnit Rule everywhere rather than hardcode table/region/store names.
The HBase cleaner chore process cleans up old WAL files and archived HFiles. Cleaner operation can affect query performance when running heavy workloads, so disable the cleaner during peak hours. The cleaner has the following HBase shell commands:
Following APIs are added in Admin:
The word ‘isPartial’ is ambiguous so we introduce a new method ‘mayHaveMoreCellsInRow’ to replace it. And the old meaning of ‘isPartial’ is not the same with ‘mayHaveMoreCellsInRow’ as for batched scan, if the number of returned cells equals to the batch, isPartial will be false. After this change the meaning of ‘isPartial’ will be same with ‘mayHaveMoreCellsInRow’. This is an incompatible change but it is not likely to break a lot of things as for batched scan the old ‘isPartial’ is just a redundant information, i.e, if the number of returned cells reaches the batch limit. You have already know the number of returned cells and the value of batch.
This patch adds support for specifying a WAL directory outside of the HBase root directory.
Multiple configuration variables were added to accomplish this: hbase.wal.dir: used to configure where the root WAL directory is located. Could be on a different FileSystem than the root directory. WAL directory can not be set to a subdirectory of the root directory. The default value of this is the root directory if unset.
hbase.rootdir.perms: Configures FileSystem permissions to set on the root directory. This is ‘700’ by default.
hbase.wal.dir.perms: Configures FileSystem permissions to set on the WAL directory FileSystem. This is ‘700’ by default.
A few bug fixes and tweaks to the fsgroup feature.
Renamed shell command move_rsgroup_servers as move_servers_rsgroup Renamed shell comand move_rsgroup_tables as move_tables_rsgroup
Made the ‘default’ group more ‘dynamic’; i.e. dead servers no longer show in the ‘default’ group.
In prior versions, the HBase Thrift handlers failed to increment per-method metrics when an exception was encountered. These metrics will now always be incremented, whether an exception is encountered or not. This change also adds exception-type metrics, similar to those exposed in regionservers, for individual exceptions which are received by the Thrift handlers.
Now the scan.setSmall method is deprecated. Consider using scan.setLimit and scan.setReadType in the future. And we will open scanner lazily when you call scanner.next. This is an incompatible change which delays the table existence check and permission check.
Mob compaction partition policy can be set by hbase> create ‘t1’, {NAME => ‘f1’, IS_MOB => true, MOB_THRESHOLD => 1000000, MOB_COMPACT_PARTITION_POLICY => ‘weekly’}
or
hbase> alter ‘t1’, {NAME => ‘f1’, IS_MOB => true, MOB_THRESHOLD => 1000000, MOB_COMPACT_PARTITION_POLICY => ‘monthly’}
Available MOB_COMPACT_PARTITION_POLICY options are “daily”, “weekly” and “monthly”, the default is “daily”.
When it is “weekly” policy, the mob compaction will try to compact files within one calendar week into one for a specific partition, similar for “daily” and “monthly”.
With “weekly” policy, one mob file normally is compacted twice during its lifetime (that is first on daily basis and then all such daily based compacted files belonging to a week at the weekly interval), for one region, there normally are 52 files for one year. With “Monthly” policy, one mob file normally is compacted 3 times during its lifetime (First daily and then weekly followed by monthly at end of every month) and normally there are 12 files for one year.
The -f argument is no longer required specifying target file; just pass the file as an argument.
In MOB-enabled column family, the lock in the major compaction is removed. All the delete markers are retained in the major compaction, and a MOB reference tag is appended to each of the retained delete markers.
Upgrades Jetty to 9.x from 6.x (Jetty9 is in different namespace from Jetty6). Also updated Jersey to 2.x and Servlet to 3.x.
Fix inability at finding static content post push of parent issue moving us to jetty9.
This issue adds two new modules, hbase-metrics and hbase-metrics-api which define and implement the “new” metric system used internally within HBase. These two modules (and some other code in hbase-hadoop2-compat) module are referred as “HBase metrics framework” which is HBase-specific and independent of any other metrics library (including Hadoop metrics2 and dropwizards metrics).
HBase Metrics API (hbase-metrics-api) contains the interface that HBase exposes internally and to third party code (including coprocessors). It is a thin abstraction over the actual implementation for backwards compatibility guarantees. The metrics API in this hbase-metrics-api module is inspired by the Dropwizard metrics 3.1 API, however, the API is completely independent.
hbase-metrics module contains implementation of the “HBase Metrics API”, including MetricRegistry, Counter, Histogram, etc. These are highly concurrent implementations of the Metric interfaces. Metrics in HBase are grouped into different sets (like WAL, RPC, RegionServer, etc). Each group of metrics should be tracked via a MetricRegistry specific to that group.
Historically, HBase has been using Hadoop‘s Metrics2 framework [3] for collecting and reporting the metrics internally. However, due to the difficultly of dealing with the Metrics2 framework, HBase is moving away from Hadoop’s metrics implementation to its custom implementation. The move will happen incrementally, and during the time, both Hadoop Metrics2-based metrics and hbase-metrics module based classes will be in the source code. All new implementations for metrics SHOULD use the new API and framework.
This jira also introduces the metrics API to coprocessor implementations. Coprocessor writes can export custom metrics using the API and have those collected via metrics2 sinks, as well as exported via JMX in regionserver metrics.
More documentation available at: hbase-metrics-api/README.txt
After HBASE-17491 all setter methods in HTable are marked as deprecated, moved into TableBuilder, and will be removed later.
Make the framework more ‘lively’; undo ‘suspend’ notion in Procedure, rely on eventing mechanism instead. Lets us remove no longer needed synchronizations. Framework can now do more ops per second.
Assign sequenceid to an edit before we go on the ringbuffer; undoes contention on WALKey latch. Adds a new config “hbase.hregion.mvcc.preassign” which defaults to true: i.e. this speedup is enabled.
User could set this per-table level, like: create ‘table’,{NAME=>‘f1’,CONFIGURATION=>{‘hbase.hregion.mvcc.preassign’=>‘false’}}
prevent creating unused objects in the WALEdit‘s construction. +If the cp#preBatchMutate returns true, the WALEdit is useless. So we should create the WALEdit after step 2. +The cells came from cp should be counted because they are added into the WALEdit . The use case is the local index of phoenix +If the mutation contains the SKIP_WAL property, its cells aren’t added into the WALEdit. So these cells shouldn't be counted.
Purges code that did zk-hosted locks for table ops (we do procedure-based locks now)
Add checking ACL when taking locks.
Move locking to be procedure (Pv2) rather than zookeeper based. All locking moved over to new infrastructure including MOBing locking.
In 1.x branches, Admin.mergeRegions calls MASTER via dispatchMergingRegions RPC; when executing dispatchMergingRegions RPC, MASTER calls RS via MergeRegions to complete the merge in RS-side.
With HBASE-16119, the merge logic moves to master-side. This JIRA cleans up unused RPCs (dispatchMergingRegions and MergeRegions) , removes dangerous tools such as Merge and HMerge, and deletes unused RegionServer-side merge region logic in 2.0 release.
Lock for HBase Entity either a Table, a Namespace, or Regions.
These are remote locks which live on master, and need periodic heartbeats to keep them alive. (Once we request the lock, internally an heartbeat thread will be started). If master doesn‘t receive the heartbeat in time, it’ll release the lock and make it available to other users.
Use {@link LockServiceClient} to build instances. Then call {@link #requestLock()}. {@link #requestLock} will contact master to queue the lock and start the heartbeat thread which will check lock's status periodically and once the lock is acquired, it will send the heartbeats to the master.
Use {@link #await} or {@link #await(long, TimeUnit)} to wait for the lock to be acquired. Always call {@link #unlock()} irrespective of whether lock was acquired or not. If the lock was acquired, it'll be released. If it was not acquired, it is possible that master grants the lock in future and the heartbeat thread keeps it alive forever by sending heartbeats. Calling {@link #unlock()} will stop the heartbeat thread and cancel the lock queued on master.
There are 4 ways in which these remote locks may be released/can be lost: * Call {@link #unlock}. * Lock times out on master: Can happen because of network issues, GC pauses, etc. Worker thread will call the given abortable as soon as it detects such a situation. Fail to contact master: If worker thread can not contact mater and thus fails to send heartbeat before the timeout expires, it assumes that lock is lost and calls the * abortable. Worker thread is interrupted.
Use example:
EntityLock lock = lockServiceClient.*Lock(...., “exampled lock”, abortable); lock.requestLock(); .... ....can do other initializations here since lock is ‘asynchronous’... .... if (lock.await(timeout)) { ....logic requiring mutual exclusion } lock.unlock();
After HBASE-14061 we support to set storage policy for HFile through “hbase.hstore.block.storage.policy” configuration, and we support CF-level setting to override the settings from configuration file. Currently supported storage policies include ALL_SSD/ONE_SSD/HOT/WARM/COLD, refer to http://hadoop.apache.org/docs/r2.6.0/hadoop-project-dist/hadoop-hdfs/ArchivalStorage.html for more details
For example, to create a table with two families: “cf1” with “ALL_SSD” storage policy and “cf2” with “ONE_SSD”, we could use below command in hbase shell: create ‘table’,{NAME=>‘f1’,STORAGE_POLICY=>‘ALL_SSD’},{NAME=>‘f2’,STORAGE_POLICY=>‘ONE_SSD’}
We could also set the configuration in table attribute like all other configurations: create ‘table’,{NAME=>‘f1’,CONFIGURATION=>{‘hbase.hstore.block.storage.policy’=>‘ONE_SSD’}}
List replication peers request will be roughed through master.
After HBASE-15172/HBASE-19016 we could set storage policy through “hbase.hstore.block.storage.policy” property for bulkload, or “hbase.hstore.block.storage.policy.<family_name>” for a specified family. Supported storage policy includes: ALL_SSD, ONE_SSD, HOT, WARM, COLD, etc.
Get/update replication peer config requests will be routed through master.
Now you can specific the inclusive of startRow and stopRow for a scan using the new methods withStartRow(byte[] startRow, boolean inclusive) and withStopRow(byte[] stopRow, boolean inclusive). The old setStartRow and setStopRow methods, and the constructors are marked as deprecated because of an strange behavior that we will include the stopRow implicitly if startRow equals to stopRow. This is used to support get scan in the old time. Use withStartRow and withStopRow instead.
For developers, the ConnectionUtils.createClosestRowBefore is also marked as deprecated as the row returned by this method is only very very close to the current row, not closest. Avoid using this method in the future.
Add a conf “replication.total.buffer.quota” to limit total size of buffered entries in all replication peers. It will prevent server getting OOM if there are many peers. Default value is 256MB.
Enable/Disable replication peer requests will be routed through master.
Changes how many tasks PE runs when clients are mapreduce. Now tasks == client count. Previous we hardcoded ten tasks per client instance.
Add/Remove replication peer requests will be routed through master. And make ReplicationAdmin as Deprecated.
Now one can start hbase services with enabled “autostart/autorestart” feature in controlled fashion with the help of “--autostart-window-size” to define the window period and the “--autostart-window-retry-limit” to define the number of times the hbase services have to be restarted upon being killed/terminated abnormally within the provided window perioid.
The following cases are supported with “autostart/autorestart”:
a) --autostart-window-size=0 and --autostart-window-retry-limit=0, indicates infinite window size and no retry limit b) not providing the args, will default to a) c) --autostart-window-size=0 and --autostart-window-retry-limit=<positive value> indicates the autostart process to bail out if the retry limit exceeds irrespective of window period d) --autostart-window-size=<x> and --autostart-window-retry-limit=<y> indicates the autostart process to bail out if the retry limit “y” is exceeded for the last window period “x”.
For each read(), old ThrottledInputStream sleeps/wakes/checks for many times for controlling the throughput. After this patch, ThrottledInputStream sleeps/wakes/checks only once. So we can reduce CPU usage.
Provide per peer throttling for replication. Add the bandwidth upper limit to ReplicationPeerConfig and a new shell cmd set_peer_bandwidth to update the bandwidth in need.
Specify the name of an alternate BufferedMutator implementation by either:
* Setting “hbase.client.bufferedmutator.classname” to the name of the alternate implementation class in Configuration * Or, by setting BufferedMutatorParams#implementationClassName and passing the amended BufferedMutatorParams when calling Connection#getBufferedMutator.
This patch provides a single external knob to control memstore compaction. It also inmemory compaction with BASIC policy as our default (AFTERWORD: inmemory compaction as default was undone in HBASE-17333 because of test failures; will be reenabled in later, dedicated issue)
Possible memstore compaction policies are: (1) None - no memory compaction, when size threshold is exceeded data is flushed to disk (2) Basic policy applies optimizations which modify the index to a more compacted representation. This is beneficial in all access patterns. The smaller the cells are the greater the benefit of this policy. This is the default policy. (3) Eager - in addition to compacting the index representation as the basic policy, eager policy eliminates duplication while the data is still in memory (much like the on-disk compaction does after the data is flushed to disk). This policy is most useful for applications with high data churn or small working sets.
Memory compaction policeman be set at the column family level at table creation time: {code} create ‘<tablename>’, {NAME => ‘<cfname>’, IN_MEMORY_COMPACTION => ‘<NONE|BASIC|EAGER>’} {code} or as a property at the global configuration level by setting the property in hbase-site.xml, with BASIC being the default value: {code} <property> <name>hbase.hregion.compacting.memstore.type</name> <value><NONE|BASIC|EAGER></value> </property> {code} The values used in this property can change as memstore compaction policies evolve over time.
Add a ReplicationZKNodeCleaner periodically check and delete the useless replication queue zk node belong to the peer which is not exist.
Adds section at http://hbase.apache.org/book.html#standalone.over.hdfs on how to make standalone persist to an hdfs instance (where standalone is all daemons in the one jvm).
Provides ability to restrict table coprocessors based on HDFS path whitelist. (Particularly useful for allowing Phoenix coprocessors but not arbitrary user created coprocessors.)
Provide an interface RpcCall on the server side. RpcServer.Call now is marked as @InterfaceAudience.Private, and implements the interface RpcCall,
The merge region logic is controlled by master in 2.0.0 (in 1.x, the core merge region logic is in the region server side). The coprocessors related to merge region in RS-side would be no-op in 2.0.0 and later release. Therefore, this is an incompatible change. Users needs to move the CP logic to new master CP and registers them.
A new mergeRegionsAsync() API is added in client. The existing mergeRegions() API will call the new API so client does not have to change its code.
Before this issue, two concurrent Increments/Appends done in same millisecond or RS's clock going back will result in two results have same TS, which is not friendly to versioning and will get wrong result in slave cluster if the replication is disordered. After this issue, the result of Increment/Append will always have an incremental TS. There is no any inconsistent in replication for these operations. But there is a rare case that if there is a Delete in same millisecond, the later result can not be masked by this Delete. This can be fixed after we have new semantics that previous Delete will never mask later Put even its timestamp is higher.
Add TThreadedSelectorServer support for HBase Thrift2
Add region balance throttling. Master execute every region balance plan per balance interval, which is equals to divide max balancing time by the size of region balance plan. And Introduce a new config hbase.master.balancer.maxRitPercent to protect availability. If config this to 0.01, then the max percent of regions in transition is 1% when balancing. Then the cluster's availability is at least 99% when balancing.
Added a new config hbase.regionserver.offheap.global.memstore.size using which one can specify the global off heap limit that all memstores can use. When this config is in MSLAB should be turned ON and we will use the entire size for the MSLAB pool. It will make off heap chunks and pool then. It will behave as if we are working with off heap memstores. When this config is having a valid value and MSLAB is turned OFF, the system will just ignore the offheap size and continue to use global max heap space % for memstores and work with on heap memstores.
Remove HLogKey and related classes and methods. Remove SequenceFile based log reader and writer. WALObserver and RegionObserver are changed so this is an incompatible change.
Added couple of API's to Admin.java:
Returns region load map of all regions hosted on a region server Map<byte[], RegionLoad> getRegionLoad(ServerName sn) throws IOException;
Returns region load map of all regions of a table hosted on a region server Map<byte[], RegionLoad> getRegionLoad(ServerName sn, TableName tableName) throws IOException
Added an API to region server:
public GetRegionLoadResponse getRegionLoad(RpcController controller, GetRegionLoadRequest request) throws ServiceException;
Primary intention is to use this API for RegionSizeCalculator and not rely on Master for ClusterStatus. On large clusters, ClusterStatus() can take a long time. IfMaster is down/busy, then some of the jobs timeout/fail. Other possible uses:
Using the ByteBuffers from ByteBufferPool to read the request bytes at server. When the size of the request is smaller than 1/6th size of a BB in the pool, we will not use that but read into an on demand created, proper sized on heap ByteBuffer.
