You are viewing a plain text version of this content. The canonical link for it is here.

Posted to mapreduce-issues@hadoop.apache.org by "Jason Lowe (JIRA)" <ji...@apache.org> on 2018/01/25 22:22:00 UTC

[jira] [Assigned] (MAPREDUCE-7033) Map outputs implicitly rely on permissive umask for shuffle

     [ https://issues.apache.org/jira/browse/MAPREDUCE-7033?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jason Lowe reassigned MAPREDUCE-7033:
-------------------------------------

            Assignee: Jason Lowe
            Priority: Critical  (was: Major)
    Target Version/s: 3.1.0, 3.0.1

Raising the priority of this since I recently discovered this long-lingering bug can actually prevent proper shuffling on secure clusters in Hadoop 3.x.

HADOOP-11347 fixed a bug in the local filesystem so it started honoring the configured fs.permissions.umask-mode for local files.  That fix now exposes the lack of explicit permissions in the MapReduce framework for any case where the configured umask is at least 0044.


> Map outputs implicitly rely on permissive umask for shuffle
> -----------------------------------------------------------
>
>                 Key: MAPREDUCE-7033
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-7033
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2
>            Reporter: Jason Lowe
>            Assignee: Jason Lowe
>            Priority: Critical
>
> Map tasks do not explicitly set the permissions of their output files for shuffle.  In a secure cluster the shuffle service is running as a different user than the map task, so the output files require group readability in order to serve up the data during the shuffle phase.  If the user's UNIX umask is too restrictive (e.g.: 077) then the map task's file.out and file.out.index permissions can be too restrictive to allow the shuffle handler to access them.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: mapreduce-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: mapreduce-issues-help@hadoop.apache.org