ulysses-you commented on a change in pull request #32816:
URL: https://github.com/apache/spark/pull/32816#discussion_r695402449
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/adaptive/AQEUtils.scala
##########
@@ -57,4 +58,20 @@ object AQEUtils {
}
case _ => Some(UnspecifiedDistribution)
}
+
+ // Add an extra shuffle if input plan does not satisfy the required
distribution.
+ // This method is invoked after optimizing skewed join in case we change
final stage
+ // output partitioning.
+ def ensureRequiredDistribution(
+ plan: SparkPlan, distribution: Option[Distribution]): SparkPlan =
distribution match {
+ case Some(d) if !plan.outputPartitioning.satisfies(d) =>
+ val numPartitions =
d.requiredNumPartitions.getOrElse(conf.numShufflePartitions)
+ val shuffleOrigin = if (d.requiredNumPartitions.isDefined) {
+ REPARTITION_BY_NUM
+ } else {
+ REPARTITION_BY_COL
+ }
+ ShuffleExchangeExec(d.createPartitioning(numPartitions), plan,
shuffleOrigin)
Review comment:
We should retain origin shuffle here, otherwise some stage optimize
rules can break the `ENSIRE_REQUIREMENT` output partitioning in next stage.
e.g. `CoalesceShufflePartitions` when the input plan's origin is
`REPARTITION_BY_NUM`
--
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]
