[
https://issues.apache.org/jira/browse/TEZ-1470?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14169013#comment-14169013
]
Bikas Saha commented on TEZ-1470:
---------------------------------
Speculation is not enabled but the existing code supports speculation because
the state machine transitions were inherited from MR which supported
speculation. The patch for speculation is available on TEZ-14. So it would be
good to not introduce a bug in the recovery code now that would be hit once
TEZ-14 gets committed. No fix would have been needed if the recovery code
followed existing state machine transitions. The successful attempt recovery
would set the successful attempt and subsequent killed attempt recovery would
trigger the RetroActiveKilledTransition that would ignore the killed attempt
because its different from the successful attempt. Since that is not the case,
the recovery code would need a simple fix to check if the killed attempt is the
same as the successful attempt and ignore the killed attempt if that is not the
case.
> Recovery fails due to TaskAttemptFinishedEvent being recorded multiple times
> for the same task
> ----------------------------------------------------------------------------------------------
>
> Key: TEZ-1470
> URL: https://issues.apache.org/jira/browse/TEZ-1470
> Project: Apache Tez
> Issue Type: Sub-task
> Reporter: Jeff Zhang
> Assignee: Jeff Zhang
> Priority: Minor
> Fix For: 0.5.2
>
> Attachments: Tez-1470-2.patch, Tez-1470.patch
>
>
> TaskAttempt can move from SUCCEEDED to KILLED due to node failure. In this
> case TaskAttemptFinishedEvent may been recorded 2 times,and will cause
> failure in recovery.
> {code}
> 14-05-16 08:07:18,386 INFO [main] org.apache.hadoop.service.AbstractService:
> Service org.apache.tez.dag.app.DAGAppMaster failed in state STARTED; cause:
> org.apache.tez.dag.api.TezUncheckedException: Invalid recovery event for
> attempt finished, more completions than starts encountered,
> taskId=task_1400226928057_0001_1_05_000005, finishedAttempts=2,
> incompleteAttempts=-1
> org.apache.tez.dag.api.TezUncheckedException: Invalid recovery event for
> attempt finished, more completions than starts encountered,
> taskId=task_1400226928057_0001_1_05_000005, finishedAttempts=2,
> incompleteAttempts=-1
> at
> org.apache.tez.dag.app.dag.impl.TaskImpl.restoreFromEvent(TaskImpl.java:592)
> at
> org.apache.tez.dag.app.RecoveryParser.parseRecoveryData(RecoveryParser.java:814)
> at
> org.apache.tez.dag.app.DAGAppMaster.recoverDAG(DAGAppMaster.java:1529)
> at
> org.apache.tez.dag.app.DAGAppMaster.serviceStart(DAGAppMaster.java:1558)
> at
> org.apache.hadoop.service.AbstractService.start(AbstractService.java:193)
> at org.apache.tez.dag.app.DAGAppMaster$5.run(DAGAppMaster.java:1957)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:415)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1557)
> at
> org.apache.tez.dag.app.DAGAppMaster.initAndStartAppMaster(DAGAppMaster.java:1953)
> at org.apache.tez.dag.app.DAGAppMaster.main(DAGAppMaster.java:1792)
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
