You are viewing a plain text version of this content. The canonical link for it is here.

Posted to yarn-issues@hadoop.apache.org by "Weiwei Yang (JIRA)" <ji...@apache.org> on 2016/01/22 12:02:39 UTC

[jira] [Commented] (YARN-4621) Job failed if time on one NM is out of sync even other nodes are sync'd

    [ https://issues.apache.org/jira/browse/YARN-4621?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15112301#comment-15112301 ] 

Weiwei Yang commented on YARN-4621:
-----------------------------------

If a NM is desync (system time out of sync with RM .. more than 5 minutes gap ?), then list it as unusable ? 

> Job failed if time on one NM is out of sync even other nodes are sync'd
> -----------------------------------------------------------------------
>
>                 Key: YARN-4621
>                 URL: https://issues.apache.org/jira/browse/YARN-4621
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: nodemanager, resourcemanager
>            Reporter: Weiwei Yang
>            Assignee: Weiwei Yang
>         Attachments: terasort_job_failed.log
>
>
> RM tried allocated more than 10 containers on a NM (on which time is out of sync), they all failed with token expired error, and job eventually failed. We can add a new state to NodeState, e.g DESYNC, if a node is not sync'd with RM, RM then can skip allocating containers on this node.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)