[
https://issues.apache.org/jira/browse/IMPALA-1270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17155799#comment-17155799
]
Tim Armstrong edited comment on IMPALA-1270 at 7/10/20, 11:51 PM:
------------------------------------------------------------------
I don't think the distinct rewrite actually helps here, per Marcel's point -
the semi join is better as far as I can tell. Maybe we'd be better off
broadcasting. I think the general point about adding the distinct to the
subquery, if it would reduce the cardinality, is worth thinking about so
repurposing the JIRA for that.
was (Author: tarmstrong):
I don't think the distinct rewrite actually helps here, per Marcel's point -
the semi join helps. Maybe we'd be better off broadcasting. I think the general
point about adding the distinct to the subquery is worth thinking about.
> Consider adding distinct aggregation to subqueries as perf optimization
> -----------------------------------------------------------------------
>
> Key: IMPALA-1270
> URL: https://issues.apache.org/jira/browse/IMPALA-1270
> Project: IMPALA
> Issue Type: Task
> Components: Frontend
> Affects Versions: Impala 2.0
> Reporter: Nong Li
> Priority: Minor
> Labels: planner
>
> We should consider other rewrites for exists. For q4, another rewrite is an
> inner join + distinct:
> {code}
> select
> o_orderpriority,
> count(distinct l_orderkey) as order_count
> from lineitem l
> inner join orders o
> on (o.o_orderkey = l.l_orderkey and
> l.l_commitdate < l.l_receiptdate)
> where
> o_orderdate >= '1993-07-01' and
> o_orderdate < '1993-10-01'
> group by
> o_orderpriority
> order by
> o_orderpriority
> {code}
> This can run much faster because we have more flexibility on how we execute
> the inner join. We get killed partitioning lineitem now.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
