[
https://issues.apache.org/jira/browse/MAPREDUCE-3728?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ahmed Radwan updated MAPREDUCE-3728:
------------------------------------
Attachment: MAPREDUCE-3728.patch
Thanks Roman! I was able to reproduce this issue on a secure cluster.
Here is more explanation of what is happening: For each application, the
$x/usercache/$user/appcache/$appId directory has the setgid permission, the
intention here is to have files and sub-directories created within it to
inherit its group ID (i.e., hadoop), and not the primary group ID of the user
who created the files/sub-directories. However the explicit creation of
$x/usercache/$user/appcache/$appId/output in ContainerLocalizer.java:
{code}
lfs.mkdir(new Path(appBase, OUTPUTDIR), null, false);
{code}
causes the "output" directory to lose the setguid, and accordingly the
attempts' outputs created inside it has the user's group ID, and hence user
"yarn" is now unable to access these map outputs. Note that, yarn has group ID
"hadoop", so if setguid was preserved, then it would have been able to access
the files.
The attached patch addresses this issue, it simply avoids this way of creation
of the output directory, as it is later correctly created when map outputs are
written.
I have tested it by running example pi and wordcount jobs on a single-node
secure cluster, and also on an insecure single-node cluster.
> ShuffleHandler can't access results when configured in a secure mode
> --------------------------------------------------------------------
>
> Key: MAPREDUCE-3728
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-3728
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mrv2, nodemanager
> Affects Versions: 0.23.0
> Reporter: Roman Shaposhnik
> Assignee: Ahmed Radwan
> Priority: Critical
> Fix For: 0.23.1
>
> Attachments: MAPREDUCE-3728.patch
>
>
> While running the simplest of jobs (Pi) on MR2 in a fully secure
> configuration I have noticed that the job was failing on the reduce side with
> the following messages littering the nodemanager logs:
> {noformat}
> 2012-01-19 08:35:32,544 ERROR org.apache.hadoop.mapred.ShuffleHandler:
> Shuffle error
> org.apache.hadoop.util.DiskChecker$DiskErrorException: Could not find
> usercache/rvs/appcache/application_1326928483038_0001/output/attempt_1326928483038_0001_m_000003_0/file.out.index
> in any of the configured local directories
> {noformat}
> While digging further I found out that the permissions on the files/dirs were
> prohibiting nodemanager (running under the user yarn) to access these files:
> {noformat}
> $ ls -l
> /data/3/yarn/usercache/testuser/appcache/application_1327102703969_0001/output/attempt_1327102703969_0001_m_000001_0
> -rw-r----- 1 testuser testuser 28 Jan 20 15:41 file.out
> -rw-r----- 1 testuser testuser 32 Jan 20 15:41 file.out.index
> {noformat}
> Digging even further revealed that the group-sticky bit that was faithfully
> put on all the subdirectories between testuser and
> application_1327102703969_0001 was gone from output and
> attempt_1327102703969_0001_m_000001_0.
> Looking into how these subdirectories are created
> (org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.ContainerLocalizer.initDirs())
> {noformat}
> // $x/usercache/$user/appcache/$appId/filecache
> Path appFileCacheDir = new Path(appBase, FILECACHE);
> appsFileCacheDirs[i] = appFileCacheDir.toString();
> lfs.mkdir(appFileCacheDir, null, false);
> // $x/usercache/$user/appcache/$appId/output
> lfs.mkdir(new Path(appBase, OUTPUTDIR), null, false);
> {noformat}
> Reveals that lfs.mkdir ends up manipulating permissions and thus clears
> sticky bit from output and filecache.
> At this point I'm at a loss about how this is supposed to work. My
> understanding was
> that the whole sequence of events here was predicated on a sticky bit set so
> that daemons running under the user yarn (default group yarn) can have access
> to the resulting files and subdirectories down at output and below. Please let
> me know if I'm missing something or whether this is just a bug that needs to
> be fixed.
> On a related note, when the shuffle side of the Pi job failed the job itself
> didn't.
> It went into the endless loop and only exited when it exhausted all the local
> storage
> for the log files (at which point the nodemanager died and thus the job
> ended). Perhaps
> this is even more serious side effect of this issue that needs to be
> investigated
> separately.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
