[
https://issues.apache.org/jira/browse/HIVE-21398?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16792979#comment-16792979
]
Zoltan Haindrich commented on HIVE-21398:
-----------------------------------------
it seems like this have clashed with some other change; pushed an addendum for
it as
[cfaeced79ee1e17f5c7afdc6e728419268dc8546|https://github.com/apache/hive/commit/cfaeced79ee1e17f5c7afdc6e728419268dc8546]
> Columns which has estimated statistics should not be considered as unique keys
> ------------------------------------------------------------------------------
>
> Key: HIVE-21398
> URL: https://issues.apache.org/jira/browse/HIVE-21398
> Project: Hive
> Issue Type: Bug
> Reporter: Zoltan Haindrich
> Assignee: Zoltan Haindrich
> Priority: Major
> Fix For: 4.0.0
>
> Attachments: HIVE-21398.01.patch, HIVE-21398.02.patch
>
>
> Right now for a column to qualify as a unique column it has to meet the
> criteria:
> {code}
> NDV >= numRows
> {code}
> when numRows is 1 this tends to be true ; but numRows is also 1 in cases when
> we are kinda operate in the blind - don't know how many row there are - more
> generatlly: with estimated column statistics.
> As a sideeffect of qualifying all columns to be unique; after a few joins all
> column combinations became unique....so for a join between 3 tables which
> have (i,j,k) columns; then it will allocate {{i*j*k}} triplets of "unique
> column triplets".
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
