[SPARK-33617][SQL][FOLLOWUP] refine the default parallelism SQL config

### What changes were proposed in this pull request? This is a followup of apache#30559 . The default parallelism config in Spark core is not good, as it's unclear where it applies. To not inherit this problem in Spark SQL, this PR refines the default parallelism SQL config, to make it clear that it only applies to leaf nodes. ### Why are the changes needed? Make the config clearer. ### Does this PR introduce _any_ user-facing change? It changes an unreleased config. ### How was this patch tested? existing tests Closes apache#30736 from cloud-fan/follow. Authored-by: Wenchen Fan <[email protected]> Signed-off-by: Wenchen Fan <[email protected]>
figo · Dec 15, 2020 · 40c37d6 · 40c37d6
1 parent 23083aa
commit 40c37d6
Show file tree

Hide file tree

Showing 9 changed files with 19 additions and 23 deletions.
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala
@@ -374,12 +374,13 @@ object SQLConf {
       .booleanConf
       .createWithDefault(true)
 
-  val DEFAULT_PARALLELISM = buildConf("spark.sql.default.parallelism")
-    .doc("The number of parallelism for Spark SQL, the default value is " +
-      "`spark.default.parallelism`.")
+  val LEAF_NODE_DEFAULT_PARALLELISM = buildConf("spark.sql.leafNodeDefaultParallelism")
+    .doc("The default parallelism of Spark SQL leaf nodes that produce data, such as the file " +
+      "scan node, the local data scan node, the range node, etc. The default value of this " +
+      "config is 'SparkContext#defaultParallelism'.")
     .version("3.2.0")
     .intConf
-    .checkValue(_ > 0, "The value of spark.sql.default.parallelism must be positive.")
+    .checkValue(_ > 0, "The value of spark.sql.leafNodeDefaultParallelism must be positive.")
     .createOptional
 
   val SHUFFLE_PARTITIONS = buildConf("spark.sql.shuffle.partitions")
@@ -3202,8 +3203,6 @@ class SQLConf extends Serializable with Logging {
 
   def cacheVectorizedReaderEnabled: Boolean = getConf(CACHE_VECTORIZED_READER_ENABLED)
 
-  def defaultParallelism: Option[Int] = getConf(DEFAULT_PARALLELISM)
-
   def defaultNumShufflePartitions: Int = getConf(SHUFFLE_PARTITIONS)
 
   def numShufflePartitions: Int = {

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/SparkSession.scala b/sql/core/src/main/scala/org/apache/spark/sql/SparkSession.scala
@@ -523,8 +523,7 @@ class SparkSession private(
    * @since 2.0.0
    */
   def range(start: Long, end: Long): Dataset[java.lang.Long] = {
-    range(start, end, step = 1,
-      numPartitions = sqlContext.conf.defaultParallelism.getOrElse(sparkContext.defaultParallelism))
+    range(start, end, step = 1, numPartitions = leafNodeDefaultParallelism)
   }
 
   /**
@@ -534,8 +533,7 @@ class SparkSession private(
    * @since 2.0.0
    */
   def range(start: Long, end: Long, step: Long): Dataset[java.lang.Long] = {
-    range(start, end, step,
-      numPartitions = sqlContext.conf.defaultParallelism.getOrElse(sparkContext.defaultParallelism))
+    range(start, end, step, numPartitions = leafNodeDefaultParallelism)
   }
 
   /**
@@ -775,6 +773,10 @@ class SparkSession private(
       SparkSession.setActiveSession(old)
     }
   }
+
+  private[sql] def leafNodeDefaultParallelism: Int = {
+    conf.get(SQLConf.LEAF_NODE_DEFAULT_PARALLELISM).getOrElse(sparkContext.defaultParallelism)
+  }
 }
 
 

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/LocalTableScanExec.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/LocalTableScanExec.scala
@@ -50,8 +50,7 @@ case class LocalTableScanExec(
       sqlContext.sparkContext.emptyRDD
     } else {
       val numSlices = math.min(
-        unsafeRows.length,
-        conf.defaultParallelism.getOrElse(sqlContext.sparkContext.defaultParallelism))
+        unsafeRows.length, sqlContext.sparkSession.leafNodeDefaultParallelism)
       sqlContext.sparkContext.parallelize(unsafeRows, numSlices)
     }
   }

