[
https://issues.apache.org/jira/browse/YARN-10902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18040729#comment-18040729
]
ASF GitHub Bot commented on YARN-10902:
---------------------------------------
github-actions[bot] closed pull request #3347: YARN-10902. Release reserved
container on blacklist node for an application
URL: https://github.com/apache/hadoop/pull/3347
> Resources on application blacklisted node with reserved container can not
> allocate to other applications
> --------------------------------------------------------------------------------------------------------
>
> Key: YARN-10902
> URL: https://issues.apache.org/jira/browse/YARN-10902
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacityscheduler
> Reporter: jackwangcs
> Assignee: jackwangcs
> Priority: Major
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
> If a node has reserved container of an application and the application adds
> this node into its blacklist, resources on the node can not allocate to other
> applications in the current allocation process.
> In RegularContainerAllocator, if it finds the node is in blacklist, it will
> not allocate resources. Furthermore, this node has a reserved container,
> other queues or application will not have opportunity to allocate.
> {code:java}
> ContainerAllocation tryAllocateOnNode(Resource clusterResource,
> FiCaSchedulerNode node, SchedulingMode schedulingMode,
> ResourceLimits resourceLimits, SchedulerRequestKey schedulerKey,
> RMContainer reservedContainer) {
> ContainerAllocation result;
> // Sanity checks before assigning to this node
> result = checkIfNodeBlackListed(node, schedulerKey);
> if (null != result) {
> return result;
> }
> // ....
> }{code}
> In this case, the reserved container should be cancelled.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]