Zhijie Shen commented on YARN-3034:

bq. then only way RMTimelineCollector can be invoked is through 
SystemMetricsPublisher's (SMP) public methods

Oh, probably I misunderstood your intention. I used to think this is way that 
you want to do put the data into RMTimelineCollector. So in this case, we could 
put RMTimelineCollector inside SystemMetricsPublisher, and whereas we invoke 
timeline client, we call RMTimelineCollector for v2.

According to this comments, it seems that you want to create a separate stack 
to put entities into RMTimelineCollector, right? If so, the current design 
makes sense.

bq. So in NM side too we req a configuration but we cannot use the existing one

I meant we keep {{yarn.resourcemanager.system-metrics-publisher.enabled}} for 
v1 SystemMetricsPublisher. For v2,  both RM and NM reads 
{{yarn.system-metrics-publisher.enabled}}? No need to have v1/v2 flag?

> [Collector wireup] Implement RM starting its timeline collector
> ---------------------------------------------------------------
>                 Key: YARN-3034
>                 URL: https://issues.apache.org/jira/browse/YARN-3034
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: timelineserver
>            Reporter: Sangjin Lee
>            Assignee: Naganarasimha G R
>         Attachments: YARN-3034-20150312-1.patch, YARN-3034.20150205-1.patch, 
> YARN-3034.20150316-1.patch, YARN-3034.20150318-1.patch, 
> YARN-3034.20150320-1.patch
> Per design in YARN-2928, implement resource managers starting their own ATS 
> writers.

This message was sent by Atlassian JIRA

Reply via email to