[
https://issues.apache.org/jira/browse/YARN-3216?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14944289#comment-14944289
]
Wangda Tan commented on YARN-3216:
----------------------------------
Thanks [~sunilg].
Went through the patch, some comments:
1. AbstractCSQueue: Instead of adding AM-used-resource to parentQueue, I think
we may only need to calculate AM-used-resource on LeafQueueu and user.
Currently we don't have limitation of AM-used-resource on parentQueue, so the
aggregated resource may not be very useful. We can add it along the hierachy if
we want to limit max-am-percent on parentQueue in the future.
2. CapacitySchedulerConfiguration: Instead of introduce a new configuration:
MAXIMUM_AM_RESOURCE_PARTITION_SUFFIX, I suggest to use the existing one:
maximum-am-resource-percent. If
{{queue.accessible-node-labels.<label>.maximum-am-resource-percent}} not set,
it uses queue.maximum-am-resource-percent. Please let me know if there's any
specific reason to add a new maximum-am-resource-partition.
3. LeafQueue: I'm wondering if we need to maintain map of {{PartitionInfo}}:
PartitionInfo.getActiveApplications is only used to check if there's any
activated apps under a partition, it is equivalent to
{{queueUsage.getAMUsed(partitionName) > 0}}.
4. SchedulerApplicationAttempt: I think return value getAMUsed should be:
- Before AM container allocated, it returns AM-Resource-Request.resource on
partition=AM-Resource-Request.node-label-request
- After AM container allocated, it returns AM-Container.resource on
partition=AM-Node.partition
- you don't have to update am-resource when AM container just allocated,
because AM-container.resource and am-resource-request.node-label-request won't
be changed, but you need to update this if partition of AM-container's NM
updated). I'm not sure if it is clear to you, please let me know if you need
more elaborate about this comment.
I found you removed some code from FiCaSchedulerApp's constructor, I think
getAMUsed should still return correct value before AM container allocated,
otherwise the computation might be wrong. Let me know if I didn't understand
your code correctly.
> Max-AM-Resource-Percentage should respect node labels
> -----------------------------------------------------
>
> Key: YARN-3216
> URL: https://issues.apache.org/jira/browse/YARN-3216
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Wangda Tan
> Assignee: Sunil G
> Priority: Critical
> Attachments: 0001-YARN-3216.patch, 0002-YARN-3216.patch,
> 0003-YARN-3216.patch
>
>
> Currently, max-am-resource-percentage considers default_partition only. When
> a queue can access multiple partitions, we should be able to compute
> max-am-resource-percentage based on that.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
