[
https://issues.apache.org/jira/browse/HIVE-28490?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17877927#comment-17877927
]
Sungwoo Park commented on HIVE-28490:
-------------------------------------
On 10TB TPC-DS benchmark (tested with Hive 4 on MR3),
query 58, before: 72.9s, after: 9.6s
query 83, before: 18.3s, after: 14.6s
> SharedWorkOptimizer sometimes removes useful DPP sources.
> ---------------------------------------------------------
>
> Key: HIVE-28490
> URL: https://issues.apache.org/jira/browse/HIVE-28490
> Project: Hive
> Issue Type: Improvement
> Reporter: Seonggon Namgung
> Assignee: Seonggon Namgung
> Priority: Major
> Attachments: 3.StopRemovingRetainableDPP.pptx
>
>
> Current SharedWorkOptimizer sometimes removes DPP sources that are not
> invalidated. I found that findAscendantWorkOperators() returns a super set of
> ascendant operators, which causes wrong DPP source removal.
> Please check out the attached slides for detailed explanation.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
