cloud-fan commented on a change in pull request #25295: [SPARK-28560][SQL]
Optimize shuffle reader to local shuffle reader when smj converted to bhj in
adaptive execution
URL: https://github.com/apache/spark/pull/25295#discussion_r323609170
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/adaptive/ReduceNumShufflePartitions.scala
##########
@@ -180,25 +180,45 @@ case class ReduceNumShufflePartitions(conf: SQLConf)
extends Rule[SparkPlan] {
case class CoalescedShuffleReaderExec(
child: QueryStageExec,
- partitionStartIndices: Array[Int]) extends UnaryExecNode {
+ partitionStartIndices: Array[Int],
+ var isLocal: Boolean = false) extends UnaryExecNode {
Review comment:
without local shuffle reader, a task of `ShuffledRDD` reads the shuffle
blocks `map1-reduce1`, `map2-reduce1`, etc. With local shuffle reader, the task
reads `map1-reduce1`, `map1-reduce2`, etc. The task output data size is
different and we can't use the algorithm in `ReduceNumShufflePartitions`
anymore.
Furthermore, the RDD numPartitions also becomes different after switching to
local shuffle reader, how can we apply the `ReduceNumShufflePartitions`?
----------------------------------------------------------------
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:
us...@infra.apache.org
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
