[
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_v4.patch
Thank you [~devaraj] for the feedback.
bq. 1. Shouldn't the "pool" in CachingHTableFactory not be shut down if the
pool was initially obtained via getTable(tableName, pool)?
Actually, we are not taking initial pool from getTable(tableName, pool), we are
creating it and passing across. so we need to shutdown during regionserver
shutdown hook.
bq. 2. For the case you create the pool, should the maxThreads be simply
Integer.MAX_VALUE? Assuming that all table accesses go through this
CachingHTableFactory, it should be okay (no leaks and such)? Saying this
because for the case the pool is passed from outside, it seems it gets created
with Integer.MAX_VALUE.
Agreed, default pool created by HTable is also setting it to Integer.MAX_VALUE,
so have also made the change to set default max threads to Integer.MAX_VALUE
in our pool. Although it can be controlled by "phoenix.index.writes.threads.max"
bq. 3. After the pool is shutdown, wondering if you should be doing a
pool.awaitTermination() or else the threads will be killed if the JVM is
exiting (this should be verified though; I am recalling this from memory).
Wondering if this (threads did partial work and then exited) actually matters
in practice or not for Phoenix.
Agreed, Updated the patch to wait for termination of threads.
> 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: 4.9.0
>
> Attachments: PHOENIX-3159.patch, PHOENIX-3159_v1.patch,
> PHOENIX-3159_v2.patch, PHOENIX-3159_v3.patch, PHOENIX-3159_v4.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)
