[
https://issues.apache.org/jira/browse/YARN-5700?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15546852#comment-15546852
]
Eric Badger commented on YARN-5700:
-----------------------------------
Looks like there are 2 bugs here.
1) TestAMRestart uses {{waitForState()}} to wait for the completed container.
However, this checks for the liveContainers list. Once the container is
completed, it will quickly be taken out of this list. I think we can instead
use {{waitForContainerToComplete()}} to check for the last set of finished
containers. Something like this:
{noformat}
- rm1.waitForState(nm1, containerId2, RMContainerState.RUNNING);
+ NMContainerStatus completedContainer =
+ TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 2,
+ ContainerState.COMPLETE);
+ rm1.waitForContainerToComplete(app1.getCurrentAppAttempt(),
completedContainer);
{noformat}
2) YARN-4807 changed {{waitForState()}} in MockRM.java so that it quietly
returns false on failure instead of throwing an exception. In 2.8 and below,
the code would call an {{assertNotNull()}} on the container to make sure that
it wasn't null and throw an exception if it was. Since 2.9+ quietly returns
false instead of throwing an exception, the test waits for the timeout and then
continues with the test once {{waitForState()}} returns (even though it
returned false). We could fix the test to check for a false return value, but
there are most likely other tests that also depend on {{waitForState()}}
throwing an exception on failure instead of checking the return value. So I
would think that it'd be better to put the {{assertNotNull()}} back in.
[~kasha], [~yufeigu] (reporter/assignee from YARN-4807), what do you think
about adding the {{assertNotNull()}} back into {{waitForState()}}
> testAMRestartNotLostContainerCompleteMsg times out intermittently in 2.8
> ------------------------------------------------------------------------
>
> Key: YARN-5700
> URL: https://issues.apache.org/jira/browse/YARN-5700
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Eric Badger
> Assignee: Eric Badger
>
> {noformat}
> java.lang.Exception: test timed out after 30000 milliseconds
> at java.lang.Thread.sleep(Native Method)
> at
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.waitForState(MockRM.java:301)
> at
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.waitForState(MockRM.java:286)
> at
> org.apache.hadoop.yarn.server.resourcemanager.MockRM.waitForState(MockRM.java:281)
> at
> org.apache.hadoop.yarn.server.resourcemanager.applicationsmanager.TestAMRestart.testAMRestartNotLostContainerCompleteMsg(TestAMRestart.java:774)
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
