[
https://issues.apache.org/jira/browse/PHOENIX-3159?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ankit Singhal updated PHOENIX-3159:
-----------------------------------
Attachment: PHOENIX-3159_v2.patch
[~devaraj],
bq. If the above can happen, it should be prevented...
it was handled in the patch because we were synchronizing on openTables(cache)
in both the methods removeLRU() and getTable(), so there will not be any
conflict between them.
Although during testing, found a thread leak in
CoprocessorHost#getTable(TableName tableName) api (HBASE-16389), Have fixed
that in the latest patch.
> CachingHTableFactory may close HTable during eviction even if it is getting
> used for writing by another thread.
> ---------------------------------------------------------------------------------------------------------------
>
> Key: PHOENIX-3159
> URL: https://issues.apache.org/jira/browse/PHOENIX-3159
> Project: Phoenix
> Issue Type: Bug
> Reporter: Ankit Singhal
> Assignee: Ankit Singhal
> Fix For: 4.8.1
>
> Attachments: PHOENIX-3159.patch, PHOENIX-3159_v1.patch,
> PHOENIX-3159_v2.patch
>
>
> CachingHTableFactory may close HTable during eviction even if it is getting
> used for writing by another thread which results in writing thread to fail
> and index is disabled.
> LRU eviction closing HTable or underlying connection when cache is full and
> new HTable is requested.
> {code}
> 2016-08-04 13:45:21,109 DEBUG
> [nat-s11-4-ioss-phoenix-1-5.openstacklocal,16020,1470297472814-index-writer--pool11-t35]
> client.ConnectionManager$HConnectionImplementation: Closing HConnection
> (debugging purposes only)
> java.lang.Exception
> at
> org.apache.hadoop.hbase.client.ConnectionManager$HConnectionImplementation.internalClose(ConnectionManager.java:2423)
> at
> org.apache.hadoop.hbase.client.ConnectionManager$HConnectionImplementation.close(ConnectionManager.java:2447)
> at
> org.apache.hadoop.hbase.client.CoprocessorHConnection.close(CoprocessorHConnection.java:41)
> at
> org.apache.hadoop.hbase.client.HTableWrapper.internalClose(HTableWrapper.java:91)
> at
> org.apache.hadoop.hbase.client.HTableWrapper.close(HTableWrapper.java:107)
> at
> org.apache.phoenix.hbase.index.table.CachingHTableFactory$HTableInterfaceLRUMap.removeLRU(CachingHTableFactory.java:61)
> at
> org.apache.commons.collections.map.LRUMap.addMapping(LRUMap.java:256)
> at
> org.apache.commons.collections.map.AbstractHashedMap.put(AbstractHashedMap.java:284)
> at
> org.apache.phoenix.hbase.index.table.CachingHTableFactory.getTable(CachingHTableFactory.java:100)
> at
> org.apache.phoenix.hbase.index.write.ParallelWriterIndexCommitter$1.call(ParallelWriterIndexCommitter.java:160)
> at
> org.apache.phoenix.hbase.index.write.ParallelWriterIndexCommitter$1.call(ParallelWriterIndexCommitter.java:136)
> at java.util.concurrent.FutureTask.run(FutureTask.java:266)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
> at java.lang.Thread.run(Thread.java:745)
> {code}
> But the IndexWriter was using this old connection to write to the table which
> was closed during LRU eviction
> {code}
> 016-08-04 13:44:59,553 ERROR [htable-pool659-t1] client.AsyncProcess: Cannot
> get replica 0 location for
> {"totalColumns":1,"row":"\\xC7\\x03\\x04\\x06X\\x1C)\\x00\\x80\\x07\\xB0X","families":{"0":[{"qualifier":"_0","vlen":2,"tag":[],"timestamp":1470318296425}]}}
> java.io.IOException: hconnection-0x21f468be closed
> at
> org.apache.hadoop.hbase.client.ConnectionManager$HConnectionImplementation.locateRegion(ConnectionManager.java:1153)
> at
> org.apache.hadoop.hbase.client.CoprocessorHConnection.locateRegion(CoprocessorHConnection.java:41)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl.findAllLocationsOrFail(AsyncProcess.java:949)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl.groupAndSendMultiAction(AsyncProcess.java:866)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl.resubmit(AsyncProcess.java:1195)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl.receiveGlobalFailure(AsyncProcess.java:1162)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl.access$1100(AsyncProcess.java:584)
> at
> org.apache.hadoop.hbase.client.AsyncProcess$AsyncRequestFutureImpl$SingleServerRequestRunnable.run(AsyncProcess.java:727)
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> at java.util.concurrent.FutureTask.run(FutureTask.java:266)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
> at java.lang.Thread.run(Thread.java:745)
> {code}
> Although the workaround is to the cache size(index.tablefactory.cache.size).
> But still we should handle the closing of working HTables to avoid index
> write failures (which in turn disables index).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
