hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Seth (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-3084) race when KILL_CONTAINER is received for a LOCALIZED container
Date Wed, 12 Oct 2011 06:55:11 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-3084?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13125635#comment-13125635

Siddharth Seth commented on MAPREDUCE-3084:

There's two races - One while ContainersLaunch is localizing files prior to launch, and the
other between generation of the CONTAINER_LAUNCHED event and the actual launch. From an initial
look at the patch, seems like both cases will be handled with the changes (will continue looking).
One case that could be problematic is if there's multiple threads in the AsyncDispatcher -
1 per event type - that could still lead to a race between CE.cancelOrSendSignal and CE.launchContainer
(similar stuff likely exists elsewhere also).
Vinod had an alternate suggestion - by using additional states and moving the START_CONTAINER
localization to an alternate state.
Would it be simpler fixing this by just using interrupts ? (Shell does currently ignore interrupts
in some cases though)
> race when KILL_CONTAINER is received for a LOCALIZED container
> --------------------------------------------------------------
>                 Key: MAPREDUCE-3084
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-3084
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2
>    Affects Versions: 0.23.0
>            Reporter: Siddharth Seth
>            Assignee: Hitesh Shah
>            Priority: Blocker
>         Attachments: MR-3084.1.patch, MR-3084.wip.patch
> Depending on when ContainersLaunch starts a container, {{KILL_CONTAINER}} when container
state is {{LOCALIZED}} ({{LAUNCH_CONTAINER}} event already sent) can end up generating a {{CONTAINER_LAUNCHED}}
event - which isn't handled by ContainerState: {{KILLING}}. Also, the launched container won't
be killed since {{CLEANUP_CONTAINER}} would have already been processed.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message