[
https://issues.apache.org/jira/browse/YARN-275?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13566786#comment-13566786
]
Siddharth Seth commented on YARN-275:
-------------------------------------
With the RM side changes in place - I'm hopeful that this jira (NM heartbeat
interval decided by RM) will not be requried - at least for the issue reported
in YARN-270. The criteria used by the current patch to decide when an NM should
heartbeat pretty much goes away after the RM side changes.
That said, I agree that we could run into other situations where such
throttling may be useful. I'd like to defer this jira till after the RM side
changes are made. 1) the criteria for NM heartbeat will be more clear, 2)NM
back-off can delay scheduling - unless out of band heartbeats are also added.
Scheduler perf does need to be looked at - I believe there may be some
optimizations, at least on the CS, which can improve performance, and reduce
lock contention.
I'll create a separate jira for the RM side fixes.
> Make NodeManagers to NOT blindly heartbeat irrespective of whether previous
> heartbeat is processed or not.
> ----------------------------------------------------------------------------------------------------------
>
> Key: YARN-275
> URL: https://issues.apache.org/jira/browse/YARN-275
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: nodemanager, resourcemanager
> Reporter: Vinod Kumar Vavilapalli
> Assignee: Xuan Gong
> Attachments: Prototype.txt, YARN-270.1.patch
>
>
> We need NMs to back off. The event handler mechanism is very scalable but not
> infinitely so :)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
