[
https://issues.apache.org/jira/browse/FLINK-10917?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16701916#comment-16701916
]
Dongwon Kim commented on FLINK-10917:
-------------------------------------
Hi [~kkl0u],
Thanks a lot for watching the issue :-)
I've been doing some tests on it by bumping up the version to 4.0.3.
It works okay for more than a week without any back pressure.
FYI, I'm using Prometheus for collecting metrics from Flink.
> Bump up the version of dropwizard metrics-core to 3.2.3 or higher
> -----------------------------------------------------------------
>
> Key: FLINK-10917
> URL: https://issues.apache.org/jira/browse/FLINK-10917
> Project: Flink
> Issue Type: Improvement
> Components: Metrics
> Affects Versions: 1.6.2
> Reporter: Dongwon Kim
> Priority: Minor
>
> I've experienced back pressure once in a while from a streaming pipeline of
> mine [1].
> I strongly suspect SlidingTimeWindowReservoir from
> io.dropwizard.metrics:metrics-core:3.1.5.
> It is known to cause long GCs [2] so a new implementation called
> SlidingTimeWindowArrayReservoir is introduced in v3.2.3.
> So I suggest to bump up dropwizard's metrics-core to v3.2.3 or higher to use
> the new implementation to prevent back pressure which actually has nothing to
> do with Flink itself.
> I just tested compatibility very simply by importing
> io.dropwizard.metrics:metrics-core:4.0.3 in my own project in order to shadow
> v3.1.5 which is introduced by flink-metrics-dropwizard.
> It works without any incompatibility issues for me; there was no
> NoSuchMethodError or something.
> However, I'm not sure whether bumping up to 3.2.x or 4.x is okay for other
> users.
> [1]
> https://www.slideshare.net/ssuser6bb12d/realtime-driving-score-service-using-flink/30
> [2] https://github.com/dropwizard/metrics/pull/1139
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
