hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Szilard Nemeth (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-7139) FairScheduler: finished applications are always restored to default queue
Date Wed, 17 Jan 2018 13:29:00 GMT

    [ https://issues.apache.org/jira/browse/YARN-7139?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16328760#comment-16328760

Szilard Nemeth commented on YARN-7139:

Checked your changes, I think everything is fine except in this code block I wouldn't mention
that the null check is here only because of the tests are "not good enough".

I guess you spent a considerable amount of time to check how easy would have been to correct
the tests and that's why you decided with the null check in the production code is safer. 
// During tests we do not always have an application object, handle
        // it here but we probably should fix the tests
        if (rmApp != null && rmApp.getApplicationSubmissionContext() != null) {
          // Before we send out the event that the app is accepted is
          // to set the queue in the submissionContext (needed on restore etc)

> FairScheduler: finished applications are always restored to default queue
> -------------------------------------------------------------------------
>                 Key: YARN-7139
>                 URL: https://issues.apache.org/jira/browse/YARN-7139
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: fairscheduler
>    Affects Versions: 2.8.1
>            Reporter: Wilfred Spiegelenburg
>            Assignee: Wilfred Spiegelenburg
>            Priority: Major
>         Attachments: YARN-7139.01.patch, YARN-7139.02.patch, YARN-7139.03.patch, YARN-7139.04.patch
> The queue an application gets submitted to is defined by the placement policy in the
FS. The placement policy returns the queue and the application object is updated. When an
application is stored in the state store the application submission context is used which
has not been updated after the placement rules have run. 
> This means that the original queue from the submission is still stored which is the incorrect
queue. On restore we then read back the wrong queue and display the wrong queue in the RM
web UI.
> We should update the submission context after we have run the placement policies to make
sure that we store the correct queue for the application.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org

View raw message