[
https://issues.apache.org/jira/browse/HIVE-1738?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12924005#action_12924005
]
Siying Dong commented on HIVE-1738:
-----------------------------------
We handle the case for comparing different size because, (for some new codes
checked in by Ning, as far as I remember), if map doesn't produce anything for
group by, it created a dummy record whose length is 0. Just handle lengh=0
sounds strange, so I made a little bit more general to this way.
Comparing from last field is definitely a good idea. It might help a lot in
reduce side for multiple key group-by when there are lots of records to reduce.
I'll open a new JIRA for it.
> Optimize Key Comparison in GroupByOperator
> ------------------------------------------
>
> Key: HIVE-1738
> URL: https://issues.apache.org/jira/browse/HIVE-1738
> Project: Hive
> Issue Type: Improvement
> Reporter: Siying Dong
> Assignee: Siying Dong
> Fix For: 0.7.0
>
> Attachments: HIVE.1738.1.patch, HIVE.1738.2.patch, HIVE.1738.3.patch
>
>
> GroupByOperator uses ObjectInspectorUtils.compare() to compare keys, which is
> written for generalized object comparisons, which is not optimized for
> group-by operator. By optimizing this logic, we expect to see obvious
> improvements in GroupByOperator.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