Adds a 1.1. item to our ‘Documentation and API’ tab. Gives access to 1.1 APIs, XRef, etc.
The following RegionObserver methods are deprecated
InternalScanner preFlushScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, final KeyValueScanner memstoreScanner, final InternalScanner s) throws IOException;
InternalScanner preCompactScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, List<? extends KeyValueScanner> scanners, final ScanType scanType, final long earliestPutTs, final InternalScanner s, CompactionRequest request)
Instead, use the following methods:
InternalScanner preFlushScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, final KeyValueScanner memstoreScanner, final InternalScanner s, final long readPoint) throws IOException;
InternalScanner preCompactScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, List<? extends KeyValueScanner> scanners, final ScanType scanType, final long earliestPutTs, final InternalScanner s, final CompactionRequest request, final long readPoint) throws IOException
Removes per-region level (get size, get time, scan size and scan time histogram) metrics that was exposed before. Per-region histogram metrics with 1000+ regions causes millions of objects to be allocated on heap. The patch introduces getCount and scanCount as counters rather than histograms. Other per-region level metrics are kept as they are.
Test that environment no longer has to have protoc (2.5 and 3.1) available. Needed small adjustment in yetus protoc build but otherwise all works.
Upgrade Apache CLI version from 1.2 to 1.3.1.
These are few good/important changes included in this update:
For full list of changes: https://commons.apache.org/proper/commons-cli/changes-report.html#a1.3
MSLAB chunk pool is on by default in hbase-2.0.0.
WARNING: No release note provided for this change.
Delimit START, STOP, and ABORT messages with ‘*****’ so denote.
Introduces a new split policy: SteppingSplitPolicy This will use a simple step function to split a region at (by default) 2 xflushSize when no other region of the same table is seen on the region server, or max-file-size when one or more other regions of the same table is seen.
In HBase 2.0 this is going to be the default. In previous versions it can be configured.
The index-compation and data-compaction variants of CompactingMemStore are introduced. In both types the active (mutable) segment is periodically flushed-in-memory and is added as immutable segment in the compaction pipeline. The CompactingMemStore of index-compaction type is merging all immutable segments of the compacting pipeline into one. The merging of N segments is explained below. The CompactingMemStore of data-compaction type is compacting all immutable segments of the compacting pipeline into one. After the merge/compaction the old segments in the compacting pipeline are replaced with one new.
Before explaining the process of merging N old segments into new one, note that segment structure includes ordered index that allows traversing the cells data efficiently. The merge is copying the ordered indexes of the old segments into one ordered index of new segment. No data is copied, no cells are filtered. Alternatively, in the process of compacting N old segments into new one, both data and index are copied. The old cells are filtered, meaning upon compaction unused versions of the cells are not copied so the new segment has less data then all old ones.
This issue introduces only the merging ability and simplifies the user intervention for switching between types. The previous CompactingMemStore structure was added by HBASE-16420 and HBASE-16421. The future refinements of the policy or merging/compacting will come in HBASE-16417.
In order to create a table with CompactingMemStore as a MemStore one should use: create ‘<tablename>’, {NAME => ‘<cfname>’, IN_MEMORY_COMPACTION => true} IN_MEMORY_COMPACTION default is false, so table created as following will have the known DefaultMemStore as a MemStore. create ‘<tablename>’, {NAME => ‘<cfname>’}
The default type of CompactingMemStore is index-compaction. In order to change it to data-compaction one should add to the hbase-site.xml <property> <name>hbase.hregion.compacting.memstore.type</name> <value>data-compaction</value> </property>
in addition to creating the table as following create ‘<tablename>’, {NAME => ‘<cfname>’, IN_MEMORY_COMPACTION => true}
Marking it as incompatible change as there is a change in behavior for region flush decision. The default flush size of 128 MB per region was tracked against both actual data bytes size + overhead of these cells in memstore memory (Overhead because of Cell java objects and CSLM entry). As part of this jira we will keep track of cell data size only in region level. So 128 MB flush size means, 128 MB of cell data bytes (key+ value+..)
Globally we will track cell data size and heap overhead separately and will consider both for forced flushes. We will not allow over consume of heap memory by all memstore. This is as old case. Only tracking way is changed.
Upgrade os-maven-plugin mvn extension which figures the os we are running on from 1.4 to 1.5.
Simplifies .proto manipulations. One step only now -- no need to keep pom.xml listing up to date with the protobuf protos directory content -- and no need to preinstall protoc; mvn does it all for you now.
Moved the Split Region logic to Master and most of split region coprocessor is in master now. Need to change dependency such as Phoenix.
This issue adds a patch to our checked in internal, shaded protobuf, but it also adds a general means of apply patches to our version of protobuf. Patches found in the new src/main/patches directory are all applied as the last task when you run a build with the -Pcompile-protobuf profile under the hbase-protocol-shaded module. This commit also includes our first patch to protobuf; it adds ByteInput to mimic pb3.1's ByteOutput (src/main/patches/HBASE-15789_V2.patch attached here).
Fixed potential infinite recursion in AssignmentManager.checkWals().
Improve transparent table/CF encryption with Commons Crypto. The change introduces a new optional CryptoCipherProvider (CommonsCryptoAES) for transparent table/CF encryption. And the encryption performance would be accelerated by hardware in modern CPU (AES-NI). This feature could be enabled by updating the configuration “hbase.crypto.cipherprovider” to “org.apache.hadoop.hbase.io.crypto.CryptoCipherProvider” in hbase-site.xml. For detailed information about transparent table/CF encryption including configuration examples see the Security section of the HBase manual.
With the security RPC and encryption enabled, introduce Apache Commons Crypto to do the encryption/decryption which supports both supports both JCE Cipher and OpenSSL Cipher. Adds new configs “hbase.rpc.crypto.encryption.aes.enabled” which defaults to false, and “hbase.rpc.crypto.encryption.aes.cipher.class” which defaults to “org.apache.commons.crypto.cipher.JceCipher” to support JCE Cipher, it also can be set as “org.apache.hadoop.crypto.OpensslCipher” to support Openssl Cipher.
Fixed a bug in sequenceId tracking for the WALs that caused WAL files to accumulate without being deleted due to a rare race condition.
Add createAsyncConnection method to ConnectionFactory for creating AsyncConnection. The default implementation is org.apache.hadoop.hbase.client.AsyncConnectionImpl. You can use ‘hbase.client.async.connection.impl’ to plug in your own AsyncConnection implementation.
Empty filter list will behave as when there is no filter added. This change is a behavioral change for those who rely on Empty filter list.
Below APIs from CP exposed Store interface are removed upsert(Iterable<Cell> cells, long readpoint) add(Cell cell) add(Iterable<Cell> cells) replayCompactionMarker(CompactionDescriptor compaction, boolean pickCompactionFiles, boolean removeFiles) assertBulkLoadHFileOk(Path srcPath) bulkLoadHFile(String srcPathStr, long sequenceId) bulkLoadHFile(StoreFileInfo fileInfo)
Add AsyncConnection, AsyncTable and AsyncTableRegionLocator. Now the AsyncTable only support get, put and delete. And the implementation of AsyncTableRegionLocator is synchronous actually.
This issue fix three bugs:
This adds a new per-region metric named “lastMajorCompactionAge” for tracking time since the last major compaction ran on a given region. If a major compaction has never run, the age will be equal to the current timestamp.
(This change will be irrelevant after HBASE-16774 lands). There is a subtle change with error handling when a connection is not able to connect to ZK. Attempts to create a connection when ZK is not up will now fail immediately instead of silently creating and then failing on a subsequent HBaseAdmin call.
In some particular deployments, the Replication code believes it has reached EOF for a WAL prior to successfully parsing all bytes known to exist in a cleanly closed file.
If an EOF is detected due to parsing or other errors while there are still unparsed bytes before the end-of-file trailer, we now reset the WAL to the very beginning and attempt a clean read-through. Because we will retry these failures indefinitely, two additional changes are made to help with diagnostics:
* On each retry attempt, a log message like the below will be emitted at the WARN level:
Processing end of WAL file '{}'. At position {}, which is too far away
from reported file length {}. Restarting WAL reading (see HBASE-15983
for details).
* additional metrics measure the use of this recovery mechanism. they are described in the reference guide.
Updates the comments and default values in a few scripts and docs to reflect our Java 1.8+ requirement.
Core is now up on protobuf 3.1.0 (Coprocessor Endpoints and REST are still on protobuf 2.5.0).
Shade/relocate and include the protobuf we use internally. See protobuf chapter in the refguide for more on how we protobuf in hbase-.2.0.0 and going forward.
See https://docs.google.com/document/d/1H4NgLXQ9Y9KejwobddCqaVMEDCGbyDcXtdF5iAfDIEk/edit# for how we arrived at this approach.
See http://mail-archives.apache.org/mod_mbox/hbase-dev/201610.mbox/%3C07850EDD-7230-431B-9AB0-C5C91B105EEC%40gmail.com%3E for discussion around merging this change and of how we might revert if an alternative to this awkward patch presents itself; e.g. an hadoop with CLASSPATH isolation (and means of dealing with Sparks use of protobuf 2.5.0, etc.)
Shade/relocate the protobuf hbase uses internally. All core now refers to new module added in this patch, hbase-protocol-shaded. Coprocessor Endpoints carry-on with references to the original hbase-protocol module. See new chapter in book on protobufs on how-to going forward.
This issue adds a config, always.copy.files, to LoadIncrementalHFiles. When set to true, source hfiles would be copied. Meaning source hfiles would be kept after bulk load is done. Default value is false.
“Please do not use DateTieredCompaction with Major Compaction unless you have a version with this. Otherwise your cluster will not compact any store files and you can end up running out of file descriptors.” @churro morales
The HBase property ‘hbase.bulkload.staging.dir’ is deprecated and is ignored from HBase 2.0. It will defaults to hbase.rootdir/staging automatically with the correct permissions.
Changed tracking of evictedBlocks count NOT to include evictions of blocks for a removed HFile. HFiles gets removed after compaction
Fixed warning error message displayed for region directory not found for non-default/ non-primary replicas in hbck
Scan#setStartRow() and Scan#setStopRow() now validate the argument passed for each row key. If the length of the byte[] passed exceeds Short.MAX_VALUE, an IllegalArgumentException will be thrown.
org.apache.hadoop.hbase.util.Counter is deprecated now and will be removed in 3.0. Use LongAdder instead.
Support replication by namespaces config in peer.
Deprecate the configuration property ‘hbase.zookeeper.useMulti’. useMulti will always be enabled. ZooKeeper 3.4.x and newer is required.
Internal:
The ZKUtil#multiOrSequential(ZooKeeperWatcher zkw, List<ZKUtilOp> ops, boolean runSequentialOnMultiFailure) will not check ‘hbase.zookeeper.useMulti’ anymore, and will always use multi. It can still fall back to sequential operations if:
RunSequentialOnMultiFailure is true On calling multi, we get a ZooKeeper exception that can be handled by a sequential call.
Add a new configuration, hbase.client.perserver.requests.threshold, to limit the max number of concurrent request to one region server. If the user still create new request after reaching the limit, client will throw ServerTooBusyException and do not send the request to the server. This is a client side feature and can prevent client's threads being blocked by one slow region server resulting in the availability of client is much lower than the availability of region servers.
For completeness, here extract on new config from hbase-default.xml:
Property: hbase.client.perserver.requests.threshold Default: 2147483647 Description: The max number of concurrent pending requests for one server in all client threads (process level). Exceeding requests will be thrown ServerTooBusyException immediately to prevent user's threads being occupied and blocked by only one slow region server. If you use a fix number of threads to access HBase in a synchronous way, set this to a suitable value which is related to the number of threads will help you. See https://issues.apache.org/jira/browse/HBASE-16388 for details.
The security admin instance available within the HBase shell now returns “false” from the namespace_exists? method for non-existent namespaces rather than raising a wrapped NamespaceNotFoundException.
As a side effect, when the “grant” and “revoke” commands in the HBase shell are invoked with a non-existent namespace the resulting error message now properly refers to said namespace rather than to the user.
push to branch-1.3+
HBase no longer includes Xerces implementation jars that were previously included via transitive dependencies. Downstream users relying on HBase for these artifacts will need to update their dependencies.
HBASE-16213 introduced a new DataBlockEncoding in name of ROW_INDEX_V1, which could improve random read (get) performance especially when the average record size (key-value size per row) is small. To use this feature, please set DATA_BLOCK_ENCODING to ROW_INDEX_V1 for CF of newly created table, or change existing CF with below command: alter ‘table_name’,{NAME => ‘cf’, DATA_BLOCK_ENCODING => ‘ROW_INDEX_V1’}.
Please note that if we turn this DBE on, HFile block will be bigger than NONE encoding because it adds some meta infos for binary search: /** * Store cells following every row‘s start offset, so we can binary search to a row’s cells. * * Format: * flat cells * integer: number of rows * integer: row0‘s offset * integer: row1’s offset * .... * integer: dataSize * */
Seek in row when random reading is one of the main consumers of CPU. This helps. See slide #7 here https://www.slideshare.net/HBaseCon/lift-the-ceiling-of-hbase-throughputs?qid=597ee2fa-8125-4faa-bb3b-2bf1ba9ccafb&v=&b=&from_search=6
--delimiter= option is added to verifyrep. The delimiter would wrap bad rows in log output.
A long, working issue that discussed Segment formats introducing CellArrayMap (delivered as the patch attached to this issue) and CellChunkMap (to be delivered later in HBASE-16421 but see patch v02 for an embryonic form named CellBlockSerialized); when to copy Segment data (and when not too); and then what to include at flush time (the suffix Segment or all Segments). Designs that evolved as discussion went on are attached. Outstanding issues turned up here, not including a CellChunkMap implementation, are listed below but are to be addressed in follow-ons (See HBASE-16417):
New tool to dump existing replication peers, configurations and queues when using HBase Replication. The tool provides two flags:
--distributed This flag will poll each RS for information about the replication queues being processed on this RS. By default this is not enabled and the information about the replication queues and configuration will be obtained from ZooKeeper. --hdfs When --distributed is used, this flag will attempt to calculate the total size of the WAL files used by the replication queues. Since its possible that multiple peers can be configured this value can be overestimated.
Adds below change to our compat guarantees:
{code} -* Example: A user using a newly deprecated api does not need to modify application code with hbase api calls until the next major version. 10 +* New APIs introduced in a patch version will only be added in a source compatible way footnote:[See ‘Source Compatibility’ https://blogs.oracle.com/darcy/entry/kinds_of_compatibility]: i.e. code that implements public APIs will continue to compile. {code}
RemoteHTable now supports binary row keys with any character or byte by properly encoding request URLs. This is a both a behavioral change from earlier versions and an important fix for protocol correctness.
Downstream users wishing to test HBase in a “distributed” fashion (multiple “nodes” running as separate containers on the same host) can now do so in an automated fashion while leveraging Docker for process isolation via the clusterdock project.
For details see the README.md in the dev-support/apache_hbase_topology folder.
This issue upgrades httpclient to 4.5.2 and httpcore to 4.4.4 which are the versions used by hadoop-2. This is to handle the following CVE's.
https://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2015-5262 : http/conn/ssl/SSLConnectionSocketFactory.java in Apache HttpComponents HttpClient before 4.3.6 ignores the http.socket.timeout configuration setting during an SSL handshake, which allows remote attackers to cause a denial of service (HTTPS call hang) via unspecified vectors.
https://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2012-6153 https://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2012-5783 Apache Commons HttpClient 3.x, as used in Amazon Flexible Payments Service (FPS) merchant Java SDK and other products, does not verify that the server hostname matches a domain name in the subject's Common Name (CN) or subjectAltName field of the X.509 certificate, which allows man-in-the-middle attackers to spoof SSL servers via an arbitrary valid certificate.
Downstream users who are exposed to commons-httpclient via the HBase classpath will have to similarly update their dependency.
Undo protobuf references through the codebase so protobuf references are contained rather than spread about the codebase. For example, moved protobuff-ing up into the various Callables rather than repeat on each method invocation cleaning up boilerplate around rpc calls. Having a few protobuf reference locations only simplifies the parent issue shading project.
