[jira] [Commented] (HIVE-22492) Amortize lock contention due to LRFU accounting

["Slim Bouguerra (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-22492?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16973569#comment-16973569 ] 

Slim Bouguerra commented on HIVE-22492:
---------------------------------------

{code}
IO-Elevator-Thread-9  Blocked CPU usage on sample: 609ms
  org.apache.hadoop.hive.llap.cache.LowLevelLrfuCachePolicy.notifyUnlock(LlapCacheableBuffer) LowLevelLrfuCachePolicy.java:125
  org.apache.hadoop.hive.llap.cache.CacheContentsTracker.notifyUnlock(LlapCacheableBuffer) CacheContentsTracker.java:173
  org.apache.hadoop.hive.llap.cache.LowLevelCacheImpl.unlockBuffer(LlapDataBuffer, boolean) LowLevelCacheImpl.java:391
  org.apache.hadoop.hive.llap.cache.LowLevelCacheImpl.decRefBuffers(List) LowLevelCacheImpl.java:379
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader.returnData(Reader$OrcEncodedColumnBatch) OrcEncodedDataReader.java:759
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader.returnData(Object) OrcEncodedDataReader.java:110
  org.apache.hadoop.hive.llap.io.decode.EncodedDataConsumer.returnSourceData(EncodedColumnBatch) EncodedDataConsumer.java:100
  org.apache.hadoop.hive.llap.io.decode.EncodedDataConsumer.consumeData(EncodedColumnBatch) EncodedDataConsumer.java:92
  org.apache.hadoop.hive.llap.io.decode.EncodedDataConsumer.consumeData(Object) EncodedDataConsumer.java:34
  org.apache.hadoop.hive.ql.io.orc.encoded.EncodedReaderImpl.readEncodedColumns(int, StripeInformation, OrcProto$RowIndex[], List, List, boolean[], boolean[], Consumer) EncodedReaderImpl.java:532
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader.performDataRead() OrcEncodedDataReader.java:407
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader$4.run() OrcEncodedDataReader.java:266
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader$4.run() OrcEncodedDataReader.java:263
  java.security.AccessController.doPrivileged(PrivilegedExceptionAction, AccessControlContext) AccessController.java (native)
  javax.security.auth.Subject.doAs(Subject, PrivilegedExceptionAction) Subject.java:422
  org.apache.hadoop.security.UserGroupInformation.doAs(PrivilegedExceptionAction) UserGroupInformation.java:1688
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader.callInternal() OrcEncodedDataReader.java:263
  org.apache.hadoop.hive.llap.io.encoded.OrcEncodedDataReader.callInternal() OrcEncodedDataReader.java:110
  org.apache.tez.common.CallableWithNdc.call() CallableWithNdc.java:36
  org.apache.hadoop.hive.llap.daemon.impl.StatsRecordingThreadPool$WrappedCallable.call() StatsRecordingThreadPool.java:110
  java.util.concurrent.FutureTask.run() FutureTask.java:266
  java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor$Worker) ThreadPoolExecutor.java:1142
  java.util.concurrent.ThreadPoolExecutor$Worker.run() ThreadPoolExecutor.java:617
  java.lang.Thread.run() Thread.java:745

{code}

> Amortize lock contention due to LRFU accounting
> -----------------------------------------------
>
>                 Key: HIVE-22492
>                 URL: https://issues.apache.org/jira/browse/HIVE-22492
>             Project: Hive
>          Issue Type: Improvement
>            Reporter: Slim Bouguerra
>            Assignee: Slim Bouguerra
>            Priority: Major
>
> LRFU eviction policy can be a major source of contention under high load.
> This can be see on the following profiles.
> To fix this the idea is to use a batching wrapper to amortize the locking contention.
> The trick i a common way to amortize locking as explained here http://www.ece.eng.wayne.edu/~sjiang/pubs/papers/ding-09-BP-Wrapper.pdf



--
This message was sent by Atlassian Jira
(v8.3.4#803005)