diff --git a/...re/src/main/scala/org/apache/spark/sql/execution/adaptive/CoalesceShufflePartitions.scala b/...re/src/main/scala/org/apache/spark/sql/execution/adaptive/CoalesceShufflePartitions.scala
@@ -67,7 +67,7 @@ case class CoalesceShufflePartitions(session: SparkSession) extends CustomShuffl
         // We fall back to Spark default parallelism if the minimum number of coalesced partitions
         // is not set, so to avoid perf regressions compared to no coalescing.
         val minPartitionNum = conf.getConf(SQLConf.COALESCE_PARTITIONS_MIN_PARTITION_NUM)
-          .orElse(conf.defaultParallelism).getOrElse(session.sparkContext.defaultParallelism)
+          .getOrElse(session.sparkContext.defaultParallelism)
         val partitionSpecs = ShufflePartitionsUtil.coalescePartitions(
           validMetrics.toArray,
           advisoryTargetSize = conf.getConf(SQLConf.ADVISORY_PARTITION_SIZE_IN_BYTES),

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/basicPhysicalOperators.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/basicPhysicalOperators.scala
@@ -382,8 +382,7 @@ case class RangeExec(range: org.apache.spark.sql.catalyst.plans.logical.Range)
   val start: Long = range.start
   val end: Long = range.end
   val step: Long = range.step
-  val numSlices: Int = range.numSlices.orElse(sqlContext.conf.defaultParallelism)
-    .getOrElse(sparkContext.defaultParallelism)
+  val numSlices: Int = range.numSlices.getOrElse(sqlContext.sparkSession.leafNodeDefaultParallelism)
   val numElements: BigInt = range.numElements
   val isEmptyRange: Boolean = start == end || (start < end ^ 0 < step)
 

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/command/ddl.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/command/ddl.scala
@@ -738,8 +738,7 @@ case class AlterTableRecoverPartitionsCommand(
       // Set the number of parallelism to prevent following file listing from generating many tasks
       // in case of large #defaultParallelism.
       val numParallelism = Math.min(serializedPaths.length,
-        Math.min(spark.sessionState.conf.defaultParallelism
-          .getOrElse(spark.sparkContext.defaultParallelism), 10000))
+        Math.min(spark.sparkContext.defaultParallelism, 10000))
       // gather the fast stats for all the partitions otherwise Hive metastore will list all the
       // files for all the new partitions in sequential way, which is super slow.
       logInfo(s"Gather the fast stats in parallel using $numParallelism tasks.")

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FilePartition.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FilePartition.scala
@@ -89,8 +89,7 @@ object FilePartition extends Logging {
     val defaultMaxSplitBytes = sparkSession.sessionState.conf.filesMaxPartitionBytes
     val openCostInBytes = sparkSession.sessionState.conf.filesOpenCostInBytes
     val minPartitionNum = sparkSession.sessionState.conf.filesMinPartitionNum
-      .orElse(sparkSession.sessionState.conf.defaultParallelism)
-      .getOrElse(sparkSession.sparkContext.defaultParallelism)
+      .getOrElse(sparkSession.leafNodeDefaultParallelism)
     val totalBytes = selectedPartitions.flatMap(_.files.map(_.getLen + openCostInBytes)).sum
     val bytesPerCore = totalBytes / minPartitionNum
 

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SchemaMergeUtils.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SchemaMergeUtils.scala
@@ -57,8 +57,7 @@ object SchemaMergeUtils extends Logging {
     // Set the number of partitions to prevent following schema reads from generating many tasks
     // in case of a small number of orc files.
     val numParallelism = Math.min(Math.max(partialFileStatusInfo.size, 1),
-      sparkSession.sessionState.conf.defaultParallelism
-        .getOrElse(sparkSession.sparkContext.defaultParallelism))
+      sparkSession.sparkContext.defaultParallelism)
 
     val ignoreCorruptFiles = sparkSession.sessionState.conf.ignoreCorruptFiles
 

diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/SparkPlanSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/SparkPlanSuite.scala
@@ -89,9 +89,9 @@ class SparkPlanSuite extends QueryTest with SharedSparkSession {
     assert(LocalTableScanExec(Nil, Nil).execute().getNumPartitions == 0)
   }
 
-  test("SPARK-33617: spark.sql.default.parallelism effective for LocalTableScan") {
+  test("SPARK-33617: change default parallelism of LocalTableScan") {
     Seq(1, 4).foreach { minPartitionNum =>
-      withSQLConf(SQLConf.DEFAULT_PARALLELISM.key -> minPartitionNum.toString) {
+      withSQLConf(SQLConf.LEAF_NODE_DEFAULT_PARALLELISM.key -> minPartitionNum.toString) {
         val df = spark.sql("SELECT * FROM VALUES (1), (2), (3), (4), (5), (6), (7), (8)")
         assert(df.rdd.partitions.length === minPartitionNum)
       }