HBase now ensures the jsr305 implementation from the findbugs project is not included in its binary artifacts or the compile / runtime dependencies of its user facing modules. Downstream users that rely on this jar will need to update their dependencies.
The MapReduce helper function `TableMapReduce.addDependencyJars(Configuration, class<?> ...)` has been deprecated since it is easy to use incorrectly. Most users should rely on addDependencyJars(Job) instead.
In order to avoid blockcache size exceed acceptable size too much, we add one configuration “hbase.lru.blockcache.hard.capacity.limit.factor” to decide whether the block could be put into LruBlockCache or not. This factor defaults to 1.2 If blockcache size >= factor*acceptableSize, we will reject the block into cache.
The HBase client artifact previously incorrectly included the hbase-common test jar as a runtime dependency. With this change, that dependency has been moved to test scope. Downstream users are not expected to be impacted, unless they relied on the transitive dependency for these HBase internal test classes.
This issue reverts fixes designed to prevent malicious content from rendering in HBase's UIs. Specifically, these changes shipped in 1.1.4+ and 1.2.0+. They were removed due to licensing issues discovered in the dependencies they introduced. Their implementation and those dependencies have been removed from HBase! Removal of these dependencies is against the strict definition of our version compatibility guidelines. However, inclusion of non-Apache approved licenses cannot be tolerated. Implementation of these fixes using an Apache-appropriate means is tracked in HBASE-16328.
A new hfile configuration “hfile.index.block.min.entries” which defaults to 16 determines how many entries the hfile index block can have at least. The configuration which determines how large the index block can be at max (hfile.index.block.max.size) is ignored as long as we have fewer than hfile.index.block.min.entries entries. This ensures that multi-level index does not build up with too many levels.
The AssignmentManager MBean was named AssignmentManger (note misspelling). This patch fixed the misspelling.
Adds logging of region and server. Helpful debugging. Logging now looks like this: {code} 2016-06-23 17:07:18,759 INFO [Thread-1] client.AsyncProcess$AsyncRequestFutureImpl(1601): #1, waiting for 1 actions to finish on table: DUMMY_TABLE 2016-06-23 17:07:18,759 INFO [Thread-1] client.AsyncProcess(1720): Left over 1 task(s) are processed on server(s): [s1:1,1,1] 2016-06-23 17:07:18,759 INFO [Thread-1] client.AsyncProcess(1728): Regions against which left over task(s) are processed: [DUMMY_TABLE,DUMMY_BYTES_1,1.3fd12ea80b4df621fb15497ba75f7368.,DUMMY_TABLE,DUMMY_BYTES_2,2.924207e242e313d2e5491c625e0a296e.] {code}
A memory metrics reveals situations happened in both MemStores and BlockCache in RegionServer. Through this metrics, users/operators can know 1). Current size of MemStores and BlockCache in bytes. 2). Occurrence for Memstore minor and major flush. (named unblocked flush and blocked flush respectively, shown in histogram) 3). Dynamic changes in size between MemStores and BlockCache. (with Increase/Decrease as prefix, shown in histogram). And a counter for no changes, named DoNothingCounter. 4). Occurrence for memory usage alarm (used more than 95% by default) in RegionServer. (named AboveHeapOccupancyLowWatermarkCounter)
SecureBulkLoadEndpoint has been integrated into HBase core as default bulk load mechanism. It is no longer needed to install it as a coprocessor endpoint. The new server is backward compatible, accommodating non-secure old client and secure old client requesting SecureBulkLoadEndpoint service. SecureBulkLoadEndpoint is deprecated. The backward compatibility support may be removed in future releases.
When LocalHBaseCluster is started from the command line the Master would give up after 30 seconds due to a hardcoded timeout meant for unit tests. This change allows the timeout to be configured via hbase-site as well as sets it to 5 minutes when LocalHBaseCluster is started from the command line.
HBASE-16052 improves the performance and scalability of HBaseFsck, especially for large clusters with a small number of large tables.
Searching for lingering reference files is now a multi-threaded operation. Loading HDFS region directory information is now multi-threaded at the region-level instead of the table-level to maximize concurrency. A performance bug in HBaseFsck that resulted in redundant I/O and RPCs was fixed by introducing a FileStatusFilter that filters FileStatus objects directly.
If zk based replication queue is used and useMulti is false, we will schedule a chore to clean up the orphan replication queue lock on zk.
MultiHFileOutputFormat support output of HFiles from multiple tables. It will output directories and hfiles as follow, --table1 --family1 --family2 --Hfiles --table2 --family3 --hfiles --family4
family directory and its hfiles match the output of HFileOutputFormat2
Prior to this change, the integration test clients (IntegrationTest*) relied on the Kerberos credential cache for authentication against secured clusters. This could lead to the tests failing due to authentication failures when the tickets in the credential cache expired. With this change, the integration test clients will make use of the configuration properties for “hbase.client.keytab.file” and “hbase.client.kerberos.principal”, when available. This will perform a login from the configured keytab file and automatically refresh the credentials in the background for the process lifetime.
For cluster upgraded from 1.0.x or older releases, master startup would not continue the in-progress enable/disable table process. If orphaned znode with ENABLING/DISABLING state exists in the cluster, run hbck or manually fix the issue.
For new cluster or cluster upgraded from 1.1.x and newer release, there is no issue to worry about.
Adds a PRIORITY property to the HTableDescriptor. PRIORITY should be in the same range as the RpcScheduler defines it (HConstants.XXX_QOS).
Table priorities are only used for region opening for now. There can be other uses later (like RpcScheduling).
Regions of high priority tables (priority >= than HIGH_QOS) are opened from a different thread pool than the regular region open thread pool. However, table priorities are not used as a global order for region assigning or opening.
When a replication endpoint is sent a shutdown request by the replication source in situations like removing a peer, we now try to gracefully shut it down by draining the items already sent for replication to the peer cluster. If the drain does not complete in the specified time (hbase.rpc.timeout * replication.source.maxterminationmultiplier), the regionserver is aborted to avoid blocking the WAL roll.
Masters will no longer start any replication threads if they are hosting only system tables.
In order to change this add something to the config for tables on master that doesn‘t start with “hbase:” ( Replicating system tables is something that’s currently unsupported and can open up security holes, so do this at your own peril)
Allow a directory containing the jars or some wildcards to be specified, such as: hdfs://namenode:port/user/hadoop-user/ or hdfs://namenode:port/user/hadoop-user/*.jar
Please note that if a directory is specified, all jar files(.jar) directly in the directory are added, but it does not search files in the subtree rooted in the directory. Do not contain any wildcard if you would like to specify a directory.
Downstream users of HBase dependencies that do not properly activate Maven profiles should now see a correct transitive dependency on the default hadoop-compatibility-module.
The dependency owasp.esapi had a compatible change from 2.1.0 to 2.1.0.1. As a result, the transitive dependency commons-fileupload had a change from 1.2 to 1.3.1, which has some minor class changes that impact binary compatibility. Interested users should check the release notes of commons-fileupload to see if any of the incompatible changes impact them.
http://commons.apache.org/proper/commons-fileupload/changes-report.html
compaction_state shell command would return compaction state in String form: NONE, MINOR, MAJOR, MAJOR_AND_MINOR
Adds new hbase-shaded-client archetype; also corrects an omission found in hbase-archetypes/README.md in the section headed “How to add a new archetype”.
This patch introduces a new infrastructure for creation and maintenance of Maven archetypes in the context of the hbase project, and it also introduces the first archetype, which end-users may utilize to generate a simple hbase-client dependent project.
NOTE that this patch should introduce two new WARNINGs (“Using platform encoding ... to copy filtered resources”) into the hbase install process. These warnings are hard-wired into the maven-archetype-plugin:create-from-project goal. See hbase/hbase-archetypes/README.md, footnote [6] for details.
After applying the patch, see hbase/hbase-archetypes/README.md for details regarding the new archetype infrastructure introduced by this patch. (The README text is also conveniently positioned at the top of the patch itself.)
The README.md file also contains several paragraphs under the heading, “Notes for contributors and committers to the HBase project”, which explains the layout of ‘hbase-archetypes’, and how archetypes are created and installed into the local Maven repository, ready for deployment to the central Maven repository. It also outlines how new archetypes may be developed and added to the collection in the future.
Done. Thanks, Dima, Andrew!
HBase Web UIs can be secured from general public access using SPNEGO to require a valid Kerberos ticket.
Setting ‘hbase.security.authentication.ui’ to ‘kerberos’ in hbase-site.xml is a global switch to have all Web UIs allow only authenticated clients via Kerberos. ‘hbase.security.authentication.spnego.kerberos.principal’ and ‘hbase.security.authentication.spnego.kerberos.keytab’ are two other required properties in hbase-site.xml, the Kerberos principal and keytab to use for the server to use to log in. The primary in the Kerberos principal must be ‘HTTP’ as required by the SPNEGO mechanism, e.g. ‘HTTP/host.domain.com@DOMAIN.COM’.
The estimates of heap usage by the memstore objects (KeyValue, object and array header sizes, etc) have been made more accurate for heap sizes up to 32G (using CompressedOops), resulting in them dropping by 10-50% in practice. This also results in less number of flushes and compactions due to “fatter” flushes. YMMV. As a result, the actual heap usage of the memstore before being flushed may increase by up to 100%. If configured memory limits for the region server had been tuned based on observed usage, this change could result in worse GC behavior or even OutOfMemory errors. Set the environment property (not hbase-site.xml) “hbase.memorylayout.use.unsafe” to false to disable.
Adds a ‘fastpath’ when using the default FIFO rpc scheduler (‘fifo’). Does direct handoff from Reader thread to Handler if there is one ready and willing. Will shine best when high random read workload (YCSB workloadc for instance)
Change the default rpc scheduler from ‘deadline’ to ‘fifo’ instead so it is the same as in branch 0.98. ‘deadline’ was of questionable benefit but with a high cost scheduling. To re-enable ‘deadline’, set hbase.ipc.server.callqueue.type to ‘deadline’ in your hbase-site.xml.
Added a new ByteBufferPool which pools N ByteBuffers. By default it makes off heap ByteBuffers when getBuffer() is called. The size of each buffer defaults to 64KB. This can be configured using ‘hbase.ipc.server.reservoir.initial.buffer.size’. The max number of buffers which can be pooled defaults to twice the number of handler threads in RS. This can be configured with key ‘hbase.ipc.server.reservoir.initial.max’. While responding to read requests and client support Codec, we will create CellBlocks and directly return it as PB payload. For making this block, we will use N ByteBuffers from pool as per the total size of the response cells. The default size of 64 KB for the buffer is inline with the number of bytes written to RPC layer in one short.(That is also 64KB). When at point of time, the calle not able to get a free buffer from the pool (it returns null then), it will make on heap Buffer of same size (as that of Buffers in pool) and use that to create cell block.
Adds a FifoRpcSchedulerFactory so you can try the FifoRpcScheduler by setting “hbase.region.server.rpc.scheduler.factory.class”
Removes the “hbase.online.schema.update.enable” property. from now, every operation that alter the schema (e.g. modifyTable, addFamily, removeFamily, ...) will use the online schema update. there is no need to disable/enable the table.
Removes reference to disabling table in docs for stripe and date-tiered compactions
After HBASE-15931, we will log more details for long-running tasks in AsyncProcess#waitForMaximumCurrentTasks every 10 seconds, including:
documentation changes only - added section to Shell tricks and cross reference from region splitting section
Use @ClassRule to set timeout on test case level (instead of @Rule which sets timeout for the test methods). CategoryBasedTimeout.forClass(..) determines the timeout value based on category annotation (small/medium/large) on the test case.
WARNING: No release note provided for this change.
WARNING: No release note provided for this change.
Adds cacheBlocks to Scan
Removes a doBulkLoad method though it has not been through a full deprecation cycle (but it is ‘damaged’ because it has a parameter that has been properly deprecated). Use the alternative {code}public void doBulkLoad(Path hfofDir, final Admin admin, Table table, RegionLocator regionLocator){code}
See http://mail-archives.apache.org/mod_mbox/hbase-dev/201605.mbox/%3CCAMUu0w-ZiLoLBLO3D76=n3AjUr=VMtTUeYA28weLHYeq8+e3bQ@mail.gmail.com%3E for NOTICE on this ‘premature’ removal.
Added two hooks around WAL restore. preReplayWALs(final ObserverContext<? extends RegionCoprocessorEnvironment> ctx, HRegionInfo info, Path edits) and postReplayWALs(final ObserverContext<? extends RegionCoprocessorEnvironment> ctx, HRegionInfo info, Path edits)
Will be called at start and end of restore of a WAL file. The other hook around WAL restore (preWALRestore ) will be called before restore of every entry within the WAL file.
During periods where DNS resolution was not available or not working correctly, we could previously cache unresolved hostnames forever, in some cases preventing further connections to these hosts even when DNS service was restored. With this change, unresolved hostnames will no longer be cached, and will instead throw an UnknownHostException during connection setup.
Add a new configuration: hbase.ipc.min.client.request.timeout Minimum allowable timeout (in milliseconds) in rpc request's header. This configuration exists to prevent the rpc service regarding this request as timeout immediately.
The core pool size and max pool size of ThreadPoolExecutor should be the same when LinkedBlockingQueue is used. Thus the configurations hbase.hconnection.threads.max, hbase.hconnection.meta.lookup.threads.max, hbase.region.replica.replication.threads.max and hbase.multihconnection.threads.max are used as the number of the core threads, and the related configurations *.thread.core are not used any more.
To find recent set of flakies, run the script added by this patch. Run it to get usage information passing -h:
{code} $ ./dev-support/report-flakies.py -h {code}
If you get the below:
{code} $ python ./dev-support/report-flakies.py Traceback (most recent call last): File “./dev-support/report-flakies.py”, line 25, in <module> import requests ImportError: No module named requests {code}
... install the requests module:
{code} $ sudo pip install requests {code}
Downstream users with long lived applications that need to communicate with secure HBase instances can now rely on the AuthUtil class to handle authenticating via keytab.
For more information, see the javadoc for the org.apache.hadoop.hbase.AuthUtil class.
We were not waiting on all executors in a batch to complete which meant a read-your-own-writes could sometimes fail -- especially if client is loaded; i.e. putting to multiple machines in a cluster. The test for no-more-executors was damaged by the 0.99/0.98.4 fix “HBASE-11403 Fix race conditions around Object#notify”
All active branches now use maven-checkstyle-plugin 2.17 and checkstyle 6.18.
This jira fixes that following bug: During bulkloading, if there are multiple hfiles corresponding to same region, and if they have same timestamps (which may have been set using importtsv.timestamp) and duplicate keys across them, then get and scan may return values coming from different hfiles.
Removed Replication source.shippedKBs metric in favor of source.shippedBytes
The CellCounter map reduce job now supports additional configuration options on the Scan instance it creates, using the org.apache.hadoop.hbase.mapreduce.TableInputFormat defined property names. For a full list of the options, run ./hbase org.apache.hadoop.hbase.mapreduce.CellCounter with no arguments.
CellCounter also no longer creates job counters for per-rowkey and per-rowkey/qualifier cell counts. For most tables, these counters would cause the job to fail due to mapreduce job counter limits.
The following RegionObserver method is deprecated and would no longer be called in hbase 2.0:
public KeyValueScanner preStoreScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, final Scan scan, final NavigableSet<byte[]> targetCols, final KeyValueScanner s) throws IOException {
Instead, override this method:
public KeyValueScanner preStoreScannerOpen(final ObserverContext<RegionCoprocessorEnvironment> c, final Store store, final Scan scan, final NavigableSet<byte[]> targetCols, final KeyValueScanner s, final long readPt) throws IOException {
Now the AsyncFSWAL can write data to a encryption zone on HDFS.
HBase now relies on version 4.3.6 of the Apache Commons HTTPClient library. Downstream users who are exposed to it via the HBase classpath will have to similarly update their dependency.
WARNING: No release note provided for this change.
Adds a section to the debug dump page listing current row locks held.
With previous deadline mode of RPC scheduling (the implementation in SimpleRpcScheduler, which is basically a FIFO except that long-running scans are de-prioritized) and FIFO-based RPC scheduler clients are getting CallQueueTooBigException when RPC call queue is full.
With this patch and when hbase.ipc.server.callqueue.type property is set to “codel” mode, clients will also be getting CallDroppedException, which means that the request was discarded by the server as it considers itself to be overloaded and starts to drop requests to avoid going down under the load. The clients will retry upon receiving this exception. It doesn't clear MetaCache with region locations.
