[
https://issues.apache.org/jira/browse/YARN-8270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16479169#comment-16479169
]
Haibo Chen commented on YARN-8270:
----------------------------------
Thanks [~Sushil-K-S] for the patch! I don't know much about the metric system
used here, so have a few questions about the patch.
1) Right now, all timelinecollectors run with NM process and
TimelineCollectorMetrics is a singleton. Does that mean the collector metric is
an aggregate of all collectors on the same node?
2) Does it make sense to have finer granularity of the getEntity/getEntities()
metrics? That is, one metric for one type of entities?
BTW, I think the metrics are meant to be singletons, so their constructors need
to be made private. We could also the set**Metrics() methods to
update**Metrics() since they are really doing aggregation rather than updating
snapshots.
> Adding JMX Metrics for Timeline Collector and Reader
> ----------------------------------------------------
>
> Key: YARN-8270
> URL: https://issues.apache.org/jira/browse/YARN-8270
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: ATSv2, timelineserver
> Reporter: Sushil Kumar S
> Assignee: Sushil Kumar S
> Priority: Major
> Attachments: YARN-8270.001.patch
>
>
> This Jira is for emitting JMX Metrics for ATS v2 Timeline Collector and
> Timeline Reader, basically for Timeline Collector it tries to capture
> success, failure and latencies for *putEntities* and *putEntitiesAsync* from
> *TimelineCollectorWebService* and all the API's success, failure and
> latencies for fetching TimelineEntities from *TimelineReaderWebServices*.
> This would actually help in monitoring and measuring performance for ATSv2 at
> scale.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
