[
https://issues.apache.org/jira/browse/HIVE-10922?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14571926#comment-14571926
]
Vikram Dixit K commented on HIVE-10922:
---------------------------------------
My only concern would be that we have code like this in other places and it can
cause the same errors in HS2. Can you raise an enhancement jira to fix this
correctly. Either by creating and passing a new fs object for each thread or by
creating UGI objects as user hive when doAs is false and using fs'
closeAllforUGI instead.
+1 for this quick fix.
> In HS2 doAs=false mode, file system related errors in one query causes other
> failures
> -------------------------------------------------------------------------------------
>
> Key: HIVE-10922
> URL: https://issues.apache.org/jira/browse/HIVE-10922
> Project: Hive
> Issue Type: Bug
> Affects Versions: 1.0.0, 1.2.0, 1.1.0
> Reporter: Thejas M Nair
> Assignee: Thejas M Nair
> Attachments: HIVE-10922.1.patch
>
>
> Warehouse class has a few methods that close file system object on errors.
> With doAs=false, since all queries use the same HS2 ugi, the filesystem
> object is shared across queries/threads. When the close on one filesystem
> object gets called, it leads to filesystem object used in other threads also
> get closed and any files registered for deletion on exit also getting deleted.
> There is also no close being done in case of the happy code path.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
