[
https://issues.apache.org/jira/browse/YARN-3938?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14631652#comment-14631652
]
Wangda Tan commented on YARN-3938:
----------------------------------
Hi [~bibinchundatt],
Thanks for reporting this issue, this is a known issue of node label. Possible
solutions:
# Make {{maxAMResource = queue's-total-guaranteed-resource (Sum of queue's
guaranteed resource on all partitions) * maxAmResourcePercent}}. It will be
straightforward, but also can lead to too many AMs launched under a single
partition.
# Make maxAMResource computed per queue per partition, this can make AM usages
under partitions are more balanced, but can also lead to hard debugging (My
application get stuck because of AMResourceLimit for a partition is violated).
I prefer 1st solution since it's easier to understand and debug.
Thoughts?
And could I take over this issue if you haven't get started?
> AM Resources for leaf queues zero when DEFAULT PARTITION resource is zero
> with NodeLabel
> ----------------------------------------------------------------------------------------
>
> Key: YARN-3938
> URL: https://issues.apache.org/jira/browse/YARN-3938
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Reporter: Bibin A Chundatt
> Assignee: Bibin A Chundatt
> Priority: Critical
> Attachments: Am limit for subqueue.jpg
>
>
> In case of leaf queue the AM resource calculation is based on
> {{absoluteCapacityResource}}. Below is the calculation for absolute capacity
> {{LeafQueue#updateAbsoluteCapacityResource()}}
> {code}
> private void updateAbsoluteCapacityResource(Resource clusterResource) {
> absoluteCapacityResource =
> Resources.multiplyAndNormalizeUp(resourceCalculator, labelManager
> .getResourceByLabel(RMNodeLabelsManager.NO_LABEL,
> clusterResource),
> queueCapacities.getAbsoluteCapacity(), minimumAllocation);
> }
> {code}
> If default partition resource is zero for all Leaf queue the resource for AM
> will be zero
> Snapshot also attached for the same
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
