[ 
https://issues.apache.org/jira/browse/YARN-5994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15857866#comment-15857866
 ] 

Sunil G edited comment on YARN-5994 at 2/8/17 11:51 AM:
--------------------------------------------------------

Thanks [~ebadger]

Ideally this make sense. Since this test is only caring about application 
activation, we dont need an NM to be registered. However for internal cluster 
resource and other data  structures, we need some resource in cluster. 

For eg: *scheduler.getMaximumResourceCapability()* is calculated as
{{Math.min(configuredMaxAllocation.getMemorySize(), maxNodeMemory)}}, we 
consider 8GB (max-allocation-mb) or max-memory across all nodes. 
Here max NM memory is 2GB. Hence {{scheduler.getMaximumResourceCapability()}} 
will become 2GB. This will cause 
*RMAppManager#validateAndCreateResourceRequest* to fail the app as AM resource 
requests is for 3GB which is greater than 
{{scheduler.getMaximumResourceCapability()}}.

So bringing NM memory to 4G is fine. [~bibinchundatt], any comments on this?


was (Author: sunilg):
Thanks [~ebadger]

Ideally this make sense. Since this test is only caring about application 
activation, we dont need an NM to be registered. However for internal cluster 
resource and other data  structures, we need some resource in cluster. For eg:
{{Math.min(configuredMaxAllocation.getMemorySize(), maxNodeMemory)}}, we 
consider 8GB (max-allocation-mb) or max-memory across all nodes. 

So bringing NM memory to 4G is fine. [~bibinchundatt], any comments on this?

> TestCapacityScheduler.testAMLimitUsage fails intermittently
> -----------------------------------------------------------
>
>                 Key: YARN-5994
>                 URL: https://issues.apache.org/jira/browse/YARN-5994
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Eric Badger
>            Assignee: Eric Badger
>         Attachments: 
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestCapacityScheduler-output.txt,
>  YARN-5994.001.patch
>
>
> {noformat}
> java.lang.AssertionError: app shouldn't be null
>       at org.junit.Assert.fail(Assert.java:88)
>       at org.junit.Assert.assertTrue(Assert.java:41)
>       at org.junit.Assert.assertNotNull(Assert.java:621)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.waitForState(MockRM.java:169)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.submitApp(MockRM.java:577)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.submitApp(MockRM.java:488)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.submitApp(MockRM.java:395)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestCapacityScheduler.verifyAMLimitForLeafQueue(TestCapacityScheduler.java:3389)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestCapacityScheduler.testAMLimitUsage(TestCapacityScheduler.java:3251)
> {noformat}



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to