[
https://issues.apache.org/jira/browse/MAPREDUCE-1475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Amareshwari Sriramadasu updated MAPREDUCE-1475:
-----------------------------------------------
Attachment: tt-logs.txt
Attaching TaskTracker logs for the above scenario
> Race condition while launching task cleanup attempt.
> ----------------------------------------------------
>
> Key: MAPREDUCE-1475
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-1475
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: tasktracker
> Affects Versions: 0.20.1
> Reporter: Amareshwari Sriramadasu
> Attachments: tt-logs.txt
>
>
> We found a race condition while launching task cleanup attempt on a
> TaskTracker which would eat up a slot.
> The scenario is the following:
> The main attempt is killed by TaskTracker because it was a speculative
> attempt. Cleanup attempt is launched on the same tracker. Cleanup attempt
> occupied the slot and is about to start. But, there was a pending RPC: done()
> from earlier attempt in the RPC queue. Before the cleanup attempt could be
> launched, TaskTracker processed the rpc from earlier attempt and made the
> state of the cleanup attempt as KILLED. Launcher did not launch it because it
> was already KILLED. But, the rpc done() failed with NullPointerException
> because of false state. In summary, the slot was occupied by the cleanup
> attempt which could not be launched. And the slot was never released.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
