blob: 781b76074d5836d638c2f3f44f92d35d4f37019c [file] [log] [blame]
////
/**
*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
////
[[snapshot_scanner]]
== Scan over snapshot
:doctype: book
:numbered:
:toc: left
:icons: font
:experimental:
:toc: left
:source-language: java
In HBase, a scan of a table costs server-side HBase resources reading, formating, and returning data back to the client.
Luckily, HBase provides a TableSnapshotScanner and TableSnapshotInputFormat (introduced by link:https://issues.apache.org/jira/browse/HBASE-8369[HBASE-8369]),
which can scan HBase-written HFiles directly in the HDFS filesystem completely by-passing hbase. This access mode
performs better than going via HBase and can be used with an offline HBase with in-place or exported
snapshot HFiles.
To read HFiles directly, the user must have sufficient permissions to access snapshots or in-place hbase HFiles.
=== TableSnapshotScanner
TableSnapshotScanner provides a means for running a single client-side scan over snapshot files.
When using TableSnapshotScanner, we must specify a temporary directory to copy the snapshot files into.
The client user should have write permissions to this directory, and the dir should not be a subdirectory of
the hbase.rootdir. The scanner deletes the contents of the directory once the scanner is closed.
.Use TableSnapshotScanner
====
[source,java]
----
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory of hbase.rootdir
Scan scan = new Scan();
try (TableSnapshotScanner scanner = new TableSnapshotScanner(conf, restoreDir, snapshotName, scan)) {
Result result = scanner.next();
while (result != null) {
...
result = scanner.next();
}
}
----
====
=== TableSnapshotInputFormat
TableSnapshotInputFormat provides a way to scan over snapshot HFiles in a MapReduce job.
.Use TableSnapshotInputFormat
====
[source,java]
----
Job job = new Job(conf);
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory of hbase.rootdir
Scan scan = new Scan();
TableMapReduceUtil.initTableSnapshotMapperJob(snapshotName, scan, MyTableMapper.class, MyMapKeyOutput.class, MyMapOutputValueWritable.class, job, true, restoreDir);
----
====
=== Permission to access snapshot and data files
Generally, only the HBase owner or the HDFS admin have the permission to access HFiles.
link:https://issues.apache.org/jira/browse/HBASE-18659[HBASE-18659] uses HDFS ACLs to make HBase granted user have permission to access snapshot files.
==== link:https://hadoop.apache.org/docs/r2.7.1/hadoop-project-dist/hadoop-hdfs/HdfsPermissionsGuide.html#ACLs_Access_Control_Lists[HDFS ACLs]
HDFS ACLs supports an "access ACL", which defines the rules to enforce during permission checks, and a "default ACL",
which defines the ACL entries that new child files or sub-directories receive automatically during creation.
Via HDFS ACLs, HBase syncs granted users with read permission to HFiles.
==== Basic idea
The HBase files are organized in the following ways:
* {hbase-rootdir}/.tmp/data/{namespace}/{table}
* {hbase-rootdir}/data/{namespace}/{table}
* {hbase-rootdir}/archive/data/{namespace}/{table}
* {hbase-rootdir}/.hbase-snapshot/{snapshotName}
So the basic idea is to add or remove HDFS ACLs to files of the global/namespace/table directory
when grant or revoke permission to global/namespace/table.
See the design doc in link:https://issues.apache.org/jira/browse/HBASE-18659[HBASE-18659] for more details.
==== Configuration to use this feature
* Firstly, make sure that HDFS ACLs are enabled and umask is set to 027
----
dfs.namenode.acls.enabled = true
fs.permissions.umask-mode = 027
----
* Add master coprocessor, please make sure the SnapshotScannerHDFSAclController is configured after AccessController
----
hbase.coprocessor.master.classes = "org.apache.hadoop.hbase.security.access.AccessController
,org.apache.hadoop.hbase.security.access.SnapshotScannerHDFSAclController"
----
* Enable this feature
----
hbase.acl.sync.to.hdfs.enable=true
----
* Modify table scheme to enable this feature for a specified table, this config is
false by default for every table, this means the HBase granted ACLs will not be synced to HDFS
----
alter 't1', CONFIGURATION => {'hbase.acl.sync.to.hdfs.enable' => 'true'}
----
==== Limitation
There are some limitations for this feature:
=====
If we enable this feature, some master operations such as grant, revoke, snapshot...
(See the design doc for more details) will be slower as we need to sync HDFS ACLs to related hfiles.
=====
=====
HDFS has a config which limits the max ACL entries num for one directory or file:
----
dfs.namenode.acls.max.entries = 32(default value)
----
The 32 entries include four fixed users for each directory or file: owner, group, other, and mask.
For a directory, the four users contain 8 ACL entries(access and default) and for a file, the four
users contain 4 ACL entries(access). This means there are 24 ACL entries left for named users or groups.
Based on this limitation, we can only sync up to 12 HBase granted users' ACLs. This means, if a table
enables this feature, then the total users with table, namespace of this table, global READ permission
should not be greater than 12.
=====
=====
There are some cases that this coprocessor has not handled or could not handle, so the user HDFS ACLs
are not synced normally. It will not make a reference link to another hfile of other tables.
=====