[SPARK-29655] Enable adaptive execution should not add more ShuffleExchange

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 3.0.0
• Component/s: SQL
• Assignee: Yuming Wang
• Reporter: Yuming Wang
• Created: 30/Oct/19 10:10
• Updated: 15/Nov/19 07:52
• Resolved: 15/Nov/19 07:52

Description

Enable adaptive execution should not add more ShuffleExchange. How to reproduce:

import org.apache.spark.sql.SaveMode

spark.conf.set("spark.sql.autoBroadcastJoinThreshold", -1)
spark.conf.set("spark.sql.shuffle.partitions", 4)

val bucketedTableName = "bucketed_table"
spark.range(10).write.bucketBy(4, "id").sortBy("id").mode(SaveMode.Overwrite).saveAsTable(bucketedTableName)
val bucketedTable = spark.table(bucketedTableName)
val df = spark.range(4)

df.join(bucketedTable, "id").explain()

spark.conf.set("spark.sql.adaptive.enabled", true)
spark.conf.set("spark.sql.adaptive.shuffle.maxNumPostShufflePartitions", 5)
df.join(bucketedTable, "id").explain()

Output:

== Physical Plan ==
AdaptiveSparkPlan(isFinalPlan=false)
+- Project [id#5L]
   +- SortMergeJoin [id#5L], [id#3L], Inner
      :- Sort [id#5L ASC NULLS FIRST], false, 0
      :  +- Exchange hashpartitioning(id#5L, 5), true, [id=#92]
      :     +- Range (0, 4, step=1, splits=16)
      +- Sort [id#3L ASC NULLS FIRST], false, 0
         +- Exchange hashpartitioning(id#3L, 5), true, [id=#93]
            +- Project [id#3L]
               +- Filter isnotnull(id#3L)
                  +- FileScan parquet default.bucketed_table[id#3L] Batched: true, DataFilters: [isnotnull(id#3L)], Format: Parquet, Location: InMemoryFileIndex[file:/root/spark-3.0.0-preview-bin-hadoop3.2/spark-warehouse/bucketed_table], PartitionFilters: [], PushedFilters: [IsNotNull(id)], ReadSchema: struct<id:bigint>, SelectedBucketsCount: 4 out of 4

Issue Links

• links to: GitHub Pull Request #26409