[
https://issues.apache.org/jira/browse/HIVE-10778?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-10778:
------------------------------------
Attachment: HIVE-10778.patch
Simple patch. [~gopalv] that relies on the assumption that these threads are
one-shots and will actually exit, doesn't it? Also how come clearWorkMap
doesn't solve the problem; should we add logging around it to see why?
[~thejas] can you take a look?
1) Is this a good way to detect HS2? I wads thinking of adding a static boolean
set to true in startHiveServer2 when it determines the options are for start;
but it looks like session is also always initialized in init. Would it be
present at all times?
2) Would the compilation threads that access this map exit after every query,
or stick around? In the latter case different fix is needed.
> LLAP: Utilities::gWorkMap needs thread-locals for HiveServer2
> -------------------------------------------------------------
>
> Key: HIVE-10778
> URL: https://issues.apache.org/jira/browse/HIVE-10778
> Project: Hive
> Issue Type: Sub-task
> Components: HiveServer2
> Affects Versions: llap
> Reporter: Gopal V
> Assignee: Sergey Shelukhin
> Fix For: llap
>
> Attachments: HIVE-10778.patch, llap-hs2-heap.png
>
>
> 95% of heap is occupied by the Utilities::gWorkMap in the llap branch HS2.
> !llap-hs2-heap.png!
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
