cloud-fan commented on a change in pull request #31756:
URL: https://github.com/apache/spark/pull/31756#discussion_r596855011
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/adaptive/PlanAdaptiveDynamicPruningFilters.scala
##########
@@ -41,15 +42,26 @@ case class PlanAdaptiveDynamicPruningFilters(
adaptivePlan: AdaptiveSparkPlanExec), exprId, _)) =>
val packedKeys = BindReferences.bindReferences(
HashJoin.rewriteKeyExpr(buildKeys), adaptivePlan.executedPlan.output)
- val mode = HashedRelationBroadcastMode(packedKeys)
- // plan a broadcast exchange of the build side of the join
- val exchange = BroadcastExchangeExec(mode, adaptivePlan.executedPlan)
- val existingStage = stageCache.get(exchange.canonicalized)
- if (existingStage.nonEmpty && conf.exchangeReuseEnabled) {
- val name = s"dynamicpruning#${exprId.id}"
- val reuseQueryStage = existingStage.get.newReuseInstance(0,
exchange.output)
- val broadcastValues =
- SubqueryBroadcastExec(name, index, buildKeys, reuseQueryStage)
+
+ val canReuseExchange = conf.exchangeReuseEnabled && buildKeys.nonEmpty
&&
+ plan.find {
+ case BroadcastHashJoinExec(_, _, _, BuildLeft, _, left, _, _) =>
+ left.sameResult(adaptivePlan.executedPlan)
+ case BroadcastHashJoinExec(_, _, _, BuildRight, _, _, right, _) =>
+ right.sameResult(adaptivePlan.executedPlan)
+ case _ => false
+ }.isDefined
+
+ if(canReuseExchange) {
+ val mode = HashedRelationBroadcastMode(packedKeys)
+ // plan a broadcast exchange of the build side of the join
+ val exchange = BroadcastExchangeExec(mode, adaptivePlan.executedPlan)
+ exchange.setLogicalLink(adaptivePlan.executedPlan.logicalLink.get)
+ val newAdaptivePlan = AdaptiveSparkPlanExec(
Review comment:
to double-check, the core change is: now we allow the run the broadcast
in DPP first, and the corresponding broadcast in other parts of the plan will
be reused later. Do I understand correctly?
If the broadcast in other parts of the plan is running (submitted) but not
finished, will we run the broadcast twice?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
