[MAPREDUCE-6675] TestJobImpl.testUnusableNode failed

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.7.3
• Fix Version/s: 2.8.0, 3.0.0-alpha1
• Component/s: mrv2
• Target Version/s: 2.9.0
• Hadoop Flags: Reviewed
• Assignee: Haibo Chen
• Reporter: Haibo Chen
• Created: 14/Apr/16 23:30
• Updated: 05/Jan/17 01:10
• Resolved: 05/May/16 05:44

Description

TestJobImpl#testUnusableNodeTransition is flaky.

2016-02-13 09:16:42 Running org.apache.hadoop.mapreduce.v2.app.job.impl.TestJobImpl
2016-02-13 09:16:50 Tests run: 17, Failures: 1, Errors: 0, Skipped: 0, Time elapsed: 8.324 sec <<< FAILURE! - in org.apache.hadoop.mapreduce.v2.app.job.impl.TestJobImpl
2016-02-13 09:16:50 testUnusableNodeTransition(org.apache.hadoop.mapreduce.v2.app.job.impl.TestJobImpl) Time elapsed: 5.165 sec <<< FAILURE!
2016-02-13 09:16:50 java.lang.AssertionError: expected:<SUCCEEDED> but was:<ERROR>
2016-02-13 09:16:50 at org.junit.Assert.fail(Assert.java:88)
2016-02-13 09:16:50 at org.junit.Assert.failNotEquals(Assert.java:743)
2016-02-13 09:16:50 at org.junit.Assert.assertEquals(Assert.java:118)
2016-02-13 09:16:50 at org.junit.Assert.assertEquals(Assert.java:144)
2016-02-13 09:16:50 at org.apache.hadoop.mapreduce.v2.app.job.impl.TestJobImpl.assertJobState(TestJobImpl.java:977)
2016-02-13 09:16:50 at org.apache.hadoop.mapreduce.v2.app.job.impl.TestJobImpl.testUnusableNodeTransition(TestJobImpl.java:627)
2016-02-13 09:16:50
2016-02-13 09:16:50
2016-02-13 09:16:50 Results :
2016-02-13 09:16:50
2016-02-13 09:16:50 Failed tests:
2016-02-13 09:16:50 TestJobImpl.testUnusableNodeTransition:627->assertJobState:977 expected:<SUCCEEDED> but was:<ERROR>
2016-02-13 09:16:50
2016-02-13 09:16:50 Tests run: 17, Failures: 1, Errors: 0, Skipped: 0.

Looking at the code, an JobUpdatedNodesEvent is handled by putting an TaskAttemptKill event on the async dispatcher queue and return immediately, but the event might not have been processed by the time all JobTaskEvents events are seen by the job (the jobTaskSucceeded events are handed to Job immediately without going through the dispatcher). Therefore, there is a slight chance that the job will see all three succeeded attempts and transition to Committing state before the taskAttemptKill event is handled by the dispatcher. Committing jobs will reject later JobTaskEvents received, transition to InternalError state and cause the test to fail.