[
https://issues.apache.org/jira/browse/HIVE-5348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13782643#comment-13782643
]
Xuefu Zhang commented on HIVE-5348:
-----------------------------------
This seems to be a useful optimization, which should help tables with large
number of partitions of largely the same serde. It seems that the patch also
includes the "optimization" that HIVE-5298 tried to address. With comments
there in HIVE-5298, I'm still not sure whether the optimization is valid. The
question I had is: whether it's possible that a mapoperator can deals multiple
alias with the same partition set. If so, then the optimization become valid
because it only go thru each partition folder once for all alias. If there can
only be one alias, then the number of loops remains the same. (My simple select
query only generates one alias.) An example will be helpful to understand.
Overall, I feel HIVE-3833 introduced certain uncertainty and performance
implications. To optimize, deep understanding of the topic is essential.
> If both table and partition have same serdes, then we don't even need to
> check for OI conversions in ObjectInspectorConverters
> ------------------------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-5348
> URL: https://issues.apache.org/jira/browse/HIVE-5348
> Project: Hive
> Issue Type: Bug
> Reporter: Hari Sankar Sivarama Subramaniyan
> Assignee: Hari Sankar Sivarama Subramaniyan
> Attachments: HIVE-5348.1.patch.txt
>
>
> Follow-up JIRA for HIVE-5202
--
This message was sent by Atlassian JIRA
(v6.1#6144)
