[
https://issues.apache.org/jira/browse/FLINK-11183?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16725716#comment-16725716
]
Chesnay Schepler commented on FLINK-11183:
------------------------------------------
The graph you showed, is that the memory usage for taskmanagers or jobmanagers?
When you ran on 1.5.3, were you using legacy mode?
No changes have been made to the GraphiteReporter. Regardless, to rule out the
reporter you could try another reporter (like the SLF4jReporter) and see
whether the values fit.
One thing you could try is enabling the [system resources
metrics|https://ci.apache.org/projects/flink/flink-docs-release-1.7/monitoring/metrics.html#system-resources].
It would be interesting to know whether these are different.
> Flink 1.7.0 wrong memory graphite metrics
> -----------------------------------------
>
> Key: FLINK-11183
> URL: https://issues.apache.org/jira/browse/FLINK-11183
> Project: Flink
> Issue Type: Bug
> Affects Versions: 1.7.0
> Reporter: Mario Georgiev
> Priority: Minor
> Attachments: memory.png
>
>
> Hello,
> After upgrading from Flink 1.5.3 to Flink 1.7.0 graphite metrics for memory
> started reporting wrong numbers. All the jobs are reporting the same memory
> used. Off heap memory, direct memory, mapped memory is almost the same across
> all our jobs, for heap memory used all jobs report memory used ~ 30mb? Is
> that even correct, we've had jobs that die sporadically and it appears they
> are OOM-ing, but we can't really see it and verify it because of the metrics.
> Was there any major change to the metrics collection in flink-graphite-1.7.0?
>
> You can see from the screenshot that it appears all our jobs are using same
> memory.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
