ekoifman commented on a change in pull request #34464:
URL: https://github.com/apache/spark/pull/34464#discussion_r750365314
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/adaptive/DynamicJoinSelection.scala
##########
@@ -50,10 +51,17 @@ object DynamicJoinSelection extends Rule[LogicalPlan] {
mapStats.bytesByPartitionId.forall(_ <=
maxShuffledHashJoinLocalMapThreshold)
}
- private def selectJoinStrategy(plan: LogicalPlan): Option[JoinStrategyHint]
= plan match {
+ private def selectJoinStrategy(
+ plan: LogicalPlan,
+ joinType: JoinType): Option[JoinStrategyHint] = plan match {
case LogicalQueryStage(_, stage: ShuffleQueryStageExec) if
stage.isMaterialized
&& stage.mapStats.isDefined =>
- val demoteBroadcastHash =
shouldDemoteBroadcastHashJoin(stage.mapStats.get)
+ val demoteBroadcastHash = joinType match {
+ // doesn't make sense for outer joins since one side is preserved and
join is not
+ // short circuited if the other side is empty
+ case LeftOuter | RightOuter | FullOuter => false
Review comment:
BHJ can't support FullOuter - I put it in the list since it seems odd
(though not wrong) to add `no_broadcast_hash` to a join that doesn't support
BHJ anyway. Makes logs harder to read.
LeftSemi & LeftAnti could in principle support it - I don't know if they do
today.
--
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]
