[
https://issues.apache.org/jira/browse/HADOOP-5374?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12677958#action_12677958
]
Vinod K V commented on HADOOP-5374:
-----------------------------------
Here's the exact scenario:
- A TT gets a task but fails to report back immediately, making the JT mark
the task as FAILED.
- The TT comes back with the task in COMMIT_PENDING state.
- JT doesn't update the task's state on itself as task cannot go from FAILED
to COMMIT_PENDING state. (See state checks in TIP.updateStatus()). Thus
TIP.doCommit() is not called.
- In getTasksToSave(), JT just checks the task's status on TT and tries to
save the task, thus hitting NPE in TIP.shouldCommit().
Effect : The TT will keep getting NPE in heartbeat and become useless. The job
will proceed as a duplicate task will be launched for this TIP.
> NPE in JobTracker.getTasksToSave() method
> -----------------------------------------
>
> Key: HADOOP-5374
> URL: https://issues.apache.org/jira/browse/HADOOP-5374
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.20.0
> Reporter: Vinod K V
>
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
