[SPARK-19426][SQL] Custom coalescer for Dataset

core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala

@@ -1158,8 +1158,17 @@ class RDDSuite extends SparkFunSuite with SharedSparkContext {
  * Took this class out of the test suite to prevent "Task not serializable" exceptions.
  */
 class SizeBasedCoalescer(val maxSize: Int) extends PartitionCoalescer with Serializable {
+
+  def getPartitions(parent: RDD[_]): Array[Partition] = {
+    parent.asInstanceOf[HadoopRDD[Any, Any]].getPartitions
+  }
+
+  def getPartitionSize(partition: Partition): Long = {
+    partition.asInstanceOf[HadoopPartition].inputSplit.value.getLength
+  }
+
   override def coalesce(maxPartitions: Int, parent: RDD[_]): Array[PartitionGroup] = {
-    val partitions: Array[Partition] = parent.asInstanceOf[HadoopRDD[Any, Any]].getPartitions
+    val partitions = getPartitions(parent)
     val groups = ArrayBuffer[PartitionGroup]()
     var currentGroup = new PartitionGroup()
     var currentSum = 0L
@@ -1168,8 +1177,8 @@ class SizeBasedCoalescer(val maxSize: Int) extends PartitionCoalescer with Seria
 
     // sort partitions based on the size of the corresponding input splits
     partitions.sortWith((partition1, partition2) => {
-      val partition1Size = partition1.asInstanceOf[HadoopPartition].inputSplit.value.getLength
-      val partition2Size = partition2.asInstanceOf[HadoopPartition].inputSplit.value.getLength
+      val partition1Size = getPartitionSize(partition1)
+      val partition2Size = getPartitionSize(partition2)
       partition1Size < partition2Size
     })
 
@@ -1187,9 +1196,7 @@ class SizeBasedCoalescer(val maxSize: Int) extends PartitionCoalescer with Seria
 
     while (index < partitions.size) {
       val partition = partitions(index)
-      val fileSplit =
-        partition.asInstanceOf[HadoopPartition].inputSplit.value.asInstanceOf[FileSplit]
-      val splitSize = fileSplit.getLength
+      val splitSize = getPartitionSize(partition)
       if (currentSum + splitSize < maxSize) {
         addPartition(partition, splitSize)
         index += 1

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala

@@ -596,14 +596,17 @@ object CollapseProject extends Rule[LogicalPlan] {
 object CollapseRepartition extends Rule[LogicalPlan] {
   def apply(plan: LogicalPlan): LogicalPlan = plan transformUp {
     // Case 1: When a Repartition has a child of Repartition or RepartitionByExpression,
-    // 1) When the top node does not enable the shuffle (i.e., coalesce API), but the child
-    //   enables the shuffle. Returns the child node if the last numPartitions is bigger;
-    //   otherwise, keep unchanged.
+    // 1) When the top node does not enable the shuffle (i.e., coalesce with no user-specified
+    //   strategy), but the child enables the shuffle. Returns the child node if the last
+    //   numPartitions is bigger; otherwise, keep unchanged.
     // 2) In the other cases, returns the top node with the child's child
-    case r @ Repartition(_, _, child: RepartitionOperation) => (r.shuffle, child.shuffle) match {
-      case (false, true) => if (r.numPartitions >= child.numPartitions) child else r
-      case _ => r.copy(child = child.child)
-    }
+    case r @ Repartition(_, _, child: RepartitionOperation, coalescer) =>
+      (r.shuffle, child.shuffle) match {
+        case (false, true) =>
+          if (coalescer.isEmpty && r.numPartitions >= child.numPartitions) child else r
+        case _ =>
+          r.copy(child = child.child)
+      }
     // Case 2: When a RepartitionByExpression has a child of Repartition or RepartitionByExpression
     // we can remove the child.
     case r @ RepartitionByExpression(_, child: RepartitionOperation, _) =>

