[
https://issues.apache.org/jira/browse/TEZ-2313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14504214#comment-14504214
]
Siddharth Seth commented on TEZ-2313:
-------------------------------------
If I'm not mistaken, this fixes the issue by ensuring that shuffleInfoEventsMap
only gets populated in case pipelining is enabled ? It'll be good to add a
comment in the obsolete method explaining this.
Otherwise this looks good. +1
One question though - do we need the same obsoletion fail-fast in the non
ordered case (ShuffleManager) ? That would be a separate jira.
> Regression in handling obsolete events in ShuffleScheduler
> ----------------------------------------------------------
>
> Key: TEZ-2313
> URL: https://issues.apache.org/jira/browse/TEZ-2313
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Bikas Saha
> Assignee: Rajesh Balamohan
> Priority: Blocker
> Attachments: TEZ-2313.1.patch, TEZ-2313.WIP.patch
>
>
> /cc [~rohini]
> When an obsolete event is received then the shuffle scheduler fails fast even
> when pipelining is disabled. IIRC, obsolete inputs were supposed to fail the
> shuffled inputs if we were reading and merging partial spilled outputs. But
> in this case, pipelining is not on. So not sure why we are failing fast.
> {noformat}
> Caused by: java.io.IOException: InputAttemptIdentifier
> [inputIdentifier=InputIdentifier [inputIndex=4485], attemptNumber=1,
> pathComponent=null, fetchTypeInfo=FINAL_MERGE_ENABLED, spillEventId=-1] is
> marked as obsoleteInput, but it exists in shuffleInfoEventMap. Some data
> could have been already merged to memory/disk outputs. Failing the fetch
> early.
> at
> org.apache.tez.runtime.library.common.shuffle.orderedgrouped.ShuffleScheduler.obsoleteInput(ShuffleScheduler.java:546)
> at
> org.apache.tez.runtime.library.common.shuffle.orderedgrouped.ShuffleInputEventHandlerOrderedGrouped.processTaskFailedEvent(ShuffleInputEventHandlerOrderedGrouped.java:122)
> at
> org.apache.tez.runtime.library.common.shuffle.orderedgrouped.ShuffleInputEventHandlerOrderedGrouped.handleEvent(ShuffleInputEventHandlerOrderedGrouped.java:73)
> at
> org.apache.tez.runtime.library.common.shuffle.orderedgrouped.ShuffleInputEventHandlerOrderedGrouped.handleEvents(ShuffleInputEventHandlerOrderedGrouped.java:63)
> at
> org.apache.tez.runtime.library.common.shuffle.orderedgrouped.Shuffle.handleEvents(Shuffle.java:246)
> at
> org.apache.tez.runtime.library.input.OrderedGroupedKVInput.handleEvents(OrderedGroupedKVInput.java:265)
> at
> org.apache.tez.runtime.LogicalIOProcessorRuntimeTask.handleEvent(LogicalIOProcessorRuntimeTask.java:620)
> at
> org.apache.tez.runtime.LogicalIOProcessorRuntimeTask.access$1100(LogicalIOProcessorRuntimeTask.java:93)
> at
> org.apache.tez.runtime.LogicalIOProcessorRuntimeTask$1.runInternal(LogicalIOProcessorRuntimeTask.java:683)
> at
> org.apache.tez.common.RunnableWithNdc.run(RunnableWithNdc.java:35){noformat}
> /cc [~rajesh.balamohan]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
