[
https://issues.apache.org/jira/browse/MAPREDUCE-1227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12780817#action_12780817
]
Dmytro Molkov commented on MAPREDUCE-1227:
------------------------------------------
Well, the current thought is we can sacrifice those tasks that are currently
running as long as they do not get rescheduled until the scheduling is resumed.
This way the job itself wont fail, but the running tasks will of course have to
fail.
When you achieve the 'don't schedule' goal with the task trackers
decommissioning the task trackers will die as disallowed by the jobtracker,
won't they, and thus will have to be restarted in the process of resuming,
which is way more heavyweight than simply stopping giving them new tasks and
then assigning those again.
> Allow JobTracker to pause task scheduling
> -----------------------------------------
>
> Key: MAPREDUCE-1227
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-1227
> Project: Hadoop Map/Reduce
> Issue Type: New Feature
> Affects Versions: 0.22.0
> Reporter: Dmytro Molkov
>
> We want to have an ability to pause task scheduling in JobTracker.
> The idea is: make job tracker still accept new jobs, but delay their running
> and do not schedule any new tasks from the currently running jobs.
> It will help for example restarting the DFS cluster without affecting jobs:
> pause execution, restart the DFS, running tasks will fail, but will not be
> scheduled until the execution is resumed, so the job does not fail.
> In general it should help fix non MR problems (DFS, network, etc.) while not
> failing running jobs and keep accepting new ones.
> What do people think of the general idea?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
