[
https://issues.apache.org/jira/browse/FLINK-22340?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17324739#comment-17324739
]
Xintong Song commented on FLINK-22340:
--------------------------------------
Thanks for the info, [~jiamo].
Despite the reported problem happens in a user defined function, which is not
how {{ThresholdMeter}} is expected to be used, the same problem can also happen
for the runtime internals. Thus, we need to fix this anyway. I've opened a PR
to make {{ThresholdMeter}} thread safe.
In your case, I would suggest to use {{MeterView}}. One of the reasons we make
{{ThresholdMeter}} an internal class is that, it is not optimized for
performance critical scenarios. Notice that a user defined function can be
invoked on every data record, which makes it extremely sensitive to the
performance. {{MeterView}} updates the rate with a background thread, and
reduces the computation overhead by maintaining a history of events count.
However, that means you would need do the threshold checking yourselves.
> Concurrent access to ThresholdMeter may lead to NPE
> ---------------------------------------------------
>
> Key: FLINK-22340
> URL: https://issues.apache.org/jira/browse/FLINK-22340
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
> Affects Versions: 1.13.0
> Reporter: Xintong Song
> Assignee: Xintong Song
> Priority: Critical
> Labels: pull-request-available
> Fix For: 1.13.0
>
>
> It is reported on github [1] that NPE is thrown from ThreasholdMeter, which
> is likely caused by concurrent accesses.
> [1]
> https://github.com/apache/flink/commit/b9e576fb845b817d804da3d68471ff8a4723dcf3#commitcomment-49681105
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
