[
https://issues.apache.org/jira/browse/HIVE-11424?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15197407#comment-15197407
]
Gopal V commented on HIVE-11424:
--------------------------------
Thanks [~jcamachorodriguez] for the patch. This patch allows us to fold
transitive inferences and fix issues.
{code}
sold_date between '2014-01-01' and '2014-02-01' and
store_sk IN (1,2,3) and
(sold_date >= '2014-01-01' and store_sk=1) or (sold_date >= '2014-01-01' and
store_sk=2)
{code}
gets neatly folded by the partition condition remover to
{code}
true and
store_sk IN (1,2,3) and
(true and store_sk=1) or (true and store_sk=2)
{code}
This is currently left in place as
{code}
store_sk IN (1,2,3) and (store_sk=1 and store_sk=2)
{code}
We need to fold that into
{code}
store_sk IN (1,2)
{code}
to get our stats right or map-join conversions can get turned on by artificial
application of the filter twice on the statistics.
This can be done via the DNF expansion of IN & then refolding it back, but it
is much slower to do that than a set intersection (or union for OR) for
unordered IN() expressions.
> Rule to transform OR clauses into IN clauses in CBO
> ---------------------------------------------------
>
> Key: HIVE-11424
> URL: https://issues.apache.org/jira/browse/HIVE-11424
> Project: Hive
> Issue Type: Bug
> Reporter: Jesus Camacho Rodriguez
> Assignee: Jesus Camacho Rodriguez
> Attachments: HIVE-11424.01.patch, HIVE-11424.01.patch,
> HIVE-11424.03.patch, HIVE-11424.03.patch, HIVE-11424.04.patch,
> HIVE-11424.2.patch, HIVE-11424.patch
>
>
> We create a rule that will transform OR clauses into IN clauses (when
> possible).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
