[
https://issues.apache.org/jira/browse/HIVE-4501?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13780368#comment-13780368
]
Thejas M Nair commented on HIVE-4501:
-------------------------------------
I will review it today.
Closing the fs is certainly better than just disabling the cache. The benefit
of explicitly closing the FS instead of just letting the GC collect it (by
disabling the cache) seems to be that the deleteOnExit() calls will actually
take effect. But I am not sure if hive code is relying on deleteOnExit() to
happen on the client side. I haven't seen file/file handle leaks in long
running HS2 . I believe the lease checker threads also go away when GC kicks in
(based on my experience with long running HS2 with many queries).
> HS2 memory leak - FileSystem objects in FileSystem.CACHE
> --------------------------------------------------------
>
> Key: HIVE-4501
> URL: https://issues.apache.org/jira/browse/HIVE-4501
> Project: Hive
> Issue Type: Bug
> Components: HiveServer2
> Affects Versions: 0.11.0
> Reporter: Thejas M Nair
> Assignee: Thejas M Nair
> Priority: Critical
> Attachments: HIVE-4501.1.patch, HIVE-4501.1.patch
>
>
> org.apache.hadoop.fs.FileSystem objects are getting accumulated in
> FileSystem.CACHE, with HS2 in unsecure mode.
> As a workaround, it is possible to set fs.hdfs.impl.disable.cache and
> fs.file.impl.disable.cache to true.
> Users should not have to bother with this extra configuration.
> As a workaround disable impersonation by setting hive.server2.enable.doAs to
> false.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
