[
https://issues.apache.org/jira/browse/PHOENIX-3159?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15411345#comment-15411345
]
Ankit Singhal edited comment on PHOENIX-3159 at 8/8/16 5:52 AM:
----------------------------------------------------------------
Thanks [~devaraj] & [~enis] for the review(done offline).
one comment from [~enis]:-
{code}
In 1.0, of course the whole Connection / HTable lifecycle is changed.
Unfortunately, since Phoenix is still bound by 0.98 semantics, changing these
would be different for Phoenix-HBase-0.98 versus the other branches. One good
thing is that for HBase-2.0, this way of HTable handling / caching is going
away. We should make it so that all of these index writers are sharing the same
connection obtained via the coprocessor and every time the index writers wants
to use HTable, it instantiates one and closes that Table once it is done
instead of pooling / caching. Connection.getTable() is very cheap in 1.1+, and
can be used to instantiate one Table per operation.
{code}
was (Author: an...@apache.org):
Thanks [~devaraj] & [~enis] for the review(done offline).
one comment from [~enis] for documentation:-
{code}
In 1.0, of course the whole Connection / HTable lifecycle is changed.
Unfortunately, since Phoenix is still bound by 0.98 semantics, changing these
would be different for Phoenix-HBase-0.98 versus the other branches. One good
thing is that for HBase-2.0, this way of HTable handling / caching is going
away. We should make it so that all of these index writers are sharing the same
connection obtained via the coprocessor and every time the index writers wants
to use HTable, it instantiates one and closes that Table once it is done
instead of pooling / caching. Connection.getTable() is very cheap in 1.1+, and
can be used to instantiate one Table per operation.
{code}
> 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
>
>
> 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)
