[
https://issues.apache.org/jira/browse/YARN-8657?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16617342#comment-16617342
]
Weiwei Yang commented on YARN-8657:
-----------------------------------
[~leftnoteasy], [~sunilg]
Sorry I just got time to this.
The patch looks fine. My only concern is the performance.
Since computeUserLimitAndSetHeadroom was called in both scheduling-phase and
commit-phase, so even the user limit was stale a bit in the scheduling-phase,
the check put in the commit-phase should be able to guarantee there is no
violations. If we add this sync block, it may reduce the performance as a
side-effect. If I remember correctly, when we were analyzing the conflicts a
few months ago, this was not one of the top-3 cause of conflicts.
BTW, the patch no longer applies to trunk.
Thanks
> User limit calculation should be read-lock-protected within LeafQueue
> ---------------------------------------------------------------------
>
> Key: YARN-8657
> URL: https://issues.apache.org/jira/browse/YARN-8657
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler
> Reporter: Sumana Sathish
> Assignee: Wangda Tan
> Priority: Critical
> Attachments: YARN-8657.001.patch
>
>
> When async scheduling is enabled, user limit calculation could be wrong:
> It is possible that scheduler calculated a user_limit, but inside
> {{canAssignToUser}} it becomes staled.
> We need to protect user limit calculation.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
