[
https://issues.apache.org/jira/browse/YARN-5841?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16003745#comment-16003745
]
Haibo Chen commented on YARN-5841:
----------------------------------
NMs maintains a set of registered collectors that they send to RM during
heartbeat. Tracking the code, a collector is added to the set of registered
collectors only if an AM container is launched. Thus, NMs are already reporting
only local collectors to RM. Unless I am missing something, I think this jira
can be closed as not a problem.
> Report only local collectors on node upon resync with RM after RM fails over
> ----------------------------------------------------------------------------
>
> Key: YARN-5841
> URL: https://issues.apache.org/jira/browse/YARN-5841
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Varun Saxena
>
> As per discussion on YARN-3359, we can potentially optimize reporting of
> collectors to RM after RM fails over.
> Currently NM would report all the collectors known to itself in HB after
> resync with RM. This would mean many NMs' may pretty much report similar set
> of collector infos in first NM HB on reconnection.
> This JIRA is to explore how to optimize this flow and if possible, fix it.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