This patch adds new configuration property - hbase.fs.wrapper. If provided, it should be fully qualified class name of the class used as a pluggable wrapper for HFileSystem. This may be useful for specific debugging/tracing needs.
Fixes issue when CallQueueTooBig exception returned to the client could print useless address info (like 0.0.0.0) if RPC server is listening on something other than the host name, making troubleshooting inconvenient.
In HBASE-15711 a new client side property hbase.client.log.batcherrors.details is introduced to allow logging full stacktrace of exceptions for batch error. It's disabled by default and set the property to true will enable it.
New coprocessor table descriptor attribute, hbase.coprocessor.classloader.included.classes, is added. User can specify class name prefixes (semicolon separated) which should be loaded by CoprocessorClassLoader through this attribute using the following syntax: {code} hbase> alter ‘t1’, ‘coprocessor’=>‘hdfs:///foo.jar|com.foo.FooRegionObserver|1001|arg1=1,arg2=2’ {code}
Fixes regression where hbase.rpc.timeout configuration was ignored in branch-1.0+
Adds new methods setOperationTimeout, getOperationTimeout, setRpcTimeout, and getRpcTimeout to Table. In branch-1.3+ they are public interfaces and in 1.0-1.2 they are labeled as @InterfaceAudience.Private.
Adds hbase.client.operation.timeout to hbase-default.xml with default of 1200000
Fix over-persisting in blockcache; no longer save the block PLUS the header of the next block (33 bytes) when writing the cache.
Also removes support for hfileblock v1; hfile block v1 was used writing hfile v1. hfile v1 was the default in hbase before hbase-0.92. hbase.96 would not start unless all v1 hfiles had been compacted out of the cluster.
Introduce an AsyncFSOutput interface which is an abstraction of the original FanOutOneBlockAsyncDFSOutput. Now you can create AsyncFSOutput on any FileSystem using the method AsyncFSOutputHelper.createOutput. The returned AsyncFSOutput will be FanOutOneBlockAsyncDFSOutput if the given FileSystem is a DistributedFileSystem.
When an explicit Get with a one or more columns specified, we at a minimum, were overseeking, reading until we tripped over the next row, regardless, and only then returning. If the next row was in-block, we‘d just do too much seeking but if the next row was in the next (or in the next block beyond that), we would keep seeking and loading blocks until we found the next row before we’d return.
There remains one case where we will still ‘overread’. It is when the row end aligns with the end of the block. In this case we will load the next block just to find that there are no more cells in the current row. See HBASE-15457.
Adds storeFileSize, memstoreSize and tableSize to the per-table metrics.
No functional change. Added javadoc, comments, and extra trace-level logging to make clear what is happening around the reading and caching of hfile blocks.
Use ‘hbase.hstore.compaction.date.tiered.window.factory.class’ to specify the window implementation you like for date tiered compaction. Now the only and default implementation is org.apache.hadoop.hbase.regionserver.compactions.ExponentialCompactionWindowFactory.
{code} <property> <name>hbase.hstore.compaction.date.tiered.window.factory.class</name> <value>org.apache.hadoop.hbase.regionserver.compactions.ExponentialCompactionWindowFactory</value> </property> <property> {code}
Adds per-table metrics aggregated from per-region metrics in region server metrics. New metrics are available under JMX section “Hadoop:service=HBase,name=RegionServer,sub=Tables” and they are available via hadoop metrics2 collectors.
The blockcache UI tab would stop refreshing at 100k blocks (configurable, see “hbase.ui.blockcache.by.file.max”), which isn't very many blocks when doing a big cache, giving a misleading picture of the content of L1 and/or L2 cache. Up the default limit to 1M blocks (UI takes a while but just a few seconds counting over 1M blocks).
Also, when beyond the limit give the user a noticeable WARNING in the UI.
This was a non-issue. The PREFETCH_... flag actually works. While here though made the following additions.
Changes the prefetch TRACE-level loggings to include the word ‘Prefetch’ in them so you know what they are about.
Changes the cryptic logging of the CacheConfig#toString to have some preamble saying why and what column family is responsible (helps figure what is going on)
Add test that verifies setting flag on HColumnDescriptor actually works.
HBASE-13372 Add unit tests for SplitTransaction and RegionMergeTransaction listeners
Protection against CSRF attack can be turned on with config parameter, hbase.rest.csrf.enabled - default value is false.
The custom header to be sent can be changed via config parameter, hbase.rest.csrf.custom.header whose default value is “X-XSRF-HEADER”.
Config parameter, hbase.rest.csrf.methods.to.ignore , controls which HTTP methods are not associated with customer header check.
Config parameter, hbase.rest-csrf.browser-useragents-regex , is a comma-separated list of regular expressions used to match against an HTTP request's User-Agent header when protection against cross-site request forgery (CSRF) is enabled for REST server by setting hbase.rest.csrf.enabled to true.
The implementation came from hadoop/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/security/http/RestCsrfPreventionFilter.java
We should periodically update the RestCsrfPreventionFilter.java in hbase codebase to include fixes to the hadoop implementation.
WALObserver coprocessors now can receive notifications of WAL rolling via the new methods preWALRoll and postWALRoll.
This change is incompatible due to the addition of these methods to the WALObserver interface. Downstream users are encouraged to instead extend the BaseWALObserver class, which remains compatible through this change.
Added update_peer_config to the HBase shell and ReplicationAdmin, and provided a callback for custom replication endpoints to be notified of changes to their configuration and peer data
Add the delegate config for multiwal back. Now you can use ‘hbase.wal.regiongrouping.delegate.provider’ to specify the wal provider you want to use for multiwal. For example: {code} <property> <name>hbase.wal.regiongrouping.delegate.provider</name> <value>asyncfs</value> </property> {code} And the default value is filesystem which is the alias of DefaultWALProvider, i.e., the FSHLog.
With this patch combined with HBASE-15389, when we compact, we can output multiple files along the current window boundaries. There are two use cases:
This will change the way to enable date tiered compaction. To turn it on: hbase.hstore.engine.class: org.apache.hadoop.hbase.regionserver.DateTieredStoreEngine
With tiered compaction all servers in the cluster will promote windows to higher tier at the same time, so using a compaction throttle is recommended: hbase.regionserver.throughput.controller:org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController hbase.hstore.compaction.throughput.higher.bound and hbase.hstore.compaction.throughput.lower.bound need to be set for desired throughput range as uncompressed rates.
Because there will most likely be more store files around, we need to adjust the configuration so that flush won't be blocked and compaction will be properly throttled: hbase.hstore.blockingStoreFiles: change to 50 if using all default parameters when turning on date tiered compaction. Use 1.5~2 x projected file count if changing the parameters, Projected file count = windows per tier x tier count + incoming window min + files older than max age
Because major compaction is turned on now, we also need to adjust the configuration for max file to compact according to the larger file count: hbase.hstore.compaction.max: set to the same number as hbase.hstore.blockingStoreFiles.
For more details, please refer to the design spec at https://docs.google.com/document/d/1_AmlNb2N8Us1xICsTeGDLKIqL6T-oHoRLZ323MG_uy8/edit#
Use hbase org.apache.hadoop.hbase.procedure2.store.wal.ProcedureWALPrettyPrinter to print the content of a Procedure WAL. e.g. hbase org.apache.hadoop.hbase.procedure2.store.wal.ProcedureWALPrettyPrinter -f /hbase/MasterProcWALs/state-00000000000000002571.log
To aid troubleshooting of MapReduce job that rely on the HBase provided input format, splits now include the encoded region name they cover.
WARNING: No release note provided for this change.
WARNING: No release note provided for this change.
Add the following config in hbase-site.xml if you want to use secure protobuf wal writer together with AsyncFSWAL {code} <property> <name>hbase.regionserver.hlog.async.writer.impl</name> <value>org.apache.hadoop.hbase.regionserver.wal.SecureAsyncProtobufLogWriter</value> </property> <property> {code}
WARNING: No release note provided for this change.
To enable, set the WALProvider as follows:
{code} <property> <name>hbase.wal.provider</name> <value>asyncfs</value> </property> <property> {code}
To check which provider is active, look for the log line:
LOG.info("Instantiating WALProvider of type " + clazz);
HBASE-14256 Correct confusing flush task message
Adds a configuration parameter “hbase.ipc.max.request.size” which defaults to 256MB to protect the server against very large incoming RPC requests. All requests larger than this size will be immediately rejected before allocating any resources (memory allocation, etc).
Adds a new metric for called “averageRegionSize” that is emitted as a regionserver metric. Metric description: Average region size over the region server including memstore and storefile sizes
This fix decreases client's memory allocation during writes by more than 50%.
WARNING: No release note provided for this change.
Adds get_peer_config and list_peer_configs to the hbase shell.
Failed taking snapshot - Manifest proto-message too large. add property (“snapshot.manifest.size.limit”) to change max size of proto-message
Fixed an issue in REST server checkAndDelete operation where the remaining cells other than the to-be-checked column are also applied in the Delete operation. Also fixed an issue in RemoteHTable where the Delete object was not passed correctly to the REST server side.
Per-region metrics related to Get histograms are changed from being response size based into being latency based similar to the per-regionserver metrics of the same name.
Added GetSize histogram metrics at the per-regionserver and per-region level for the response sizes.
[ADVANCED USERS ONLY] This patch adds a new experimental module hbase-rsgroup. It is an advanced feature for partitioning regionservers into distinctive groups for strict isolation, and should only be used by users who are sophisticated enough to understand the full implications and have a sufficient background in managing HBase clusters.
RSGroups can be defined and managed with shell commands or corresponding Java APIs. A server can be added to a group with hostname and port pair, and tables can be moved to this group so that only regionservers in the same rsgroup can host the regions of the table. RegionServers and tables can only belong to 1 group at a time. By default, all tables and regionservers belong to the “default” group. System tables can also be put into a group using the regular APIs. A custom balancer implementation tracks assignments per rsgroup and makes sure to move regions to the relevant regionservers in that group. The group information is stored in a regular HBase table, and a zookeeper-based read-only cache is used at the cluster bootstrap time.
To enable, add the following to your hbase-site.xml and restart your Master:
<property> <name>hbase.coprocessor.master.classes</name> <value>org.apache.hadoop.hbase.rsgroup.RSGroupAdminEndpoint</value> </property> <property> <name>hbase.master.loadbalancer.class</name> <value>org.apache.hadoop.hbase.rsgroup.RSGroupBasedLoadBalancer</value> </property>
Then use the shell ‘rsgroup’ commands to create and manipulate regionserver groups: e.g. to add a group and then add a server to it, do as follows:
hbase(main):008:0> add_rsgroup ‘my_group’ Took 0.5610 seconds
This adds a group to the ‘hbase:rsgroup’ system table. Add a server (hostname + port) to the group using the ‘move_rsgroup_servers’ command as follows:
hbase(main):010:0> move_rsgroup_servers ‘my_group’,[‘k.att.net:51129’]
Adds a new metric named “writtenBytes” as a per-regionserver metric. Metric Description: Size (in bytes) of the data written to the WAL.
HBase now ensures that the JDK tools jar used during the build process is not exposed to downstream clients as a transitive dependency of hbase-annotations.
If you need to have the JDK tools jar in your classpath, you should add a system dependency on it. See the hbase-annotations pom for an example of the necessary pom additions.
When using the bulk load helper provided by the hbase-spark module, output files will now be written into temporary files and only made available when the executor has successfully completed.
Previously, failed executors would leave their files in place in a way that would be picked up by a bulk load command. This caused retried failures to include spurious copies of some cells.
HBASE-15364 Fix unescaped < and > characters in Javadoc
When all filters in a MUST_PASS_ONE FilterList return a SEEK_USING_NEXT_HINT code, we return SEEK_NEXT_USING_HINT from the FilterList#filterKeyValue() to utilize the lowest seek value.
This patch fixes some issues when MetaCache (region location cache) gets unnecessarily dropped on the client.
On master branch we now in RegionServerCallable and RegionServerAdminCallable pass the actual exception down to Connection#updateCachedLocation, so we could check there if the exception is “meta-clearing” or not.
on branch-1, branch-1.2 and branch 1.3 we now check if the exception is meta-clearing or not in AsyncProcess (this check was there on master, but not on earlier branches)
Removed ScanNext histogram metrics as regionserver level and per-region level metrics since the semantics is not compatible with other similar metrics (size histogram vs latency histogram).
Instead, this patch adds ScanTime and ScanSize histogram metrics at the regionserver and per-region level.
Add a new config: hbase.block.data.cacheonread, which is a global switch for caching data blocks on read. The default value of this switch is true, and data blocks will be cached on read if the block cache is enabled for the family and cacheBlocks flag is set to be true for get and scan operations. If this global switch is set to false, data blocks won't be cached even if the block cache is enabled for the family and the cacheBlocks flag of Gets or Scans are sets as true. Bloom blocks and index blocks are always be cached if the block cache of the regionserver is enabled. One usage of this switch is for the performance tests for the extreme case that the cache for data blocks all missed and all data blocks are read from underlying file system.
Previously RPC request scheduler in HBase had 2 modes in could operate in:
This patch adds new type of scheduler to HBase, based on the research around controlled delay (CoDel) algorithm [1], used in networking to combat bufferbloat, as well as some analysis on generalizing it to generic request queues [2]. The purpose of that work is to prevent long standing call queues caused by discrepancy between request rate and available throughput, caused by kernel/disk IO/networking stalls.
New RPC scheduler could be enabled by setting hbase.ipc.server.callqueue.type=codel in configuration. Several additional params allow to configure algorithm behavior -
hbase.ipc.server.callqueue.codel.target.delay hbase.ipc.server.callqueue.codel.interval hbase.ipc.server.callqueue.codel.lifo.threshold
[1] Controlling Queue Delay / A modern AQM is just one piece of the solution to bufferbloat. http://queue.acm.org/detail.cfm?id=2209336 [2] Fail at Scale / Reliability in the face of rapid change. http://queue.acm.org/detail.cfm?id=2839461
Date tiered compaction policy is a date-aware store file layout that is beneficial for time-range scans for time-series data.
When it performs well:
reads for limited time ranges, especially scans of recent data
When it doesn't perform as well:
random gets without a time range frequent deletes and updates out of order data writes, especially writes with timestamps in the future bulk loads of historical data
Recommended configuration: To turn on Date Tiered Compaction (It is not recommended to turn on for the whole cluster because that will put meta table on it too and random get on meta table will be impacted): hbase.hstore.compaction.compaction.policy: org.apache.hadoop.hbase.regionserver.compactions.DateTieredCompactionPolicy
Parameters for Date Tiered Compaction: hbase.hstore.compaction.date.tiered.max.storefile.age.millis: Files with max-timestamp smaller than this will no longer be compacted.Default at Long.MAX_VALUE. hbase.hstore.compaction.date.tiered.base.window.millis: base window size in milliseconds. Default at 6 hours. hbase.hstore.compaction.date.tiered.windows.per.tier: number of windows per tier. Default at 4. hbase.hstore.compaction.date.tiered.incoming.window.min: minimal number of files to compact in the incoming window. Set it to expected number of files in the window to avoid wasteful compaction. Default at 6. hbase.hstore.compaction.date.tiered.window.policy.class: the policy to select store files within the same time window. It doesn’t apply to the incoming window. Default at exploring compaction. This is to avoid wasteful compaction.
With tiered compaction all servers in the cluster will promote windows to higher tier at the same time, so using a compaction throttle is recommended: hbase.regionserver.throughput.controller:org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController
Because there will most likely be more store files around, we need to adjust the configuration so that flush won't be blocked and compaction will be properly throttled: hbase.hstore.blockingStoreFiles: change to 50 if using all default parameters when turning on date tiered compaction. Use 1.5~2 x projected file count if changing the parameters, Projected file count = windows per tier x tier count + incoming window min + files older than max age
For more details, please refer to the design spec at https://docs.google.com/document/d/1_AmlNb2N8Us1xICsTeGDLKIqL6T-oHoRLZ323MG_uy8/edit#
Fixed an issue in REST server checkAndPut operation where the remaining cells other than the to-be-checked column are also applied in the put operation .
