[
https://issues.apache.org/jira/browse/HIVE-26075?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17517981#comment-17517981
]
liuguanghua commented on HIVE-26075:
------------------------------------
I have tested this problem is reproduction on hive version 1.2.2 . But the
version 2.3.3 does not have the problem. The Master doesn't have tested.
So I will push a PR on version 1.2.2.
> hive metastore connection leaking when hiveserver2 kerberos enable and
> hive.server2.enable.doAs set to true
> ------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-26075
> URL: https://issues.apache.org/jira/browse/HIVE-26075
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Affects Versions: 1.2.0
> Reporter: liuguanghua
> Assignee: liuguanghua
> Priority: Major
> Attachments: HIVE-26075.patch
>
>
> (1)When hadoop cluster kerberos is enable
> (2) HiveServer2 config hive.server2.enable.doAs is set true
> After a beeline scripte has been executed, hivemetastore connection is
> created are in ESTABLISHED state and never closed.
> If we submit a lot of task to hiveserver2 ,this will result in hive metastore
> thrift thread(default is 1000) full ,thus new task will fail.
>
> HiveServer2 use ThreadLocal<Hive> to store multithreading metastore
> connection,the application should call Hive.closeCurrent() to close
> connection after task finished.
>
> When HiveServer2 impersonate is enable (hive.server2.enable.doAs is set
> true), the ugi will create proxy user via
> UserGroupInformation.createProxyUser(
> owner, UserGroupInformation.getLoginUser()),the old metastore client is never
> closed.
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
