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 "Miklos Szegedi (JIRA)" <ji...@apache.org> on 2017/06/19 22:42:00 UTC
[jira] [Commented] (YARN-6723) NM overallocation based on over-time
rather than snapshot utilization
[ https://issues.apache.org/jira/browse/YARN-6723?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16054885#comment-16054885 ]
Miklos Szegedi commented on YARN-6723:
--------------------------------------
The suggestion is to add a period counter for NM_OVERALLOCATION_GENERAL_THRESHOLD and similar preemption thresholds. This would specify how log the threshold needs to be met so that any action is taken (oversubscription or preemption). This should reduce flakiness especially in case of CPU.
> NM overallocation based on over-time rather than snapshot utilization
> ---------------------------------------------------------------------
>
> Key: YARN-6723
> URL: https://issues.apache.org/jira/browse/YARN-6723
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: yarn
> Affects Versions: 3.0.0-alpha3
> Reporter: Haibo Chen
> Assignee: Haibo Chen
>
> To continue discussion on Miklos's idea in YARN-6670 of
> "Usually the CPU usage fluctuates quite a bit. Do not we need a time period for NM_OVERALLOCATION_GENERAL_THRESHOLD, etc. to avoid allocating on small glitches, even worse preempting in those cases?"
--
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