[
https://issues.apache.org/jira/browse/FLINK-29801?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17636421#comment-17636421
]
Hang Ruan commented on FLINK-29801:
-----------------------------------
Hi, [~MengYue] ,
For the question 1, what I means is better to pass the metric group in the
`OperatorCoordinator.Context` instead of using it by invoking the method
`registeMetric` provided in the PR.
For the question 2, it seems that the implementations of `OperatorCoordinator`
need to distinguish the metric group name by self, which may need to be done by
Flink.
IMO, I think we need a new metric group for the `OperatorCoordinator`. And we
need the FLIP for the change and a discuss it in the community.
This needs to spend some time in the discussion it in the community.
> OperatorCoordinator need open the way to operate metricGroup interface
> ----------------------------------------------------------------------
>
> Key: FLINK-29801
> URL: https://issues.apache.org/jira/browse/FLINK-29801
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Metrics
> Reporter: yuemeng
> Assignee: yuemeng
> Priority: Major
> Labels: pull-request-available
>
> Currently, We have no way to get metric group instances in OperatorCoordinator
> In some cases, we may report some metric in OperatorCoordinator such as Flink
> hudi integrate scene， some meta will send to operator coordinator to commit
> to hdfs or hms
> but we also need to report some metrics in operator coordinator for monitor
> purpose
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
