[
https://issues.apache.org/jira/browse/FLINK-21585?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17295853#comment-17295853
]
Chesnay Schepler commented on FLINK-21585:
------------------------------------------
This has come up in the past, but we haven't an implementation plan yet.
Exclusions based on metric names are not really a viable option; they are
rather tedious for the user and need to be updated rigorously on every upgrade
or connector change. The same naturally applies to inclusions.
Another approach might be to define these rules on the logical scope. For
example, you could disable all Kinesis connector metrics by specifying an
exclusion for {{tm.job.task.operator.KinesisConsumer.*}}, or maybe just
{{*.KinesisConsumer.*}}. But now what if you do want one particular Kinesis
metric? Maybe we'd still need inclusions for singular metrics.
Finally, we could think about classifying metrics via internal tags by
semantic, like "availability", "kinesis", "throughput_records" and such. But
that's a fair bit of maintenance/documentation overhead.
> Provide a configurable way to skip metrics
> ------------------------------------------
>
> Key: FLINK-21585
> URL: https://issues.apache.org/jira/browse/FLINK-21585
> Project: Flink
> Issue Type: Wish
> Components: Runtime / Metrics
> Reporter: Ashish Khatkar
> Priority: Minor
>
> I work on Stream Processing team at Yelp. We provide centralized streaming
> infrastructure for Flink applications within our company. We have many Flink
> applications running in their own clusters with multiple applications having
> multiple jobs. Different applications have different usage of Flink metrics
> and not all applications use all the metrics provided by Flink. I was
> wondering if Flink can provide a configurable key for ex:
> {{metrics.reporter.<name>.scope.metrics.excludes}} or
> {{metrics.scope.excludes}} which can take a list of metric names that can be
> excluded by Flink metrics library. This will help us to scrape only the
> relevant metrics at an applications level and not bloat up our metrics
> infrastructure with metrics that are never used. We would like to use this
> with Prometheus Exporter.
> We can solve this at our end by writing a custom metric reporter which can
> provide this functionality. I think community can benefit from such an option
> directly provided by Flink. What do you think?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
