Accumulo supports building a set of sample data that can be efficiently accessed by scanners. What data is included in the sample set is configurable. Below, some data representing documents are inserted.
root@instance sampex> createtable sampex root@instance sampex> insert 9255 doc content 'abcde' root@instance sampex> insert 9255 doc url file://foo.txt root@instance sampex> insert 8934 doc content 'accumulo scales' root@instance sampex> insert 8934 doc url file://accumulo_notes.txt root@instance sampex> insert 2317 doc content 'milk, eggs, bread, parmigiano-reggiano' root@instance sampex> insert 2317 doc url file://groceries/9.txt root@instance sampex> insert 3900 doc content 'EC2 ate my homework' root@instance sampex> insert 3900 doc uril file://final_project.txt
Below the table sampex is configured to build a sample set. The configuration causes Accumulo to include any row where murmur3_32(row) % 3 ==0
in the tables sample data.
root@instance sampex> config -t sampex -s table.sampler.opt.hasher=murmur3_32 root@instance sampex> config -t sampex -s table.sampler.opt.modulus=3 root@instance sampex> config -t sampex -s table.sampler=org.apache.accumulo.core.client.sample.RowSampler
Below, attempting to scan the sample returns an error. This is because data was inserted before the sample set was configured.
root@instance sampex> scan --sample 2015-09-09 12:21:50,643 [shell.Shell] ERROR: org.apache.accumulo.core.client.SampleNotPresentException: Table sampex(ID:2) does not have sampling configured or built
To remedy this problem, the following command will flush in memory data and compact any files that do not contain the correct sample data.
root@instance sampex> compact -t sampex --sf-no-sample
After the compaction, the sample scan works.
root@instance sampex> scan --sample 2317 doc:content [] milk, eggs, bread, parmigiano-reggiano 2317 doc:url [] file://groceries/9.txt
The commands below show that updates to data in the sample are seen when scanning the sample.
root@instance sampex> insert 2317 doc content 'milk, eggs, bread, parmigiano-reggiano, butter' root@instance sampex> scan --sample 2317 doc:content [] milk, eggs, bread, parmigiano-reggiano, butter 2317 doc:url [] file://groceries/9.txt
Inorder to make scanning the sample fast, sample data is partitioned as data is written to Accumulo. This means if the sample configuration is changed, that data written previously is partitioned using a different criteria. Accumulo will detect this situation and fail sample scans. The commands below show this failure and fixiing the problem with a compaction.
root@instance sampex> config -t sampex -s table.sampler.opt.modulus=2 root@instance sampex> scan --sample 2015-09-09 12:22:51,058 [shell.Shell] ERROR: org.apache.accumulo.core.client.SampleNotPresentException: Table sampex(ID:2) does not have sampling configured or built root@instance sampex> compact -t sampex --sf-no-sample 2015-09-09 12:23:07,242 [shell.Shell] INFO : Compaction of table sampex started for given range root@instance sampex> scan --sample 2317 doc:content [] milk, eggs, bread, parmigiano-reggiano 2317 doc:url [] file://groceries/9.txt 3900 doc:content [] EC2 ate my homework 3900 doc:uril [] file://final_project.txt 9255 doc:content [] abcde 9255 doc:url [] file://foo.txt
The example above is replicated in a java program using the Accumulo API. Below is the program name and the command to run it.
./bin/runex sample.SampleExample
The commands below look under the hood to give some insight into how this feature works. The commands determine what files the sampex table is using.
root@instance sampex> tables -l accumulo.metadata => !0 accumulo.replication => +rep accumulo.root => +r sampex => 2 trace => 1 root@instance sampex> scan -t accumulo.metadata -c file -b 2 -e 2< 2< file:hdfs://localhost:10000/accumulo/tables/2/default_tablet/A000000s.rf [] 702,8
Below shows running accumulo rfile-info
on the file above. This shows the rfile has a normal default locality group and a sample default locality group. The output also shows the configuration used to create the sample locality group. The sample configuration within a rfile must match the tables sample configuration for sample scan to work.
$ accumulo rfile-info hdfs://localhost:10000/accumulo/tables/2/default_tablet/A000000s.rf Reading file: hdfs://localhost:10000/accumulo/tables/2/default_tablet/A000000s.rf RFile Version : 8 Locality group : <DEFAULT> Start block : 0 Num blocks : 1 Index level 0 : 35 bytes 1 blocks First key : 2317 doc:content [] 1437672014986 false Last key : 9255 doc:url [] 1437672014875 false Num entries : 8 Column families : [doc] Sample Configuration : Sampler class : org.apache.accumulo.core.client.sample.RowSampler Sampler options : {hasher=murmur3_32, modulus=2} Sample Locality group : <DEFAULT> Start block : 0 Num blocks : 1 Index level 0 : 36 bytes 1 blocks First key : 2317 doc:content [] 1437672014986 false Last key : 9255 doc:url [] 1437672014875 false Num entries : 6 Column families : [doc] Meta block : BCFile.index Raw size : 4 bytes Compressed size : 12 bytes Compression type : gz Meta block : RFile.index Raw size : 309 bytes Compressed size : 176 bytes Compression type : gz
The [shard example][shard] shows how to index and search files using Accumulo. That example indexes documents into a table named shard
. The indexing scheme used in that example places the document name in the column qualifier. A useful sample of this indexing scheme should contain all data for any document in the sample. To accomplish this, the following commands build a sample for the shard table based on the column qualifier.
root@instance shard> config -t shard -s table.sampler.opt.hasher=murmur3_32 root@instance shard> config -t shard -s table.sampler.opt.modulus=101 root@instance shard> config -t shard -s table.sampler.opt.qualifier=true root@instance shard> config -t shard -s table.sampler=org.apache.accumulo.core.client.sample.RowColumnSampler root@instance shard> compact -t shard --sf-no-sample -w 2015-07-23 15:00:09,280 [shell.Shell] INFO : Compacting table ... 2015-07-23 15:00:10,134 [shell.Shell] INFO : Compaction of table shard completed for given range
After enabling sampling, the command below counts the number of documents in the sample containing the words import
and int
.
$ ./bin/runex shard.Query --sample -t shard import int | fgrep '.java' | wc 11 11 1246
The command below counts the total number of documents containing the words import
and int
.
$ ./bin/runex shard.Query -t shard import int | fgrep '.java' | wc 1085 1085 118175
The counts 11 out of 1085 total are around what would be expected for a modulus of 101. Querying the sample first provides a quick way to estimate how much data the real query will bring back.
Another way sample data could be used with the shard example is with a specialized iterator. In the examples source code there is an iterator named CutoffIntersectingIterator. This iterator first checks how many documents are found in the sample data. If too many documents are found in the sample data, then it returns nothing. Otherwise it proceeds to query the full data set. To experiment with this iterator, use the following command. The --sampleCutoff
option below will cause the query to return nothing if based on the sample it appears a query would return more than 1000 documents.
$ ./bin/runex shard.Query --sampleCutoff 1000 -t shard import int | fgrep '.java' | wc