Implement a fan-out asynchronous DFSOutputStream for implementing new WAL writer.
mmap() based bucket cache can be configured by specifying the property {code} <property> <name>hbase.bucketcache.ioengine</name> <value> mmap://filepath </value> </property> {code} This mode of bucket cache is ideal when your file based bucket cache size is lesser than then available RAM. When the cache is bigger than the available RAM then the kernel page faults will make this cache perform lesser particularly in case of scans.
Checksumming is cpu intensive. HBase computes additional checksums for HFiles (hdfs does checksums too) and stores them inline with file data. During reading, these checksums are verified to ensure data is not corrupted. This patch tries to use Hadoop Native Library for checksum computation, if it’s available, otherwise falls back to standard Java libraries. Instructions to load NHL in HBase can be found here (http://hbase.apache.org/book.html#hadoop.native.lib).
Default checksum algorithm has been changed from CRC32 to CRC32C primarily because of two reasons: 1) CRC32C has better error detection properties, and 2) New Intel processors have a dedicated instruction for crc32c computation (SSE4.2 instruction set)*. This change is fully backward compatible. Also, users should not see any differences except decrease in cpu usage. To keep old settings, set configuration ‘hbase.hstore.checksum.algorithm’ to ‘CRC32’.
* On linux, run 'cat /proc/cpuinfo’ and look for sse4_2 in list of flags to see if your processor supports SSE4.2.
A new flag is added for Canary tool: -treatFailureAsError When this flag is specified, read / write failure would result in Canary tool exit code of 5.
Now we can write duplicated WAL entries into different WAL files. This feature is required by the replication consistency fix and new implementation of WAL writer.
The constructor for o.a.h.hbase.ProcedureInfo was mistakenly labeled IA.Public in previous releases and has now changed to IA.Private. Downstream users are safe to consume ProcedureInfo objects returned from HBase public interfaces, but should not expect to be able to reliably create new instances themselves.
The method ProcedureInfo.setNonceKey has been removed, because it should not have been exposed to clients.
Adds being able to Scan each column family with a different time range. Adds new methods setColumnFamilyTimeRange and getColumnFamilyTimeRange to Scan.
This release note tries to tell the general story. Dive into sub-tasks for more specific release noting.
Increments, appends, checkAnd* have been slow since hbase-.1.0.0. The unification of mvcc and sequence id done by HBASE-8763 was responsible.
A ‘fast-path’ workaround was added by HBASE-15031 “Fix merge of MVCC and SequenceID performance regression in branch-1.0 for Increments”. It became available in 1.0.3 and 1.1.3. To enable the fast path, set “hbase.increment.fast.but.narrow.consistency” and then rolling restart. The workaround was for increments only (appends, checkAndPut, etc., were not addressed. See HBASE-15031 release note for more detail).
Subsequently, the regression was properly identified and fixed in HBASE-15213 and the fix applied to branch-1.0 and branch-1.1. As it happens, hbase-1.2.0 does not suffer from the performance regression (though the thought was that it did -- and so it got the fast-path patch too via HBASE-15092) nor does the master branch. HBASE-15213 identified that HBASE-12751 (as a side effect) had cured the regression.
hbase-1.0.4 (if it is ever released -- 1.0 has been end-of-lifed) and hbase-1.1.4 will have the HBASE-15213 fix. If you are suffering from the increment regression and you are on 1.0.3 or 1.1.3, you can enable the work around to get back your increment performance but you should upgrade.
In here we perf tested a realignment of the write pipeline and mvcc handling. Thought was that this work was a predicate for a general fix of HBASE-14460 (turns out, realignment of write path was not needed to fix the increment perf regression). The perf testing here made it so we were able to simplify writing. HBASE-15158 was just committed. This work is done.
Changed the write pipeline order; made it more rational, easier-to-reason-about doing all updates to WA, MemStore, and mvcc while read/write rowlock is held where before we'd release after WAL append and then do sync and mvcc.
Add append, increment, checkAndMutate, checkAndPut, and checkAndDelete tests to PerformanceEvaluation tool. Below are excerpts from new usage from PE:
.... Command: append Append on each row; clients overlap on keyspace so some concurrent operations checkAndDelete CheckAndDelete on each row; clients overlap on keyspace so some concurrent operations checkAndMutate CheckAndMutate on each row; clients overlap on keyspace so some concurrent operations checkAndPut CheckAndPut on each row; clients overlap on keyspace so some concurrent operations filterScan Run scan test using a filter to find a specific row based on it's value (make sure to use --rows=20) increment Increment on each row; clients overlap on keyspace so some concurrent operations randomRead Run random read test .... Examples: ... To run 10 clients doing increments over ten rows: $ bin/hbase org.apache.hadoop.hbase.PerformanceEvaluation --rows=10 --nomapred increment 10
Removed IncrementPerformanceTest. It is not as configurable as the additions made here.
This issue fixes
As we use tags for Cell level security, this fixes 2 security issues
Before HBASE-15129, if somehow hbase-default.xml is not on classpath, default values for hbase.fs.tmp.dir and hbase.bulkload.staging.dir are left empty. After HBASE-15129, default values of both properties are set to “/user/<user.name>/hbase-staging”.
Adds means of throttling flush throughput. By default there is no limit; we use NoLimitThroughputController. An alternative controller, PressureAwareFlushThroughputController, allows specifying throughput bounds. A new simple factor, flush pressure, influences throughput. See PressureAwareFlushThroughputController.java class for detail.
For E2E off heaped read path, first of all there should be an off heap backed BucketCache(BC). Configure ‘hbase.bucketcache.ioengine’ to offheap in hbase-site.xml. Also specify the total capacity of the BC using hbase.bucketcache.size config. Please remember to adjust value of ‘HBASE_OFFHEAPSIZE’ in hbase-env.sh as per this capacity. Here-by we specify the max possible off-heap memory allocation for the RS java process. So this should be bigger than the off-heap BC size. Please keep in mind that there is no default for hbase.bucketcache.ioengine which means the BC is turned OFF by default.
Next thing to tune is the ByteBuffer pool in the RPC server side. The buffers from this pool will be used to accumulate the cell bytes and create a result cell block to send back to the client side. ‘hbase.ipc.server.reservoir.enabled’ can be used to turn this pool ON or OFF. By default this pool is ON and available. HBase will create off heap ByteBuffers and pool them. Please make sure not to turn this OFF if you want E2E off heaping in read path. If this pool is turned off, the server will create temp buffers on heap to accumulate the cell bytes and make a result cell block. This can impact the GC on a highly read loaded server. The user can tune this pool with respect to how many buffers are in the pool and what should be the size of each ByteBuffer. Use the config ‘hbase.ipc.server.reservoir.initial.buffer.size’ to tune each of the buffer sizes. Defaults is 64 KB.
When the read pattern is a random row read and each of the rows are smaller in size compared to this 64 KB, try reducing this. When the result size is larger than one ByteBuffer size, the server will try to grab more than one buffer and make a result cell block out of these. When the pool is running out of buffers, the server will end up creating temporary on-heap buffers.
The maximum number of ByteBuffers in the pool can be tuned using the config ‘hbase.ipc.server.reservoir.initial.max’. Its value defaults to 64 * region server handlers configured (See the config ‘hbase.regionserver.handler.count’). The math is such that by default we consider 2 MB as the result cell block size per read result and each handler will be handling a read. For 2 MB size, we need 32 buffers each of size 64 KB (See default buffer size in pool). So per handler 32 ByteBuffers(BB). We allocate twice this size as the max BBs count such that one handler can be creating the response and handing it to the RPC Responder thread and then handling a new request creating a new response cell block (using pooled buffers). Even if the responder could not send back the first TCP reply immediately, our count should allow that we should still have enough buffers in our pool without having to make temporary buffers on the heap. Again for smaller sized random row reads, tune this max count. There are lazily created buffers and the count is the max count to be pooled.
The setting for HBASE_OFFHEAPSIZE in hbase-env.sh should consider this off heap buffer pool at the RPC side also. We need to config this max off heap size for RS as a bit higher than the sum of this max pool size and the off heap cache size. The TCP layer will also need to create direct bytebuffers for TCP communication. Also the DFS client will need some off-heap to do its workings especially if short-circuit reads are configured. Allocating an extra of 1 - 2 GB for the max direct memory size has worked in tests.
If you still see GC issues even after making E2E read path off heap, look for issues in the appropriate buffer pool. Check the below RS log with INFO level:
“Pool already reached its max capacity : XXX and no free buffers now. Consider increasing the value for ‘hbase.ipc.server.reservoir.initial.max’ ?”
If you are using co processors and refer the Cells in the read results, DO NOT store reference to these Cells out of the scope of the CP hook methods. Some times the CPs need store info about the cell (Like its row key) for considering in the next CP hook call etc. For such cases, pls clone the required fields of the entire Cell as per the use cases. [ See CellUtil#cloneXXX(Cell) APIs ]
Added a new command line argument: --auth-as-server to enable authenticating to ZooKeeper as the HBase Server principal. This is required for secure clusters for doing replication operations like add_peer, list_peers, etc until HBASE-11392 is fixed. This advanced option can also be used for manually fixing secure znodes.
Commands can now be invoked like: hbase --auth-as-server shell hbase --auth-as-server zkcli
HBCK in secure setup also needs to authenticate to ZK using servers principals.This is turned on by default (no need to pass additional argument).
When authenticating as server, HBASE_SERVER_JAAS_OPTS is concatenated to HBASE_OPTS if defined in hbase-env.sh. Otherwise, HBASE_REGIONSERVER_OPTS is concatenated.
WARNING: No release note provided for this change.
After this JIRA we will not be doing any scanner reset after compaction during a course of a scan. The files that were compacted will still be continued to be used in the scan process. The compacted files will be archived by a background thread that runs every 2 mins by default only when there are no active scanners on those comapcted files. The above duration can be controlled using the knob ‘hbase.hfile.compactions.cleaner.interval’.
With this patch, hbase.rpc.protection can now take multiple comma-separate QOP values. Accepted QOP values remain unchanged and are ‘authentication’, ‘integrity’, and ‘privacy’. Server or client can use this configuration to specify their preference (in decreasing order) while negotiating QOP. This feature can be used to upgrade or downgrade QOP in an online cluster without compromising availability (i.e. taking cluster offline). For e.g. to change qop from A to B, typical steps would be: “A” --> “B,A” --> rolling restart --> “B” --> rolling restart
Sidenote: Based on experimentation, server‘s choice is given higher preference than client’s choice. i.e. if server‘s choices are “A,B,C” and client’s choices are “B,C,A”, both A and B are acceptable, but A is chosen.
The config parameter, hbase.normalizer.enabled, has been dropped since it is not used in the code base.
The `hbase version` command now outputs directly to stdout rather than to a logger. This change allows the version information to be output consistently regardless of logger configuration. Naturally, this also means the command output ignores all logger configuration. Furthermore, the move from loggers to direct output changes the output of the command to omit metadata commonly included in logger ouput such as a timestamp, log level, and logger name.
The property ‘hbase.hfile.compactions.discharger.interval’ has been renamed to ‘hbase.hfile.compaction.discharger.interval’ that describes the interval after which the compaction discharger chore service should run. The property ‘hbase.hfile.compaction.discharger.thread.count’ describes the thread count that does the compaction discharge work. The CompactedHFilesDischarger is a chore service now started as part of the RegionServer and this chore service iterates over all the onlineRegions in that RS and uses the RegionServer's executor service to launch a set of threads that does this job of compaction files clean up.
FIFO compaction policy selects only files which have all cells expired. The column family MUST have non-default TTL. Essentially, FIFO compactor does only one job: collects expired store files.
Because we do not do any real compaction, we do not use CPU and IO (disk and network), we do not evict hot data from a block cache. The result: improved throughput and latency both write and read. See: https://github.com/facebook/rocksdb/wiki/FIFO-compaction-style
This patch changes the semantic around namespace create/delete/modify when coprocessor asks that the invocation be by-passed. Previous the by-pass was done silently -- the method would just return with no indication as to whether by-pass route had been taken or not. This patch adds throwing of a BypassCoprocessorException which is thrown if we have been asked to bypass a call.
The bypass facility has been in place since hbase 1.0.0 when namespace creation/deletion, etc.., was originally added in HBASE-8408 (HBASE-15071 is about addressing bypass handling in a general way)
When using the new AsyncRpcClient introduced in HBase 1.1.0 (HBASE-12684), time outs now result in an IOException wrapped around a CallTimeoutException instead of a bare CallTimeoutException. This change makes the AsyncRpcClient behave the same as the default HBase 1.y RPC client implementation.
spark.hbase.bulkGetSize in HBaseSparkConf is for grouping bulkGet, and default value is 1000.
Adds column displaying current aggregated call queues size in region server queues tab UI.
And 1.1, 1.0, and 0.98.
WARNING: No release note provided for this change.
Limiting the amount of memory resident for any one request allows the server to handle concurrent requests smoothly. To this end we added the ability to limit the size of responses to a multi request. That worked well however it correctly represent the amount of memory resident. So this issue adds on a an approximation of the number of blocks held for a request.
All clients before 1.2.0 will not get this multi request chunking based upon blocks kept. All clients 1.2.0 and after will.
Rolling WAL events across a cluster can be highly correlated, hence flushing memstores, hence triggering minor compactions, that can be promoted to major ones. These events are highly correlated in time if there is a balanced write-load on the regions in a table. Default value for maximum WAL files (* hbase.regionserver.maxlogs*), which controls WAL rolling events - 32 is too small for many modern deployments. Now we calculate this value dynamically (if not defined by user), using the following formula:
maxLogs = Math.max( 32, HBASE_HEAP_SIZE * memstoreRatio * 2/ LogRollSize), where
memstoreRatio is *hbase.regionserver.global.memstore.size* LogRollSize is maximum WAL file size (default 0.95 * HDFS block size)
We need to make sure that we avoid fully or minimize events when RS has to flush memstores prematurely only because it reached artificial limit of hbase.regionserver.maxlogs, this is why we put this 2 x multiplier in equation, this gives us maximum WAL capacity of 2 x RS memstore-size.
Runaway WAL files.
The default log rolling period (1h) allows to accumulate up to 2 X Memstore Size data in a WAL. For heap size - 32G and all other default setting, this gives ~ 26GB of data. Under heavy write load, the number of WAL files can increase dramatically. RegionServer LogRoller will be archiving old WALs periodically. User has three options, either override default hbase.regionserver.maxlogs or override default hbase.regionserver.logroll.period (decrease), or both to control runaway WALs.
For system with bursty write load, the hbase.regionserver.logroll.period can be decreased to lower value. In this case the maximum number of wal files will be defined by the total size of memstore (unflushed data), not by the hbase.regionserver.maxlogs. But for majority of applications there will be no issues with defaults. Data will be flushed periodically from memstore, the LogRoller will archive old wal files and the system will never reach the new defaults for hbase.regionserver.maxlogs, unless the system is under extreme load for prolonged period of time, but in this case, decreasing hbase.regionserver.logroll.period allows us to control runaway wal files.
The following table gives the new default maximum log files values for several different Region Server heap sizes:
heap memstore perc maxLogs 1G 40% 32 2G 40% 32 10G 40% 80 20G 40% 160 32G 40% 256
Setting hbase.cache.memcached.spy.optimze to true will allow the spy memcached client to try and optimize for the number of requests outstanding. This can increase throughput but can also increase variance for request times.
Setting it to true will help when round trip times are longer. Setting it to false ( the default ) will help ensure a more even distribution of response times.
Updated yammer metrics to version 3.1.2 (now it's been renamed to dropwizard). API has changed quite a bit, consult https://dropwizard.github.io/metrics/3.1.0/manual/core/ for additional information.
Note that among other things, in yammer 2.2.0 histograms were by default created in non-biased mode (uniform sampling), while in 3.1.0 histograms created via MetricsRegistry.histogram(...) are by default exponentially decayed. This shouldn't affect end users, though.
If the configured RPC controller factory (via hbase.rpc.controllerfactory.class) cannot be found in the classpath or loaded, we fall back to using the default RPC controller factory in HBase.
The HBase region server will now send a chunk of get responses to a client if the total response size is too large. This will only be done for clients 1.2.0 and beyond. Older clients by default will have the old behavior.
This patch is for the case where the basic flow is like this:
I want to get a single column from lots of rows. So I create a list of gets. Then I send them to table.get(List<Get>). If the regions for that table are spread out then those requests get chunked out to all the region servers. No one regionserver gets too many. However if one region server contains lots of regions for that table then a multi action can contain lots of gets. No single get is too onerous. However the regionserver won't return until every get is complete. So if there are thousands of gets that are sent in one multi then the regionserver can retain lots of data in one thread.
