[
https://issues.apache.org/jira/browse/HADOOP-3940?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Douglas updated HADOOP-3940:
----------------------------------
Attachment: 3940-1.patch
Added some comments to waitForDataToMerge, particularly explaining why
numRequiredMapOutputs == 0 is a special case, per Arun's suggestion.
Test failures are due to HADOOP-3946.
> Reduce often attempts in memory merge with no work
> --------------------------------------------------
>
> Key: HADOOP-3940
> URL: https://issues.apache.org/jira/browse/HADOOP-3940
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Reporter: Chris Douglas
> Priority: Minor
> Attachments: 3940-0.patch, 3940-1.patch
>
>
> ReduceTask.ReduceCopier.ShuffleRamManager initializes numRequiredMapOutputs
> to 0, so one of the predicates in ShuffleRamManager::waitForDataToMerge,
> {{numPendingRequests < numRequiredMapOutputs}}, is false until the first map
> output is fetched and false again after the last map output is fetched. The
> InMemFSMergeThread thread will loop busily in this state.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
