c21 commented on a change in pull request #35216:
URL: https://github.com/apache/spark/pull/35216#discussion_r787253434
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
##########
@@ -684,7 +684,9 @@ object LimitPushDown extends Rule[LogicalPlan] {
left = maybePushLocalLimit(limitExpr, join.left),
right = maybePushLocalLimit(limitExpr, join.right))
case LeftSemi | LeftAnti if join.condition.isEmpty =>
- join.copy(left = maybePushLocalLimit(limitExpr, join.left))
+ join.copy(
+ left = maybePushLocalLimit(limitExpr, join.left),
+ right = maybePushLocalLimit(Literal(1, IntegerType), join.right))
Review comment:
Actually I think this PR is orthogonal to what we have did during join
execution - https://github.com/apache/spark/pull/34247. During join execution,
the optimization was to keep one row per key in hash table of SHJ. SHJ still
shuffles the build side without any row reduction. This optimizer rule here
will affect join without condition, so the join would be BNLJ or
`CartesianProduct`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