In HBASE-14906 we use “hbase.hregion.memstore.flush.size/column_family_number” as the default threshold for memstore flush instead of the fixed value through “hbase.hregion.percolumnfamilyflush.size.lower.bound” property, which makes the default threshold more flexible to various use case. We also introduce a new property in name of “hbase.hregion.percolumnfamilyflush.size.lower.bound.min” with 16M as the default value to avoid small flush in cases like hundreds of column families.
After this change setting “hbase.hregion.percolumnfamilyflush.size.lower.bound” in hbase-site.xml won't take effect anymore, but expert users could still set this property in table descriptor to override the default value just as before
This enhances the HBase replication to support replication of bulk loaded data. This is configurable, by default it is set to false which means it will not replicate the bulk loaded data to its peer(s). To enable it set “hbase.replication.bulkload.enabled” to true.
Following are the additional configurations added for this enhancement, a. hbase.replication.cluster.id - This is manadatory to configure in cluster where replication for bulk loaded data is enabled. A source cluster is uniquely identified by sink cluster using this id. This should be configured in the source cluster configuration file for all the RS. b. hbase.replication.conf.dir - This represents the directory where all the active cluster's file system client configurations are defined in subfolders corresponding to their respective replication cluster id in peer cluster. This should be configured in the peer cluster configuration file for all the RS. Default is HBASE_CONF_DIR. c. hbase.replication.source.fs.conf.provider - This represents the class which provides the source cluster file system client configuration to peer cluster. This should be configured in the peer cluster configuration file for all the RS. Default is org.apache.hadoop.hbase.replication.regionserver.DefaultSourceFSConfigurationProvider
For example: If source cluster FS client configurations are copied in peer cluster under directory /home/user/dc1/ then hbase.replication.cluster.id should be configured as dc1 and hbase.replication.conf.dir as /home/user
Note: a. Any modification to source cluster FS client configuration files in peer cluster side replication configuration directory then it needs to restart all its peer(s) cluster RS with default hbase.replication.source.fs.conf.provider. b. Only ‘xml’ type files will be loaded by the default hbase.replication.source.fs.conf.provider.
As part of this we have made following changes to LoadIncrementalHFiles class which is marked as Public and Stable class, a. Raised the visibility scope of LoadQueueItem class from package private to public. b. Added a new method loadHFileQueue, which loads the queue of LoadQueueItem into the table as per the region keys provided.
HBASE-7171 adds 2 new pages to the region server Web UI to ease debugging and provide greater insight into the physical data layout.
Region names in UI table listing all regions (on the RS status page) are now hyperlinks leading to region detail page which shows some aggregate memstore information (currently just memory used) along with the list of all Store Files (HFiles) in the region. Names of Store Files are also hyperlinks leading to Store File detail page, which currently runs ‘hbase hfile’ command behind the scene and displays statistics about store file.
Region observer notifications w.r.t. compaction request are now audited with request user through proper scope of doAs() calls.
Region observer notifications w.r.t. merge request are now audited with request user through proper scope of doAs() calls.
When split is requested by non-super user, split related notifications for Coprocessor are executed using the login of the request user. Previously the notifications were carried out as super user.
Adds a timeout to server read from clients. Adds new configs hbase.thrift.server.socket.read.timeout for setting read timeout on server socket in milliseconds. Default is 60000;
Corrections to content of “book.html”, which is pulled from various *.adoc files and *.xml files. -- corrects typos/misspellings -- corrects incorrectly formatted links
Configuration properties for org.apache.hadoop.hbase.mapreduce.TableOutputFormat can now be overridden by prefixing the property keys with “hbase.mapred.output.”. When the configuration is applied to TableOutputFormat, these entries will be rewritten with the prefix removed -- ie. “hbase.mapred.output.hbase.security.authentication” becomes “hbase.security.authentication”. This can be useful when directing output to a peer cluster with different security configuration, for example.
This issue resolves a potential security vulnerability. For all versions we update our commons-collections dependency to the release that fixes the reported vulnerability in that library. In 0.98 we additionally disable by default a feature of code carried from 0.94 for backwards compatibility that is not needed.
Locks on row are now reader/writer rather than exclusive.
Moves sequenceid out of HRegion and into MVCC class; MVCC is now in charge. A WAL append is still stamped in same way (we pass MVCC context in a few places where we previously we did not).
MVCC methods cleaned up. Make a bit more sense now. Less of them.
Simplifies our update of MemStore/WAL. Now we update memstore AFTER we add to WAL (but before we sync). This fixes possible dataloss when two edits came in with same coordinates; we could order the edits in memstore differently to how they arrived in the WAL.
Marked as an incompatible change because it breaks Distributed Log Replay, a feature we'd determined already was unreliable and to be removed (See http://search-hadoop.com/m/YGbbhTJpoal8GD1).
Very large blocks can fragment the heap and cause bad issues for the garbage collector, especially the G1GC. Now there is a maximum size that a block can be and still stick in the LruBlockCache. That size defaults to 16mb but can be controlled by changing “hbase.lru.max.block.size”
New configuration option: hbase.hstore.compaction.max.size.offpeak - maximum selection size eligible for minor compaction during off peak hours. hbase.hstore.compaction.max.size - this is default maximum if no off-peak hours are defined or if no maximum off-peak maximum size is defined.
Incorporated in HBASE-13014.
Secure HBase now supports a permissive mode to allow mixed secure and insecure clients. This allows clients to be incrementally migrated over to a secure configuration. To enable clients to continue to connect using SIMPLE authentication when the cluster is configured for security, set “hbase.ipc.server.fallback-to-simple-auth-allowed” equal to “true” in hbase-site.xml. NOTE: This setting should ONLY be used as a temporary measure while converting clients over to secure authentication. It MUST BE DISABLED for secure operation.
Memstore periodic flusher used to flush META table every 5 minutes but not any other system tables. This jira extends it to flush all system tables within this time period.
You can specify one of the predefined set of Monkeys when you run Integration Tests by passing the -m|--monkey arguments on the command line; e.g -m CALM or -m SLOW_DETERMINISTIC
This patch makes it so you can pass the name of a class as the monkey to run: e.g. -m org.example.KingKong
After this change, hbase.client.reties.number universally means the number of retry which is one less than total tries number, for both non-batch operations like get/scan/increment etc. which uses RpcRetryingCallerImpl#callWithRetries to submit the call or batch operations like put through AsyncProcess#submit.
Note that previously this property means total tries number for puts, so please adjust the setting of its value if necessary. Please also be cautious when setting it to zero since retry is necessary for client cache update when region move happens.
For master branch(2.0 version), the BoundedByteBufferPool always create Direct (off heap) ByteBuffers and return that. For branch-1(1.3 version), byte default the buffers returned will be off heap. This can be changed to return on heap ByteBuffers by configuring ‘hbase.ipc.server.reservoir.direct.buffer’ to false.
Adds server version to the listing of regionservers on the master home page.
if a cluster where the versions deviate, at the bottom of the ‘Version’ column on the master home page listing of ‘Region Servers’, you will see a note in red that says something like: ‘Total:10 9 nodes with inconsistent version’
Introduces collection and reporting of various client-perceived metrics. Metrics are exposed via JMX under “org.apache.hadoop.hbase.client.MetricsConnection”. Metrics are scoped according to connection instance, so multiple connection objects (ie, to different clusters) will report their metrics separately. Metrics are disabled by default, must be enabled by configuring “hbase.client.metrics.enable=true”.
HBase daemons can now be signaled to reload their config by sending SIGHUP to the java process. Not all config parameters can be reloaded.
In order for this new feature to work the hbase-daemon.sh script was changed to use disown rather than nohup. Functionally this shouldn't change anything but the processes will have a different parent when being run from a connected login shell.
HBASE-14502 Purge use of jmock and remove as dependency
By setting hbase.resolve.hostnames.on.failure to false you can reduce the number of dns name resolutions that a client will do. However if machines leave and come back with different ip's the changes will not be noticed by the clients. So only set hbase.resolve.hostnames.on.failure to false if your cluster dns is not changing while clients are connected.
This patch adds shell support for region normalizer (see HBASE-13103).
3 commands have been added to hbase shell ‘tools’ command group (modeled on how the balancer works):
Also ‘alter’ command has been extended to allow user to enable/disable region normalization per table (disabled by default). Use it as
alter ‘testtable’, {NORMALIZATION_MODE => ‘true’}
Here is the help for the normalize command:
{code} hbase(main):008:0> help ‘normalize’ Trigger region normalizer for all tables which have NORMALIZATION_MODE flag set. Returns true if normalizer ran successfully, false otherwise. Note that this command has no effect if region normalizer is disabled (make sure it's turned on using ‘normalizer_switch’ command).
Examples:
hbase> normalize {code}
Region observer notifications w.r.t. split request are now audited with request user through proper scope of doAs() calls.
Remove calling getNumCurrentReplicas on HdfsDataOutputStream via reflection. getNumCurrentReplicas showed up in hadoop 1+ and hadoop 0.2x. In hadoop-2 it was deprecated.
The WAL append was changed by HBASE-12751. Every append now sets a latch on an edit. The latch needs to be cleared or else the WAL will hang. The original failures in TestHRegion turned up ‘holes’ where we were failing to throw the latch if we skipped out early because we were interrupted. Other ‘holes’ were found where we had mocked up a WAL so the latch would just stay in place. Futher holes were found appending WAL markers... here we were skipping the mvcc completely for a few edits. A clean up of WALUtils made all markers take the same code paths.
Patch will effectively work with Hadoop version 2.6 or greater with a launch of “internal.nameservices”. There will be no change in versions older than 2.6.
Move external block cache to it's own module. This will reduce dependencies for people who use hbase-server. Currently Memcached is the reference implementation for external block cache. External block caches allow HBase to take advantage of other more complex caches that can live longer than the HBase regionserver process and are not necessarily tied to a single computer life time. However external block caches add in extra operational overhead.
Tests run with client executors that have core thread count of 4 and a keepalive of 3 seconds. They used to default to 256 core threads and 60 seconds for keepalive.
To use rpc protection in HBase, set the value of ‘hbase.rpc.protection’ to: ‘authentication’ : simple authentication using kerberos ‘integrity’ : authentication and integrity ‘privacy’ : authentication and confidentiality
Earlier, HBase reference guide erroneously mentioned in some places to set the value to ‘auth-conf’. This patch fixes the guide and adds temporary support for erroneously recommended values.
In HBASE-14306 we‘ve changed default strategy of RegionGroupingProvider from “identify” to “bounded”, so it’s required to explicitly set “hbase.wal.regiongrouping.strategy” to “identify” if user still wants to use one WAL per region
Please also notice that in the new framework there will be one WAL per group, and the region-group mapping is decided by RegionGroupingStrategy. Accordingly, we've removed BoundedRegionGroupingProvider and added BoundedRegionGroupingStrategy as a replacement. If you already have a customized class for hbase.wal.regiongrouping.strategy, please check the new logic and make updates if necessary.
ReplicationSourceManager now could track multiple wal paths. Notice that although most changes are internal and all metrics names remain the same, signature of below methods in MetricsSource are changed:
The following metrics for primary region replica are added:
blockCacheHitCountPrimary blockCacheMissCountPrimary blockCacheEvictionCountPrimary
Tighten up WAL-use semantic.
The above rules have been applied laxly up to this; it used to be possible to get a good sync to go in over the top of a failed append. This has been fixed in this patch.
Also fixed a hang in the WAL subsystem if a request to pause the write pipeline took on a failed sync. before the roll requests sync got scheduled.
TODO: Revisit our WAL system. HBASE-12751 helps rationalize our write pipeline. In particular, it manages sequenceid inside mvcc which should make it so we can purge mechanism that writes empty, unflushed appends just to get the next sequenceid... problematic when WAL goes lame-duck. Lets get it in. TODO: A successful append followed by a failed sync probably only needs us replace the WAL (if we have signalled the client that the appends failed). Bummer is that replicating, these last appends might make it to the sink cluster or get replayed during recovery. HBase should keep its own WAL length? Or sequenceid of last successful sync should be passed when doing recovery and replication?
This change augments existing chaos monkey framework with actions for restarting underlying zookeeper quorum and hdfs nodes of distributed hbase cluster. One assumption made while creating zk actions are that zookeper ensemble is an independent external service and won't be managed by hbase cluster. For these actions to work as expected, the following parameters need to be configured appropriately.
{code} <property> <name>hbase.it.clustermanager.hadoop.home</name> <value>$HADOOP_HOME</value> </property> <property> <name>hbase.it.clustermanager.zookeeper.home</name> <value>$ZOOKEEPER_HOME</value> </property> <property> <name>hbase.it.clustermanager.hbase.user</name> <value>hbase</value> </property> <property> <name>hbase.it.clustermanager.hadoop.hdfs.user</name> <value>hdfs</value> </property> <property> <name>hbase.it.clustermanager.zookeeper.user</name> <value>zookeeper</value> </property> {code}
The service user related configurations are newly introduced since in prod/test environments each service is managed by different user. Once the above parameters are configured properly, you can start using them as needed. An example usage for invoking these new actions is:
{{./hbase org.apache.hadoop.hbase.IntegrationTestAcidGuarantees -m serverAndDependenciesKilling}}
This issue adds boolean parameter, force, to ‘balancer’ command so that admin can force region balancing even when there is region (other than hbase:meta) in transition - assuming RIT being transient. If hbase:meta is in transition, balancer command returns false.
WARNING: For experts only. Forcing a balance may do more damage than repair when assignment is confused Note: enclose the force parameter in double quotes
HConnection could get stuck when talking to a host that went down and then returned. This has been fixed by closing the connection in all paths.
Master/2.0.0 now builds on the latest stable hadoop by default.
Prevent Coprocessors being doubly-loaded; a particular coprocessor can only be loaded once.
Use junit facility to impose timeout on test. Use test category to chose which timeout to apply: small tests timeout after 30 seconds, medium tests after 180 seconds, and large tests after ten minutes.
Updated junit version from 4.11 to 4.12. 4.12 has support for feature used here.
Add this at the head of your junit4 class to add a category-based timeout:
{code} @Rule public final TestRule timeout = CategoryBasedTimeout.builder().withTimeout(this.getClass()). withLookingForStuckThread(true).build(); {code}
For example:
Security fix: Adds protection from clickjacking using X-Frame-Options header. This will prevent use of HBase UI in frames. To disable this feature, set the configuration ‘hbase.http.filter.xframeoptions.mode’ to ‘ALLOW’ (default is ‘DENY’).
Promotes an -ea assert to logged FATAL and RS abort when memstore is found to be in an inconsistent state.
Wraps region, start key, end key columns if too long.
Starting from HBase 2.0, CoprocessorClassLoader will not exempt hadoop classes or zookeeper classes. This means that if the custom coprocessor jar contains hadoop or zookeeper packages and classes, they will be loaded by the CoprocessorClassLoader. Only hbase packages and classes are exempted from the CoprocessorClassLoader. They (and their dependencies) are loaded by the parent server class loader.
In {{checkAndPut}} write path use max(max timestamp for the row, System.currentTimeMillis()) in the, instead of blindly taking System.currentTimeMillis() to ensure that checkAndPut() cannot do writes which is already eclipsed. This is similar to what has been done in HBASE-12449 for increment and append.
A new config, hbase.loadincremental.validate.hfile , is introduced - default to true When set to false, checking hfile format is skipped during bulkloading.
HBCK no longer takes a lock until there are changes to the cluster being made.
The old behavior can be achieved by passing the -exclusive flag.
HBASE-14081 Remove (outdated) references to SVN/trunk from documentation
Increase default hbase.hregion.memstore.block.multiplier from 2 to 4 in the code to match the default value in the config files.
We try to delay the eviction of the block till the cellblocks are formed at the Rpc layer. A simple reference counting mechanism is introduced when ever a block is accessed from the Bucket cache. Once a scanner completes using a block the reference count is decremented. The eviction of the block happens only when the reference count of that block is 0. We also introduce a concept of ShareableMemory based on the type of blocks we create from the Block cache. The blocks from the ByteBufferIOEngine directly refer to the buckets in offheap and such blocks are marked SHARED memory type. The blocks from LRU, HDFS and file mode of Bucket cache are all marked EXCLUSIVE because these blocks have their own exclusive memory. For the CP case, any cell coming out of SHARED memory block is copied before returning the results, because CPs can use the results as its state so that eviction cannot corrupt the results.
