[
https://issues.apache.org/jira/browse/MAPREDUCE-4637?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tom White updated MAPREDUCE-4637:
---------------------------------
Attachment: MapReduce.png
Looking at the state transition diagrams (attached) it's possible to kill a
job, task or task attempt in any of their pre-running states: NEW,
INITED/SCHEDULED/(UN)ASSIGNED. I don't see a reason to change that since users
may have a reason to kill a task attempt before it is running. The bug here is
that the unassigned state of the task attempt does not have a
TA_DIAGNOSTICS_UPDATE transition which causes the kill to fail.
I just noticed that job has a JOB_DIAGNOSTIC_UPDATE transition from both the
NEW and INITED states, so perhaps we should extend the patch to add a
TA_DIAGNOSTICS_UPDATE from the NEW state of task attempt too for completeness.
> Killing an unassigned task attempt causes the job to fail
> ---------------------------------------------------------
>
> Key: MAPREDUCE-4637
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-4637
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mrv2
> Affects Versions: 2.0.0-alpha
> Reporter: Tom White
> Assignee: Mayank Bansal
> Attachments: MAPREDUCE-4637-trunk.patch,
> MAPREDUCE-4637-trunk-v2.patch, MapReduce.png
>
>
> Attempting to kill a task attempt that has been scheduled but is not running
> causes an invalid state transition and the AM to stop with an error.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
