[
https://issues.apache.org/jira/browse/HADOOP-5338?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12677918#action_12677918
]
Amar Kamat commented on HADOOP-5338:
------------------------------------
Looks like the problem lies with the _SYNC_ information passed by the
job-tracker to the task-tracker. Upon restart, the job-tracker passes the size
of task completion events that the jobtracker has rebuilt. Events at
job-tracker's comprises of all the tasks (map/reduce/cleanup/setup) while the
one at the tracker comprises of just the map tasks. Will upload a patch soon.
> Reduce tasks are stuck waiting for map outputs when none are in progress
> ------------------------------------------------------------------------
>
> Key: HADOOP-5338
> URL: https://issues.apache.org/jira/browse/HADOOP-5338
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.20.0
> Reporter: Ramya R
> Assignee: Amar Kamat
> Fix For: 0.20.0
>
> Attachments: log.txt
>
>
> When JT is restarted several times, a situation is encountered when the
> reduce tasks are stuck forever waiting for map outputs. However 100%map is
> complete and none of the map tasks are in progress. The reduce tasks wait
> infinitely.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