The Moderate Object Storage (MOB) feature (HBASE-11339[1]) is modified I/O and compaction path that allows individual moderately sized values (100KB-10MB) to be stored in a way that write amplification is reduced when compared to the normal I/O path. MOB is defined in the column family and it is almost isolated with other components, the features and performance cannot be effected in normal columns.
For more details on how to use the feature please consult the HBase Reference Guide
Removed Table#getRowOrBefore, Region#getClosestRowBefore, Store#getRowKeyAtOrBefore, RemoteHTable#getRowOrBefore apis and Thrift support for getRowOrBefore. Also removed two coprocessor hooks preGetClosestRowBefore and postGetClosestRowBefore. User using this api can instead use reverse scan something like below, {code} Scan scan = new Scan(row); scan.setSmall(true); scan.setCaching(1); scan.setReversed(true); scan.addFamily(family); {code} pass this scan object to the scanner and retrieve the first Result from scanner output.
Change to support offheaping.
Incompatible change for filters ColumnPrefixFilter and MultipleColumnPrefixFilter
Changes parameters to filterColumn so takes a Cell rather than a byte [].
hbase-client-1.2.7-SNAPSHOT.jar, ColumnPrefixFilter.class package org.apache.hadoop.hbase.filter ColumnPrefixFilter.filterColumn ( byte[ ] buffer, int qualifierOffset, int qualifierLength ) : Filter.ReturnCode org/apache/hadoop/hbase/filter/ColumnPrefixFilter.filterColumn:([BII)Lorg/apache/hadoop/hbase/filter/Filter$ReturnCode;
Ditto for filterColumnValue in SingleColumnValueFilter. Takes a Cell instead of byte array.
This changes upgrades thrift dependency of HBase to 0.9.2. Though this doesn't break any HBase compatibility promises, it might impact any downstream projects that share thrift dependency with HBase.
HBase's convenience binary artifact no longer contains the netty 3.2.4 jar . This jar was not directly used by HBase, but may have been relied on by downstream applications.
HBaseTestingUtility now uses the truncate API added in HBASE-8332 so that calls to HBTU.truncateTable will behave like the shell command: effectively dropping the table and recreating a new one with the same split points.
Previously, HBTU.truncateTable instead issued deletes for all the data already in the table. If you wish to maintain the same behavior, you should use the newly added HBTU.deleteTableData method.
The following API from Cell (which were deprecated since past few major versions) are removed now. getRow getFamily getQualifier getValue getMvccVersion The above apis can be replaced with their respective CellUtil#cloneXXX (allocates a copy) or Cell#getXXXArray (essentially just returns a pointer) based on the use case.
HBASE-14029 Correct documentation for Hadoop version specific artifacts
The HBase master status web page no longer allows operators to clone snapshots nor restore snapshots.
When RegionServerCoprocessors throw an exception we will no longer attempt to build an incomplete RPC response message. Instead, the response message will be null.
Tool to sync two tables that tries to send the differences only like rsync.
Adds two new MapReduce jobs, SyncTable and HashTable. See usage for these jobs on how to use. See design doc for generally overview: https://docs.google.com/document/d/1-2c9kJEWNrXf5V4q_wBcoIXfdchN7Pxvxv1IO6PW0-U/edit
From comments below, “It can be challenging to run against a table getting live writes, if those writes are updates/overwrites. In general, you can run it against a time range to ignore new writes, but if those writes update existing cells, then the time range scan may or may not see older versions of those cells depending on whether major compaction has happened, which may be different in remote clusters.”
If the master went to assign a region concurrent with a RegionServer abort, the returned RegionServerAbortedException was being handled as though the region had been cleanly offlined so assign was allowed proceed. If the region was opened in its new location before WAL replay completion, the replayed edits were ignored, worst case, or were later played over the top of edits that had come in since open and so susceptible to overwrite. In either case, DATALOSS.
Adds extra doc on getStartKeys, getEndKeys, and getStartEndKeys in HTable explaining that they will be removed in 2.0.0 (these methods did not get the proper full major version deprecation cycle).
In this issue, we actually also remove these methods in master/2.0.0 branch.
Java 8 is considered supported and tested as of HBase 1.2+
The performance of region splitting has been improved by using a thread pool to split the store files concurrently. Prior to this change, the store files were always split sequentially in a single thread, so a region with multiple store files ended up taking several seconds. The thread pool is sized dynamically with the aim of getting maximum concurrency, without exceeding the number of cores available for HBase Java process. A lower limit for the thread pool can be explicitly set using the property hbase.regionserver.region.split.threads.max.
Exclude Findbugs packages from shaded jars
WARNING: No release note provided for this change.
Fix typo in HBase book
Use the master's timestamp when sending hbase:meta edits on region merge to ensure proper ordering of new region addition and old region deletes.
Correct Javadoc generation errors
This patch adds optional ability for HMaster to normalize regions in size (disabled by default, change hbase.normalizer.enabled property to true to turn it on). If enabled, HMaster periodically (every 30 minutes by default) monitors tables for which normalization is enabled in table configuration and performs splits/merges as seems appropriate. Users may implement their own normalization strategies by implementing RegionNormalizer interface and configuring it in hbase-site.xml.
Use ProtobufMagic methods in ProtobufUtil
In previous versions of HBase, the ReplicationAdmin utility erroneously used the string key “columnFamlyName” when listing replicated column families. It now uses the corrected spelling of “columnFamilyName” (note the added “i”).
Downstream code that parsed the replication entries returned from listReplicated will need to be updated to use the new key. Previously compiled code that relied on the static CFNAME member of ReplicationAdmin will need to be recompiled in order to see the updated value.
By default the Get/Scan will throw Exception when it is not able to find a mob cell because the mob file is missing/corrupted. This jira adds a facility to continue scan/get and get other cells with mob cell value as empty. Set an attribute MobConstants.EMPTY_VALUE_ON_MOBCELL_MISS = true in Scan/Get for getting this behaviour
As per this jira contribution. We now support two kinds of RateLimiter.
org.apache.hadoop.hbase.quotas.AverageIntervalRateLimiter : This limiter will refill resources at every TimeUnit/resources interval. Example: For a limiter configured with 10resources/second, then 1resource will be refilled after every 100ms.
org.apache.hadoop.hbase.quotas.FixedIntervalRateLimiter: This limiter will refill resources only after a given fixed interval of time.
Client can configure anyone of this rate limiter for the cluster by setting the value for the property “hbase.quota.rate.limiter” in the hbase-site.xml. org.apache.hadoop.hbase.quotas.AverageIntervalRateLimiter is the default value. Note: Client needs to restart the cluster for the configuration to take into effect.
hbase-shaded-client and hbase-shaded-server modules will not build the actual jars unless -Prelease is supplied in mvn.
This jira has removed the already deprecated method KeyValue#oswrite(final KeyValue kv, final OutputStream out)
This JIRA modifies the signature of PriorityFunction#getPriority() method to also take request user as a parameter; all RPC requests sent by super users (as determined by cluster configuration) are executed with Admin QoS.
Adds scan metrics to the result. In the shell, set the ALL_METRICS attribute to true on your scan to see dump of metrics after results (see the scan help for examples).
If you would prefer to see only a subset of the metrics, the METRICS array can be defined to include the names of only the metrics you care about.
Added getRegionLocation and getAllRegionLocations to the thrift2 interface.
Purge support for parsing zookeepers zoo.cfg deprecated since hbase-0.96.0
MOTIVATION
A pipelined scan API is introduced for speeding up applications that combine massive data traversal with compute-intensive processing. Traditional HBase scans save network trips through prefetching the data to the client side cache. However, they prefetch synchronously: the fetch request to regionserver is invoked only when the entire cache is consumed. This leads to a stop-and-wait access pattern, in which the client stalls until the next chunk of data is fetched. Applications that do significant processing can benefit from background data prefetching, which eliminates this bottleneck. The pipelined scan implementation overlaps the cache population at the client side with application processing. Namely, it issues a new scan RPC when the iteration retrieves 50% of the cache. If the application processing (that is, the time between invocations of next()) is substantial, the new chunk of data will be available before the previous one is exhausted, and the client will not experience any delay. Ideally, the prefetch and the processing times should be balanced.
API AND CONFIGURATION
Asynchronous scanning can be configured either globally for all tables and scans, or on per-scan basis via a new Scan class API.
Configuration in hbase-site.xml: hbase.client.scanner.async.prefetch, default false:
<property> <name>hbase.client.scanner.async.prefetch</name> <value>true</value> </property>
API - Scan#setAsyncPrefetch(boolean)
Scan scan = new Scan();
scan.setCaching(1000);
scan.setMaxResultSize(BIG\_SIZE);
scan.setAsyncPrefetch(true);
...
ResultScanner scanner = table.getScanner(scan);
IMPLEMENTATION NOTES
Pipelined scan is implemented by a new ClientAsyncPrefetchScanner class, which is fully API-compatible with the synchronous ClientSimpleScanner. ClientAsyncPrefetchScanner is not instantiated in case of small (Scan#setSmall) and reversed (Scan#setReversed) scanners. The application is responsible for setting the prefetch size in a way that the prefetch time and the processing times are balanced. Note that due to double buffering, the client side cache can use twice as much memory as the synchronous scanner.
Generally, this feature will put more load on the server (higher fetch rate -- which is the whole point). Also, YMMV.
Correct setting.xml anchor in book
Introduces a new config hbase.fs.tmp.dir which is a directory in HDFS (or default file system) to use as a staging directory for HFileOutputFormat2. This is also used as the default for hbase.bulkload.staging.dir
From 2.0 branch onwards KVComparator and its subclasses MetaComparator, RawBytesComparator are all deprecated. All the comparators are moved to CellComparator. MetaCellComparator, a subclass of CellComparator, will be used to compare hbase:meta cells.
Previously exposed static instances KeyValue.COMPARATOR, KeyValue.META_COMPARATOR and KeyValue.RAW_COMPARATOR are deprecated instead use CellComparator.COMPARATOR and CellComparator.META_COMPARATOR. Also note that there will be no RawBytesComparator. Where ever we need to compare raw bytes use Bytes.BYTES_RAWCOMPARATOR. CellComparator will always operate on cells and its components, abstracting the fact that a cell can be backed by a single byte[] as opposed to how KVComparators were working.
Adds a renewLease call to ClientScanner
To use the coprocessor-based JMX implementation provided by HBase for Master. Add below property in hbase-site.xml file:
<property> <name>hbase.coprocessor.master.classes</name> <value>org.apache.hadoop.hbase.JMXListener</value> </property>
NOTE: DO NOT set `com.sun.management.jmxremote.port` for Java VM at the same time.
By default, the JMX listens on TCP port 10101 for Master, we can further configure the port using below properties:
The registry port can be shared with connector port in most cases, so you only need to configure master.rmi.registry.port. However if you want to use SSL communication, the 2 ports must be configured to different values.
As we made changes to return types in asynchronous methods of Admin API, this change is going to break binary compatibility. The source compatibility is kept intact though. The applications running against this change needs to be recompiled to keep things working.
HBase now provides added convenience artifacts that shade most dependencies. These jars hbase-shaded-client and hbase-shaded-server are meant to be used when dependency conflicts can not be solved any other way. The normal jars hbase-client and hbase-server should still be preferred when possible.
Do not use hbase-shaded-server or hbase-shaded-client inside of a co-processor as bad things will happen.
In HBase 1.1.0 and above we have upgraded the version of Jackson dependencies (jackson-core-asl, jackson-mapper-asl, jackson-jaxrs and jackson-xc) from 1.8.8 to 1.9.13. This is to follow the upgrade to Jackson 1.9.13 in Hadoop 2.5 and above which causes Jackson class incompatibility for HBase as reported in HBASE-13149. Refer to HADOOP-10104 and YARN-2092 for additional information. Jackson1.9.13 is not completely backward compatible with the prior version 1.8.8 used in HBase. See the Compatibility reports attached in HBASE-13149 and http://svn.codehaus.org/jackson/trunk/release-notes/VERSION for more information.
This upgrade does not have direct impact on HBase users and HBase applications in most cases. In the rare case where your HBase application uses Jackson directly AND your application has compatibility issue with Jackson 1.9.13, you can do the following to mitigate the problem.
Note that it may not be tested or guaranteed that using Jackson 1.8.8 in $HBASE_HOME/lib will work in future HBase releases. It is recommended that your HBase application matches the Jackson version provided in HBase.
In HBase 0.98.x and HBase 1.0.x, we have NOT upgraded the version of Jackson dependencies. If you are on Hadoop 2.5 or above, and your HBase application involves running Yarn jobs, you may encounter Jackson class incomparability issue, as reported in HBASE-13149.
You can do the following to mitigate the problem:
Master now honors configuration options as was before 1.0.0 releases: hbase.master.ipc.address hbase.master.dns.interface hbase.master.dns.nameserver hbase.master.info.bindAddress This jira also adds hbase.master.hostname parameter as an extension to HBASE-12954.
Previously, there was no way to enforce a time limit on scan RPC requests. The server would receive a scan RPC request and take as much time as it needed to accumulate enough results to reach a limit or exhaust the region. The problem with this approach was that, in the case of a very selective scan, the processing of the scan could take too long and cause timeouts client side.
With this fix, the server will now enforce a time limit on the execution of scan RPC requests. When a scan RPC request arrives to the server, a time limit is calculated to be half of whichever timeout value is more restictive between the configurations (“hbase.client.scanner.timeout.period” and “hbase.rpc.timeout”). When the time limit is reached, the server will return whatever results it has accumulated up to that point. The results may be empty.
To ensure that timeout checks do not occur too often (which would hurt the performance of scans), the configuration “hbase.cells.scanned.per.heartbeat.check” has been introduced. This configuration controls how often System.currentTimeMillis() is called to update the progress towards the time limit. Currently, the default value of this configuration value is 10000. Specifying a smaller value will provide a tighter bound on the time limit, but may hurt scan performance due to the higher frequency of calls to System.currentTimeMillis().
Protobuf models for ScanRequest and ScanResponse have been updated so that heartbeat support can be communicated. Support for heartbeat messages is specified in the request sent to the server via ScanRequest.Builder#setClientHandlesHeartbeats. Only when the server sees that ScanRequest#getClientHandlesHeartbeats() is true will it send heartbeat messages back to the client. A response is marked as a heartbeat message via the boolean flag ScanResponse#getHeartbeatMessage
Made methods smaller under Scanner#next so inlinable and compilable (was getting ‘too big to compile’ from hotspot). Use of unsafe to parse shorts rather than use BB#getShort... faster, etc.
In 1.0.x, master by default binds to the region server ports (both rpc and info). This change brings back the usage of old master rpc and info ports in 1.1+ and master (2.0) branches. The motivation for this change is to ease the life of the user so that he does not need to do anything to bring up a RS on the same host and also to make the migration from 0.98 to 1.1 hassle free. However, the users going from 1.0 to 1.1 would see the change in the master ports.
Compose thrift exception text from the text of the entire cause chain of the underlying exception.
Prior to this change the configuration setting ‘hbase.security.authorization’ had no effect if security coprocessor were installed. The act of installing the security coprocessors was assumed to indicate active authorizaton was desired and required. Now it is possible to install the security coprocessors yet have them operate in a passive state with active authorization disabled by setting ‘hbase.security.authorization’ to false. This can be useful but is probably not what you want. For more information, consult the Security section of the HBase online manual.
‘hbase.security.authorization’ defaults to true for backwards comptatible behavior.
Adds a --columns option to PE so you can write more than one column (changes default qualifier from ‘data’ to ‘0’).
Deprecates Result#addResults in favor of Result#setStatistics
This introduces a new config option: hbase.server.scanner.max.result.size This setting enforces a maximum result size (in bytes), when reached the server will return the results is has so far. This is a safety setting and should be kept large. The default is inifinite in 0.98 and 1.0.x and 100mb in 1.1 and later.
Use hbase.client.scanner.max.result.size instead to enforce practical chunk sizes of a few mb (defaults to 2mb)
Results returned from RPC calls may now be returned as partials
When is a Result marked as a partial? When the server must stop the scan because the max size limit has been reached. Means that the LAST Result returned within the ScanResult‘s Result array may be marked as a partial if the scan’s max size limit caused it to stop in the middle of a row.
