[
https://issues.apache.org/jira/browse/HADOOP-7551?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13126382#comment-13126382
]
Ravi Gummadi commented on HADOOP-7551:
--------------------------------------
>> Therefore if there's a disk failure then every call to get a local path will
>> result in LocalDirAllocator#confChanged doing a disk check of all the
>> configured local dirs.
LocalDirAllocator#AllocatorPerContext#confChanged is actually updating the
savedLocalDirs everytime conf is changed. So the main concern raised in the
above description i.e. every call to confChanged is not resulting into disk
checks (except once per configuration change i.e. once per new bad disk).
> LocalDirAllocator should incorporate LocalStorage
> -------------------------------------------------
>
> Key: HADOOP-7551
> URL: https://issues.apache.org/jira/browse/HADOOP-7551
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs
> Affects Versions: 0.20.204.0
> Reporter: Eli Collins
>
> The o.a.h.fs.LocalDirAllocator is not aware of o.a.h.m.t.LocalStorage
> (introduced in MAPREDUCE-2413) - it always considers the configured local
> dirs, not just the ones that happen to be good. Therefore if there's a disk
> failure then *every* call to get a local path will result in
> LocalDirAllocator#confChanged doing a disk check of *all* the configured
> local dirs. It seems like LocalStorage should be a private class to
> LocalAllocator so that all users of LocalDirAllocator benefit from the disk
> failure handling and all the various users of LocalDirAllocator don't have to
> be modified to handle disk failures. Note that LocalDirAllocator already
> handles faulty directories.
--
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
