[
https://issues.apache.org/jira/browse/FLINK-9013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16538523#comment-16538523
]
ASF GitHub Bot commented on FLINK-9013:
---------------------------------------
Github user zentol commented on a diff in the pull request:
https://github.com/apache/flink/pull/6294#discussion_r201329612
--- Diff: docs/ops/deployment/yarn_setup.md ---
@@ -132,7 +132,10 @@ Stop the YARN session by stopping the unix process
(using CTRL+C) or by entering
Flink on YARN will only start all requested containers if enough resources
are available on the cluster. Most YARN schedulers account for the requested
memory of the containers,
some account also for the number of vcores. By default, the number of
vcores is equal to the processing slots (`-s`) argument. The
`yarn.containers.vcores` allows overwriting the
-number of vcores with a custom value.
+number of vcores with a custom value. In order for this parameter to be
used your cluster must have CPU scheduling enabled. You can do this e.g. by
+
+ * setting the
`org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler` or
+ * enabling
`org.apache.hadoop.yarn.util.resource.DominantResourceCalculator` for
`org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler`
--- End diff --
I would suggest adding this to the options documentation, and then simply
link to the configuration section. (you can link to individual options!)
> Document yarn.containers.vcores only being effective when adapting YARN config
> ------------------------------------------------------------------------------
>
> Key: FLINK-9013
> URL: https://issues.apache.org/jira/browse/FLINK-9013
> Project: Flink
> Issue Type: Improvement
> Components: Documentation, YARN
> Affects Versions: 1.5.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.5.2
>
>
> Even after specifying {{yarn.containers.vcores}} and having Flink request
> such a container from YARN, it may not take these into account at all and
> return a container with 1 vcore.
> The YARN configuration needs to be adapted to take the vcores into account,
> e.g. by setting the {{FairScheduler}} in {{yarn-site.xml}}:
> {code}
> <property>
> <name>yarn.resourcemanager.scheduler.class</name>
>
> <value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler</value>
> </property>
> {code}
> This fact should be documented at least at the configuration parameter
> documentation of {{yarn.containers.vcores}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
