[jira] Updated: (HADOOP-5460) Job recovery should fail or kill a job that fails ACL checks upon restart, if the job was running previously

["Amar Kamat (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HADOOP-5460?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Amar Kamat updated HADOOP-5460:
-------------------------------

    Attachment: HADOOP-5460-v1.0.patch

I think a job should not be failed or killed upon acls change on restart. Ideally whatever jobs are recovered should be allowed to continue. Attaching a patch that allows job to continue even if acls change across restart. Also jobs that fail in recovery are added to the system and then *failed*. Testing in progress.

> Job recovery should fail or kill a job that fails ACL checks upon restart, if the job was running previously
> ------------------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-5460
>                 URL: https://issues.apache.org/jira/browse/HADOOP-5460
>             Project: Hadoop Core
>          Issue Type: Improvement
>          Components: mapred
>            Reporter: Hemanth Yamijala
>         Attachments: HADOOP-5460-v1.0.patch, HADOOP-5460-v1.0.patch
>
>
> Consider a scenario where a job was submitted to the M/R system and runs for a while. Then say the JT is restarted, and before that the ACLs for the user are changed so that that user can no longer submit jobs to that queue. Since the job could potentially be using resources alloted to that queue and could be account for it, this might lead to accounting inconsistencies. A suggestion is for the jobtracker to fail / kill this job.

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