[jira] Commented: (HADOOP-1472) Timed-out tasks are marked as 'KILLED' rather than as 'FAILED' which means the framework doesn't fail a TIP with 4 or more timed-out attempts

["Arun C Murthy (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-1472?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12502493 ] 

Arun C Murthy commented on HADOOP-1472:
---------------------------------------

Forgot to mention:

Thanks to Alejandro and his team for pointing this one...

> Timed-out tasks are marked as 'KILLED' rather than as 'FAILED' which means the framework doesn't fail a TIP with 4 or more timed-out attempts
> ---------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-1472
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1472
>             Project: Hadoop
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.13.0
>            Reporter: Arun C Murthy
>            Assignee: Arun C Murthy
>            Priority: Blocker
>             Fix For: 0.14.0
>
>
> Timed-out tasks (and also tasks which fail with {{FSError}}) are marked as {{KILLED}} rather than as {{FAILED}}. The major issue with this is that post HADOOP-1050 only {{FAILED}} task-attempts are considered to decide if the {{TIP}} has failed, and hence there exists a corner case where a {{TIP}} which has 4 timed-out tasks isn't marked as {{FAILED}} and thus the job keeps running too...
> Considering this is a corner-case and is going to entail not-too-insignificant changes to {{TaskTracker}}'s control-flow (ugly as it is right now), I'm proposing to fix this either for 0.13.1 (if need be) or better: 0.14.
> Thoughts?

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.