[
https://issues.apache.org/jira/browse/YARN-3044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14522790#comment-14522790
]
Vinod Kumar Vavilapalli commented on YARN-3044:
-----------------------------------------------
Apologies for dropping off the "send info from RM vs from NM" discussion mid
way through.
We all agree that sending information from NMs is *more scalable*.
The concern isn't really about information ownership. RM and NM both form the
platform, so we can rely on NMs to publish information. But it's really about
potential *loss of information* in many not-so rare cases like when container
gets allocated but gets preempted or released by AM before it really starts.
As long as containers successfully start on NMs (which will be the vast
majority assuming the cluster isn't bad), we can rely on NMs to post all sorts
of information - allocation time, wait time, execution time, information like
priority, host, port , resource-usage-over-time etc. We can just tunnel some of
the RM-originated information through AMs to the NM.
The missing dots occur when a container's life-cycle ends either on the RM or
the AM. We can take a dual pronged approach here? That or we make the
RM-publisher itself a distributed push.
> [Event producers] Implement RM writing app lifecycle events to ATS
> ------------------------------------------------------------------
>
> Key: YARN-3044
> URL: https://issues.apache.org/jira/browse/YARN-3044
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Sangjin Lee
> Assignee: Naganarasimha G R
> Attachments: YARN-3044-YARN-2928.004.patch,
> YARN-3044.20150325-1.patch, YARN-3044.20150406-1.patch,
> YARN-3044.20150416-1.patch
>
>
> Per design in YARN-2928, implement RM writing app lifecycle events to ATS.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
