[
https://issues.apache.org/jira/browse/YARN-4865?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15210869#comment-15210869
]
Wangda Tan commented on YARN-4865:
----------------------------------
1. CSQueue: getReservedCapacity is not used by anyone, could be removed.
2. FiCaSchedulerApp:
Shouldn't call
{code}
this.queue.decReservedResource(node.getPartition(), resource);
{code}
Inside FiCaSchedulerApp, because SchedulerApplicationAttempt shouldn't modify
queue's state directly.
Instead you can do at LeafQueue#completedContainer:
{code}
// Book-keeping
if (removed) {
// Inform the ordering policy
orderingPolicy.containerReleased(application, rmContainer);
releaseResource(clusterResource, application, container.getResource(),
node.getPartition(), rmContainer, false);
}
{code}
Similarily, instead of call {{queue.incReservedResource}} in
SchedulerApplicationAttempt, we should do this in LeafQueue
3. We need to add sufficient tests for this, since reserved resource will be
used by many downstream modules like UI/REST-API and scheduler internally to
make decisions, following candidates to look at:
- Simple container reserve / simple unreserve /
reserved-and-container-completed (like preempted) /
reserved-and-application-completed (Could be added to TestContainerAllocation)
- Also it will be helpful to add simple tests for container reserve / unreserve
for node partitions. (Could be added to TestContainerAllocationForNodeLabel)
- Some tests could be merged or not necessary needed.
> Track Reserved resources in ResourceUsage and QueueCapacities
> --------------------------------------------------------------
>
> Key: YARN-4865
> URL: https://issues.apache.org/jira/browse/YARN-4865
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Affects Versions: 2.7.2
> Reporter: Sunil G
> Assignee: Sunil G
> Attachments: 0001-YARN-4865.patch
>
>
> As discussed in YARN-4678, capture reserved capacity separately in
> QueueCapcities for better tracking.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
