[
https://issues.apache.org/jira/browse/MAPREDUCE-5795?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vinod Kumar Vavilapalli updated MAPREDUCE-5795:
-----------------------------------------------
Target Version/s: 2.4.0
Status: Open (was: Patch Available)
Looked at the patch. Few comments:
- We cannot always mark the final state as FAILED too. It can be any of the
final states. So, instead of passing errorHappenedShutDown to AMStartedEvent,
you should pass forcedState. May be call the argument as
forcedJobStateOnShutDown and then use it depend on whether it is set or
pointing to null.
- Add tests to validate these other possible final states.
- There is no point in storing this extra information forcedJobStateOnShutDown
into the history-file. So changes to Events.apr etc are not needed. We can
localize this change only to the event and the ApplicationMaster.
> Job should be marked as Failed if it is recovered from commit.
> --------------------------------------------------------------
>
> Key: MAPREDUCE-5795
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5795
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Reporter: Yesha Vora
> Assignee: Xuan Gong
> Attachments: MAPREDUCE-5795.1.patch
>
>
> If Resource manager is restarted when a job is in commit state, The job is
> not able to recovered after RM restart and it is marked as Killed.
> The job status should be Failed instead killed.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
