[
https://issues.apache.org/jira/browse/HIVE-13754?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15416690#comment-15416690
]
Lefty Leverenz commented on HIVE-13754:
---------------------------------------
Wow, you're so quick, I didn't have time to put a similar comment on
HIVE-13756. Thanks!
> Fix resource leak in HiveClientCache
> ------------------------------------
>
> Key: HIVE-13754
> URL: https://issues.apache.org/jira/browse/HIVE-13754
> Project: Hive
> Issue Type: Bug
> Components: Clients
> Affects Versions: 1.2.1, 2.0.0
> Reporter: Chris Drome
> Assignee: Chris Drome
> Fix For: 2.2.0
>
> Attachments: HIVE-13754-branch-1.patch, HIVE-13754.1-branch-1.patch,
> HIVE-13754.1.patch, HIVE-13754.patch
>
>
> Found that the {{users}} reference count can go into negative values, which
> prevents {{tearDownIfUnused}} from closing the client connection when called.
> This leads to a build up of clients which have been evicted from the cache,
> are no longer in use, but have not been shutdown.
> GC will eventually call {{finalize}}, which forcibly closes the connection
> and cleans up the client, but I have seen as many as several hundred open
> client connections as a result.
> The main resource for this is caused by RetryingMetaStoreClient, which will
> call {{reconnect}} on acquire, which calls {{close}}. This will decrement
> {{users}} to -1 on the reconnect, then acquire will increase this to 0 while
> using it, and back to -1 when it releases it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
