[
https://issues.apache.org/jira/browse/HIVE-6395?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13943635#comment-13943635
]
Xuefu Zhang commented on HIVE-6395:
-----------------------------------
[~szehon] The patch looks good to me. However, just one thing to consider: your
patch would make PPD off in the given case, which has performance implications.
However, if we push down ( state == 1 || state == 2 ) in the test case while
keeping the filter at the filter, wouldn't that work? However, I'm not sure if
this is possible or have covets. What you think?
> multi-table insert from select transform fails if optimize.ppd enabled
> ----------------------------------------------------------------------
>
> Key: HIVE-6395
> URL: https://issues.apache.org/jira/browse/HIVE-6395
> Project: Hive
> Issue Type: Bug
> Components: Query Processor
> Affects Versions: 0.13.0
> Reporter: Szehon Ho
> Assignee: Szehon Ho
> Attachments: HIVE-6395.patch, test.py
>
>
> {noformat}
> set hive.optimize.ppd=true;
> add file ./test.py;
> from (select transform(test.*) using 'python ./test.py'
> as id,name,state from test) t0
> insert overwrite table test2 select * where state=1
> insert overwrite table test3 select * where state=2;
> {noformat}
> In the above example, the select transform returns an extra column, and that
> column is used in where clause of the multi-insert selects. However, if
> optimize is on, the query plan is wrong:
> filter (state=1 and state=2) //impossible
> --> select, insert into test1
> --> select, insert into test2
> The correct query plan for hive.optimize.ppd=false is:
> filter (state=1)
> --> select, insert into test1
> filter (state=2)
> --> select, insert into test2
> For reference
> {noformat}
> create table test (id int, name string)
> create table test2(id int, name string, state int)
> create table test3(id int, name string, state int)
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
