[jira] [Commented] (FLINK-19876) Latency metrics can merge chain task latency metrics to reduce the number of latency metrics

["Matthias (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-19876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17222830#comment-17222830 ] 

Matthias commented on FLINK-19876:
----------------------------------

Hi [~zlzhang0122] for suggesting this improvement. We discussed this feature offline. We see your point that this would mean an improvement from a technical point of view. On the downside, we're not sure whether it's worth the effort as the latency metrics are usually not used in production but rather in debugging contexts as it puts quite some load on the system. Due to this reason we would rather not add more complexity to the code and stick to the current version.

> Latency metrics  can merge chain task latency metrics to reduce the number of latency metrics
> ---------------------------------------------------------------------------------------------
>
>                 Key: FLINK-19876
>                 URL: https://issues.apache.org/jira/browse/FLINK-19876
>             Project: Flink
>          Issue Type: Improvement
>          Components: Runtime / Metrics
>            Reporter: zlzhang0122
>            Priority: Major
>
> In latency metrics report, each operator will report themselves latency metics, but actually the communication between chain operators which in one task is very fast and the latency can be ignored.So maybe we can merge the chain operator latency metrics into one to decrease the number of it.
> For example, if source operator and downstream flat Map operator has been chained, now it will report the latency metrics separately, while we can merge it and make them report their latency metrics only once. 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)