You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Yang Wang (JIRA)" <ji...@apache.org> on 2019/08/12 04:08:00 UTC
[jira] [Commented] (FLINK-13685) YARN deployment hangs when
requested container is above the maximum vcore size
[ https://issues.apache.org/jira/browse/FLINK-13685?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16904830#comment-16904830 ]
Yang Wang commented on FLINK-13685:
-----------------------------------
Hi [~gyfora],
Thanks for creating this ticket. You are right. Currently we only check whether the requested vcores is smaller than the max vcores of each NodeManager in the Yarn cluster. It is not enough when the yarn.scheduler.maximum-allocation-vcores is smaller. The requested vcores should be also checked in AbstractYarnClusterDescriptor#validateClusterResources() like memory.
Will you attach a PR to fix this?
> YARN deployment hangs when requested container is above the maximum vcore size
> ------------------------------------------------------------------------------
>
> Key: FLINK-13685
> URL: https://issues.apache.org/jira/browse/FLINK-13685
> Project: Flink
> Issue Type: New Feature
> Components: Deployment / YARN
> Affects Versions: 1.8.1
> Reporter: Gyula Fora
> Priority: Major
>
> When deploying on YARN flink explicitly checks whether the requested container memory is smaller than yarn.nodemanager.resource.cpu-vcores.
> Unfortunately this doesn't guarantee that the request can be fulfilled if yarn.scheduler.maximum-allocation-vcores is set to a lower number.
> In these cases flink will just keep trying to request a container but will never succeed.
>
>
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)