[
https://issues.apache.org/jira/browse/HIVE-9937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14382857#comment-14382857
]
Gopal V commented on HIVE-9937:
-------------------------------
[~mmccline]: Ran a few scale tests last night and there seems to be no visible
issues with the patch from last night.
General comment about asserts - the regular runtime turns off asserts, so you
should be using Preconditions.check operations particularly if it is outside
the core loop (like the futures.size).
Need to re-verify the TODO in VectorAppMasterEventOperator - make sure nothing
in the super.process actually buffers the Object[] row, since now the data is
modified in-place, while earlier it was generating a new array for each row.
This has no safety switch to turn off other than "turn off vectorization", I'd
like to see if [~mmokhtar] can get a full TPC-DS run for this.
With this epic patch, the slowest part of a group-by is now the full-sort,
which gives me something else to fix :)
> LLAP: Vectorized Field-By-Field Serialize / Deserialize to support new
> Vectorized Map Join
> ------------------------------------------------------------------------------------------
>
> Key: HIVE-9937
> URL: https://issues.apache.org/jira/browse/HIVE-9937
> Project: Hive
> Issue Type: Sub-task
> Reporter: Matt McCline
> Assignee: Matt McCline
> Attachments: HIVE-9937.01.patch, HIVE-9937.02.patch,
> HIVE-9937.03.patch, HIVE-9937.04.patch, HIVE-9937.05.patch,
> HIVE-9937.06.patch, HIVE-9937.07.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