Incompatible Change: The return type of InternalScanners#next and RegionScanners#nextRaw has been changed to NextState from boolean The previous boolean return value can be accessed via NextState#hasMoreValues() Provides more context as to what happened inside the scanner
Scan caching default has been changed to Integer.Max_Value This value works together with the new maxResultSize value from HBASE-12976 (defaults to 2MB) Results returned from server on basis of size rather than number of rows Provides better use of network since row size varies amongst tables
Protobuf models have changed for Result, ScanRequest, and ScanResponse to support new partial Results
Partial Results should be invisible to application layer unless Scan#setAllowPartials is set
Scan#setAllowPartials has been added to allow the application to request to see the partial Results returned by the server rather than have the ClientScanner form the complete Result prior to returning it to the application
To disable the use of partial Results on the server, set ScanRequest.Builder#setClientHandlesPartials() to be false in the ScanRequest issued to server
Partial Results should allow the server to return large rows in parts rather than accumulate all the cells for that particular row and run out of memory
Enhance WALPrettyPrinter to print information (writer classnames and cell codec classname) from WAL Header
In hbase 1.0.0, 0.98.10, 0.98.10.1, 0.98.11, and 0.98.12 ‘splitSuccessCount’ was misspelled as ‘splitSuccessCounnt’
Removes MetaScanner. Use MetaTableAccessor instead.
Marking as incompatible change. Requires hfiles be major version >= 2 and >= minor version 3. Version 3 files are enabled by default in 1.0. 0.98 writes version 2 minor version 3. You cannot go to 1.0 from anything before 0.98.
For a long time, HBase supported 2 types of connections - managed, which were cached and closed automatically when not needed, and unmanaged, where user is responsible for closing the connections by calling #close() on them.
The concept of managed connections in HBase (deprecated before) has now been extinguished completely, and now all callers are responsible for managing the lifecycle of connections they acquire.
The following config is added by this JIRA:
hbase.regionserver.hostname
This config is for experts: don't set its value unless you really know what you are doing. When set to a non-empty value, this represents the (external facing) hostname for the underlying server. See https://issues.apache.org/jira/browse/HBASE-12954 for details.
Caution: please make sure rolling upgrade succeeds before turning on this feature.
Pass the -D flag generator.multiple.columnfamilies on the command-line if you want the generator to write three column families rather than the default one. When set, we will write the usual ‘meta’ column family and use it checking linked-list is wholesome but we will also write a ‘tiny’ column family and a ‘big’ column family to provoke uneven flushing; good for testing the flush-by-columnfamily feature.
Removed getScannerCaching and setScannerCaching from Table
for 0.98 and 1.0 changes are compatible (due to mitigation by HBASE-13433):
* The “get_counter” command no longer requires a dummy 4th argument. Downstream users are encouraged to migrate code to not pass this argument because it will result in an error for HBase 1.1+. * The “incr” command now outputs the current value of the counter to stdout. ex: {code} jruby-1.6.8 :005 > incr ‘counter_example’, ‘r1’, ‘cf1:foo’, 10 COUNTER VALUE = 1772 0 row(s) in 0.1180 seconds {code}
for 1.1+ changes are incompatible:
* The “get_counter” command no longer accepts a dummy 4th argument. Downstream users will need to update their code to not pass this argument. ex: {code} jruby-1.6.8 :006 > get_counter ‘counter_example’, ‘r1’, ‘cf1:foo’ COUNTER VALUE = 1772
{code} * The “incr” command now outputs the current value of the counter to stdout. ex: {code} jruby-1.6.8 :005 > incr ‘counter_example’, ‘r1’, ‘cf1:foo’, 10 COUNTER VALUE = 1772 0 row(s) in 0.1180 seconds {code}
HBase can use memcached as an external block cache. To use this change your config to set hbase.blockcache.use.external to true and hbase.cache.memcached.servers to contain the list of memcached servers to use.
When issuing an Admin.move command the RegionServer that receive the region will try and open the StoreFiles of that region to prime the block cache with index blocks.
Deprecate said methods. They were mistakenly included in Table Interface.
WARNING: No release note provided for this change.
Fixes an issue where files from a split region that were still referenced were erroneously deleted leading to data loss.
The Result.EMPTY_RESULT object is now immutable. In previous releases, the object could be modified by a caller to no longer be empty. Code that relies on this behavior will now receive an UnsupportedOperationException.
Adds support to add_peer in hbase shell to add a custom replication endpoint from HBASE-12254.
WARNING: No release note provided for this change.
HTable class has been marked as private API before, and now it's no longer directly instantiable from client code (all public constructors have been removed). All clients should use Connection#getTable() and Connection#getRegionLocator() when appropriate to obtain Table and RegionLocator implementations to work with.
WARNING: No release note provided for this change.
hbase.zookeeper.quorum configuration now allows servers together with client ports consistent with the way Zookeeper java client accepts the quorum string. In this case, using hbase.zookeeper.clientPort is not needed. eg. hbase.zookeeper.quorum=myserver1:2181,myserver2:20000,myserver3:31111
Adds buffer reuse sending Cell results. It is on by default and should not need configuration. Improves GC profile and ups throughput. The benefit gets better the larger the row size returned.
The buffer reservoir is bounded at a maximum count after which we will start logging at WARN level that the reservoir is running at capacity (returned buffers will be discarded and not added back to the reservoir pool). Default maximum is twice the handler count: i.e. 2 * hbase.regionserver.handler.count. This should be more than enough. Set the maximum with the new configuration: hbase.ipc.server.reservoir.max
The reservoir will not cache buffers in excess of hbase.ipc.server.reservoir.max.buffer.size The default is 10MB. This means that if a row is very large, then we will allocate a buffer of the average size that is currently in the pool and we will then resize it till we can accommodate the return. These resizes are expensive. The resultant buffer will be used and then discarded.
To check how the reservoir is doing, enable trace level logging for a few seconds on a regionserver. You can do this from the regionserver UI. See ‘Log Level’. Set org.apache.hadoop.hbase.io.BoundedByteBufferPool to TRACE. The BoundedByteBufferPool will spew report to the log. Disable the TRACE level and then check the log. You'll see allocation rate, size of pool, size of buffers in pool, etc.
This adds two new shell commands -- compact_mob and major_compact_mob to the hbase shell.
Run compaction on a mob enabled column family or all mob enabled column families within a table Examples: Compact a column family within a table: hbase> compact_mob ‘t1’, ‘c1’ Compact all mob enabled column families hbase> compact_mob ‘t1’
Run major compaction on a mob enabled column family or all mob enabled column families within a table Examples: Compact a column family within a table: hbase> major_compact_mob ‘t1’, ‘c1’ Compact all mob enabled column families within a table hbase> major_compact_mob ‘t1’
Adds an implementation of ClusterManager to control REST API-managed HBase clusters.
Add a ‘/conf’ link to UI
This change adds two new configuration options:
Change read and write request count in ServerLoad from int to long
Adds counts for various regions states to the table listing on main page. See attached screenshot.
Adds compaction throughput limit mechanism(the word “throttle” is already used when choosing compaction thread pool, so use a different word here to avoid ambiguity). Default is org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController, will limit throughput as follow:
The throttle is on by default in hbase-2.0.0. There is no limit in hbase-1.x.
Corresponding usages for new ScheduledChore vs. Deprecated Chore: Chore.interrupt() -> ScheduledChore.cancel(mayInterruptWhileRunning = true) Threads.setDaemonThreadRunning(Chore) -> ChoreService.scheduleChore(ScheduledChore) Chore.isAlive -> ScheduledChore.isScheduled() Chore.getSleeper().skipSleepCycle() -> ScheduledChore.triggerNow()
On the server side, set hbase.meta.replica.count to the number of replicas of meta that you want to have in the cluster (defaults to 1). hbase.regionserver. meta.storefile.refresh.period should be set to a non-zero number in milliseconds - something like 30000 (defaults to 0). On the client/user side, set hbase.meta.replicas.use to true.
Adds a dev-support/check_compatibility.sh script for comparing versions. Run the script to see usage.
Retrofit a new, netty-based rpc transport on the client. This client is slightly slower if little contention given the extra tier or so that netty adds and that we block on a Future waiting on the call to finish. This client opens the way for HBase having a native Async API.
This client is on by default in master branch (2.0 hbase). It is off in branch-1.0 (hbase-1.1.x). To enable it, set “hbase.rpc.client.impl” to “org.apache.hadoop.hbase.ipc.AsyncRpcClient”
Namespace auditor provides basic quota support for namespaces in terms of number of tables and number of regions. In order to use namespace quotas, quota support must be enabled by setting “hbase.quota.enabled” property to true in hbase-site.xml file.
The users can add quota information to namespace, while creating new namespaces or by altering existing ones.
Examples:
The quotas can be modified/added to namespace at any point of time. To remove quotas, the following command can be used:
alter_namespace ‘ns3’, {METHOD => ‘unset’, NAME => ‘hbase.namespace.quota.maxtables’} alter_namespace ‘ns3’, {METHOD => ‘unset’, NAME => ‘hbase.namespace.quota.maxregions’}
Pushed to master. Shout if there are any issues.
For users on a version of Hadoop that supports tiered storage policies (i.e. Apache Hadoop 2.6.0+), HBase now allows users to opt-in to having the write ahead log placed on the SSD tier. Users on earlier versions of Hadoop will be unable to take advantage of this feature.
Use of tiered storage is controlled by a new RegionServer config, hbase.wal.storage.policy. It defaults to the value ‘NONE’, which will rely on HDFS defaults for a policy decision.
User can specify ONE_SSD or ALL_SSD as the value: ONE_SSD: place only one replica of WAL files in SSD and the remaining in default storage ALL_SSD: all replica for WAL files are placed on SSD
See [the HDFS docs on storage policy|http://hadoop.apache.org/docs/r2.6.0/hadoop-project-dist/hadoop-hdfs/ArchivalStorage.html]
MultiRowRangeFilter is a filter to support scanning multiple row key ranges. If the number of the ranges is small, using multiple scans can also do the same thing and can work well. But when the number of ranges are quite big (e.g. millions), use the MultiRowRangeFilter will be nice. In this filter, the ranges will be sorted and merged, so users do not have to take care of ranges are not continuous. And if users are using something like rest, thrift or pig to access the data the filter might be the practical solution.
Added new option, ROWPREFIXFILTER, to the scan command in the HBase shell to easily scan for a specific row prefix.
CompressionTest will now abort when the target path exists.
Use the -Pjavac maven profile in order to compile HBase using the compiler provided by the JDK instead of the default error-prone compiler plugin. This is useful for now if you are building HBase with JDK 1.8 or a JDK that doesn't support error-prone.
Adds new flushing policy mechanism. Default, org.apache.hadoop.hbase.regionserver.FlushLargeStoresPolicy, will try to avoid flushing out the small column families in a region, those whose memstores are < hbase.hregion.percolumnfamilyflush.size.lower.bound. To restore the old behavior of flushes writing out all column families, set hbase.regionserver.flush.policy to org.apache.hadoop.hbase.regionserver.FlushAllStoresPolicy either in hbase-default.xml or on a per-table basis by setting the policy to use with HTableDescriptor.getFlushPolicyClassName().
updateConfiguration(ServerName server) method of Admin now updates config for HMaster as well. Specifically, config update would be taken by load balancer.
HBase internals for the write ahead log have been refactored. Advanced users of HBase should be aware of the following changes.
Public Audience
LimitedPrivate(COPROC) Audience, LimitedPrivate(PHOENIX)
LimitedPrivate(REPLICATION) Audience
Adds new mob related metrics:
mobCompactedIntoMobCellsCount mobCompactedIntoMobCellsSize mobCompactedFromMobCellsCount mobCompactedFromMobCellsSize mobFlushCount mobFlushedCellsCount mobFlushedCellsSize mobScanCellsCount mobScanCellsSize mobFileCacheAccessCount mobFileCacheMissCount mobFileCacheHitPercent mobFileCacheEvictedCount mobFileCacheCount
Errors from error-prone will fail the build in the compile phase. Warnings look like Javac warnings and are counted as such by test-patch etc
Adds metrics hedgedReads and hedgedReadWins counts.
Adds a script to mark a server as dead.
Usage: considerAsDead.sh --hostname serverName
Removed the below from hbase-2 (were deprecated on release of hbase-1.0.0)
Mutation setWriteToWAL(boolean) boolean getWriteToWAL() Mutation setFamilyMap(NavigableMap<byte [], List<KeyValue>>) NavigableMap<byte [], List<KeyValue>> getFamilyMap()
The below KeyValue based APIs are removed from Result KeyValue[] raw() List<KeyValue> list() List<KeyValue> getColumn(byte [] family, byte [] qualifier) KeyValue getColumnLatest(byte [] family, byte [] qualifier) KeyValue getColumnLatest(byte [] family, int foffset, int flength, byte [] qualifier, int qoffset, int qlength)
They are replaced with Cell[] rawCells() List<Cell> listCells() List<Cell> getColumnCells(byte [] family, byte [] qualifier) Cell getColumnLatestCell(byte [] family, byte [] qualifier) Cell getColumnLatestCell(byte [] family, int foffset, int flength, byte [] qualifier, int qoffset, int qlength) respectively
Also the constructors which were taking KeyValues also removed Result(KeyValue [] cells) Result(List<KeyValue> kvs)
The following APIs are removed from Filter KeyValue transform(KeyValue) KeyValue getNextKeyHint(KeyValue) and replaced with Cell transformCell(Cell) Cell getNextCellHint(Cell) respectively. If a custom Filter implementation have overridden any of these methods, we will no longer call them. User has to change the custom Filter to override cell based methods as shown above
Keeps table enabled/disabled state in HDFS rather than up in ZooKeeper. Auto-migrates any existing zk state.
Adds new test categories besides the class smalltests, mediumtests, and largetests. Adds:
ClientTests CoprocessorTests FilterTests FlakeyTests IOTests MapReduceTests MasterTests MiscTests RegionServerTests ReplicationTests RestTests SecurityTests VerySlowMapReduceTests VerySlowRegionServerTests
See description for examples on how to use them.
Adds a noninteractive mode (-n or --noninteractive) to the hbase shell that exits with a non-zero error code on failed or invalid shell command executions, and exits with a zero error code upon successful execution.
This got committed, so I guess it is safe to resolve it?
By default, we don't use ZK for region assignment now. To fall back to the old way, you can set hbase.assignment.usezk to true.
All MR jobs implement Tool Interface, http://hadoop.apache.org/docs/current/api/org/apache/hadoop/util/Tool.html, so now you can pass properties on command line with the -D flag, etc.
HTablePool was deprecated in 0.98.1 but was still present and usable by apps built against versions before HBase 2.0. It has been moved and is not intended to be used by user applications, and is now an internal part of the thrift2 proxy server only.
Adds --cycles=N argument.
Configure “hbase.display.keys” to false (default: true) in the master/regionservers if the row-keys should be hidden in the webUIs (like in the webUI for table details).
This issue introduces a few new APIs: * HConnectionManager: {code} public static HConnection createConnection(Configuration conf) public static HConnection createConnection(Configuration conf, ExecutorService pool) {code} * HConnection: {code} public HTableInterface getTable(String tableName) throws IOException public HTableInterface getTable(byte[] tableName) throws IOException public HTableInterface getTable(String tableName, ExecutorService pool) throws IOException public HTableInterface getTable(byte[] tableName, ExecutorService pool) throws IOException {code}
By default HConnectionImplementation will create an ExecutorService when needed. The ExecutorService can optionally passed be passed in. HTableInterfaces are retrieved from the HConnection. By default the HConnection's ExecutorService is used, but optionally that can be overridden for each HTable.
Changed defaults:
The Apache ZooKeeper config file zoo.cfg will no longer be read when instantiating a HBaseConfiguration object, as it causes various inconsistency issues. Instead, users have to specify all HBase-relevant ZooKeeper properties in the hbase-site.xml using the various “hbase.zookeeper” prefixed properties. For example, specify “hbase.zookeeper.quorum” to provide a ZK quorum server list.
To enable zoo.cfg reading, for which support may be removed in a future release, set the property “hbase.config.read.zookeeper.config” to true in the hbase-site.xml at the client and servers like so:
<property> <name>hbase.config.read.zookeeper.config</name> <value>true</value> <description> Set to true to allow HBaseConfiguration to read the zoo.cfg file for ZooKeeper properties. Switching this to true is not recommended, since the functionality of reading ZK properties from a zoo.cfg file has been deprecated. </description> </property>