[ 
https://issues.apache.org/jira/browse/YARN-9085?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16711709#comment-16711709
 ] 

Anthony Hsu commented on YARN-9085:
-----------------------------------

Looks good overall. A few questions/comments:
 * In CSQueueMetrics.java, I see the comment

{code:java}
//Metrics updated only for "default" partition{code}
How come metrics are not updated for non-default partitions? Are any metrics 
available for non-default partitions?
 * What about GPU metrics? Can CSQueueMetrics collect those, too?
 * Regarding
{noformat}
     if (nodePartition == null) {
       for (String partition : 
Sets.union(queueCapacities.getNodePartitionsSet(),
           queueResourceUsage.getNodePartitionsSet())) {
-        updateUsedCapacity(rc, nlm.getResourceByLabel(partition, cluster),
-            partition, childQueue);
+        updateUsedCapacity(rc, partitionResource, partition, childQueue);
       }
+      updateConfiguredCapacities(rc, partitionResource, childQueue);
     } else {
-      updateUsedCapacity(rc, nlm.getResourceByLabel(nodePartition, cluster),
-          nodePartition, childQueue);
+      updateUsedCapacity(rc, partitionResource, nodePartition, childQueue);
     }
{noformat}
Seems to me the *updateConfiguredCapacities* call you added should be inside 
the for loop and should also take in a *partition* parameter like the 
*updateUsedCapacity* call does. In the future, metrics may be collected for 
non-default partitions as well.
Also, I think the *else* block should also have a *updateConfiguredCapacities* 
call (in case in the future, we collect non-default partition metrics, too.

> Guaranteed and MaxCapacity CSQueueMetrics
> -----------------------------------------
>
>                 Key: YARN-9085
>                 URL: https://issues.apache.org/jira/browse/YARN-9085
>             Project: Hadoop YARN
>          Issue Type: Improvement
>    Affects Versions: 2.9.3
>            Reporter: Jonathan Hung
>            Assignee: Jonathan Hung
>            Priority: Major
>         Attachments: YARN-9085.001.patch
>
>
> Would be useful to have Absolute Capacity/Absolute Max Capacity for queues to 
> compare against allocated/pending/etc.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to