[
https://issues.apache.org/jira/browse/YARN-8784?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16620747#comment-16620747
]
Jason Lowe commented on YARN-8784:
----------------------------------
Thanks for the patch! We should be fine bind-mounting the full and good disk
locations and ignoring the bad disks. After YARN-3591 containers should not
longer be referencing bad disks when launching.
+1 lgtm. I'll commit this later today if there are no objections.
> DockerLinuxContainerRuntime prevents access to distributed cache entries on a
> full disk
> ---------------------------------------------------------------------------------------
>
> Key: YARN-8784
> URL: https://issues.apache.org/jira/browse/YARN-8784
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 3.2.0, 3.1.1
> Reporter: Jason Lowe
> Assignee: Eric Badger
> Priority: Major
> Labels: Docker
> Attachments: YARN-8784.001.patch
>
>
> DockerLinuxContainerRuntime bind mounts the filecache and usercache
> directories into the container to allow tasks to access entries in the
> distributed cache. However it only bind mounts directories on disks that
> are considered good, and disks that are full or bad are not in that list. If
> a container tries to run with a distributed cache entry that has been
> previously localized to a disk that is now considered full/bad, the dist
> cache directory will _not_ be bind-mounted into the container's filesystem
> namespace. At that point any symlinks in the container's current working
> directory that point to those disks will reference invalid paths.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
