[
https://issues.apache.org/jira/browse/HIVE-13194?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15253183#comment-15253183
]
Sergey Shelukhin commented on HIVE-13194:
-----------------------------------------
See HIVE-13002. There's at least one place where HMS client is not threadsafe.
I wanted to do a local fix there but [~ashutoshc] had concerns with it.
We saw the issue with the same callstack even after HIVE-13002, so I am
removing the last place where that object is shared outside of compile's
threadlocals (compile is always single threaded).
If we are sure that it is now thread-safe, perhaps threadlocal should be
removed. Although I'm not sure about perf implications on the synchronized
client shared between threads
Why don't we have a pool of Hive objects?
> Hive object is not thread safe, is shared via a threadlocal and thus should
> not be shared via session - part 2
> --------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-13194
> URL: https://issues.apache.org/jira/browse/HIVE-13194
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-13194.patch
>
>
> Session has session Hive object stored in a field. Given that Hive object is
> not thread safe and is also taken from a threadlocal to start with, that is
> not a good idea.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
