hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jian He (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-6872) Ensure apps could run given NodeLabels are disabled post RM switchover/restart
Date Fri, 28 Jul 2017 21:51:00 GMT

    [ https://issues.apache.org/jira/browse/YARN-6872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16105743#comment-16105743
] 

Jian He commented on YARN-6872:
-------------------------------

isRecovery flag is already passed into SchedulerUtils#normalizeAndValidateRequest, I think
we can use that flag directly ?

And this block of code can be removed now ?
{code}
    // If null amReq has been returned, check if it is the case that
    // application has specified node label expression while node label
    // has been disabled. Reject the recovery of this application if it
    // is true and give clear message so that user can react properly.
    if (!appContext.getUnmanagedAM() &&
        (application.getAMResourceRequests() == null ||
            application.getAMResourceRequests().isEmpty()) &&
        !YarnConfiguration.areNodeLabelsEnabled(this.conf)) {
      // check application submission context and see if am resource request
      // or application itself contains any node label expression.
      List<ResourceRequest> amReqsFromAppContext =
          appContext.getAMContainerResourceRequests();
      String labelExp =
          (amReqsFromAppContext != null && !amReqsFromAppContext.isEmpty()) ?
          amReqsFromAppContext.get(0).getNodeLabelExpression() : null;
      if (labelExp == null) {
        labelExp = appContext.getNodeLabelExpression();
      }
      if (labelExp != null &&
          !labelExp.equals(RMNodeLabelsManager.NO_LABEL)) {
        String message = "Application recovered " + appId
            + ". NodeLabel is not enabled in cluster, but AM resource request "
            + "contains a label expression. Consider for NO_LABEL.";
        LOG.warn(message);
      }
    }
{code}
Did you verify that the labeled resource will be counted as non-labeled resource after RM
restart with node label disabled?

> Ensure apps could run given NodeLabels are disabled post RM switchover/restart
> ------------------------------------------------------------------------------
>
>                 Key: YARN-6872
>                 URL: https://issues.apache.org/jira/browse/YARN-6872
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: resourcemanager
>            Reporter: Sunil G
>            Assignee: Sunil G
>         Attachments: YARN-6872.001.patch
>
>
> Post YARN-6031, few apps could be failed during recovery provided they had some label
requirements for AM and labels were disable post RM restart/switchover. As discussed in YARN-6031,
its better to run such apps as it may be long running apps as well.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

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


Mime
View raw message