[
https://issues.apache.org/jira/browse/FLINK-23321?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nicolas Raga updated FLINK-23321:
---------------------------------
Description:
Currently
[rest.server.numThreads|https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/config/#rest-server-numthreads]
works to create more DispatcherRestEndpoint threads which help in high usage
of functionality in the purview of job the dispatcher functionality. However,
we cannot tune the number of netty threads use to handle incoming connections.
Especially when using the REST server as a monitoring endpoint, a lack of netty
threads cause connection times and a long lags in responses.
Such configuration already exists for taskmanagers
(*taskmanager.network.netty.client/server.numThreads*) and such implementation
of a *server.netty-boss.numThreads* and *server.netty-worker.numThreads* would
be of great use.
was:Currently
[rest.server.numThreads|https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/config/#rest-server-numthreads]
works to create more DispatcherRestEndpoint threads which help in the purview
of job the dispatcher functionality. However, when used for the REST server for
monitoring of a running job, AbstractRestHandlers actually run on a netty
thread pool we have no control over as seen
[here|https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/rest/RestServerEndpoint.java#L188-L193].
Especially when monitoring unhealthy jobs, we would want to allow creation of
more netty-worker threads that can process incoming connection requests.
> Make REST Netty thread pool count configurable
> ----------------------------------------------
>
> Key: FLINK-23321
> URL: https://issues.apache.org/jira/browse/FLINK-23321
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Metrics, Runtime / REST
> Reporter: Nicolas Raga
> Priority: Minor
> Labels: Monitoring, Netty, REST
> Original Estimate: 1h
> Remaining Estimate: 1h
>
> Currently
> [rest.server.numThreads|https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/config/#rest-server-numthreads]
> works to create more DispatcherRestEndpoint threads which help in high usage
> of functionality in the purview of job the dispatcher functionality. However,
> we cannot tune the number of netty threads use to handle incoming
> connections. Especially when using the REST server as a monitoring endpoint,
> a lack of netty threads cause connection times and a long lags in responses.
>
> Such configuration already exists for taskmanagers
> (*taskmanager.network.netty.client/server.numThreads*) and such
> implementation of a *server.netty-boss.numThreads* and
> *server.netty-worker.numThreads* would be of great use.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
