[
https://issues.apache.org/jira/browse/HADOOP-3055?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12580686#action_12580686
]
Christian Kunz commented on HADOOP-3055:
----------------------------------------
Amareshwari was so kind to check out the situation and found that some
TaskTracker was still running 0.16.0. This TaskTracker was on a node which was
down during upgrade and got revived afterwards.
Would it make sense for JobTracker to refuse connections from TaskTrackers
running an incompatible version?
> missing output for 'successful' jobs
> ------------------------------------
>
> Key: HADOOP-3055
> URL: https://issues.apache.org/jira/browse/HADOOP-3055
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.16.1
> Reporter: Christian Kunz
>
> Same job with 2000 reducers ended apparently successfully, but twice 2 reduce
> output directories (different directories each time) were missing and the
> corresponding temporary _task_xxx directories still existed (these tasks did
> not have speculatively executed companions).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