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala

@@ -17,12 +17,12 @@
 
 package org.apache.spark.sql.catalyst.plans.logical
 
+import org.apache.spark.rdd.PartitionCoalescer
 import org.apache.spark.sql.catalyst.analysis.MultiInstanceRelation
 import org.apache.spark.sql.catalyst.catalog.CatalogTable
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression
 import org.apache.spark.sql.catalyst.plans._
-import org.apache.spark.sql.catalyst.plans.logical.statsEstimation._
 import org.apache.spark.sql.types._
 import org.apache.spark.util.Utils
 import org.apache.spark.util.random.RandomSampler
@@ -746,10 +746,24 @@ abstract class RepartitionOperation extends UnaryNode {
  * [[RepartitionByExpression]] as this method is called directly by DataFrame's, because the user
  * asked for `coalesce` or `repartition`. [[RepartitionByExpression]] is used when the consumer
  * of the output requires some specific ordering or distribution of the data.
+ *
+ * If `shuffle` = false (`coalesce` cases), this logical plan can have an user-specified strategy
+ * to coalesce input partitions.
+ *
+ * @param numPartitions How many partitions to use in the output RDD
+ * @param shuffle Whether to shuffle when repartitioning
+ * @param child the LogicalPlan
+ * @param coalescer Optional coalescer that an user specifies
  */
-case class Repartition(numPartitions: Int, shuffle: Boolean, child: LogicalPlan)
+case class Repartition(
+    numPartitions: Int,
+    shuffle: Boolean,
+    child: LogicalPlan,
+    coalescer: Option[PartitionCoalescer] = None)
   extends RepartitionOperation {
   require(numPartitions > 0, s"Number of partitions ($numPartitions) must be positive.")
+  require(!shuffle || coalescer.isEmpty,
+    "Custom coalescer is not allowed for repartition(shuffle=true)")
 }
 
 /**

sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala

@@ -33,7 +33,7 @@ import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.api.java.function._
 import org.apache.spark.api.python.{PythonRDD, SerDeUtil}
 import org.apache.spark.broadcast.Broadcast
-import org.apache.spark.rdd.RDD
+import org.apache.spark.rdd.{PartitionCoalescer, RDD}
 import org.apache.spark.sql.catalyst._
 import org.apache.spark.sql.catalyst.analysis._
 import org.apache.spark.sql.catalyst.catalog.CatalogRelation
@@ -2661,6 +2661,31 @@ class Dataset[T] private[sql](
       partitionExprs.map(_.expr), logicalPlan, sparkSession.sessionState.conf.numShufflePartitions)
   }
 
+  /**
+   * Returns a new Dataset that an user-defined `PartitionCoalescer` reduces into fewer partitions.
+   * `userDefinedCoalescer` is the same with a coalescer used in the `RDD` coalesce function.
+   *
+   * If a larger number of partitions is requested, it will stay at the current
+   * number of partitions. Similar to coalesce defined on an `RDD`, this operation results in
+   * a narrow dependency, e.g. if you go from 1000 partitions to 100 partitions, there will not
+   * be a shuffle, instead each of the 100 new partitions will claim 10 of the current partitions.
+   *
+   * However, if you're doing a drastic coalesce, e.g. to numPartitions = 1,
+   * this may result in your computation taking place on fewer nodes than
+   * you like (e.g. one node in the case of numPartitions = 1). To avoid this,
+   * you can call repartition. This will add a shuffle step, but means the
+   * current upstream partitions will be executed in parallel (per whatever
+   * the current partitioning is).
+   *
+   * @group typedrel
+   * @since 2.3.0
+   */
+  def coalesce(
+      numPartitions: Int,
+      userDefinedCoalescer: Option[PartitionCoalescer]): Dataset[T] = withTypedPlan {
+    Repartition(numPartitions, shuffle = false, logicalPlan, userDefinedCoalescer)
+  }
+
   /**
    * Returns a new Dataset that has exactly `numPartitions` partitions, when the fewer partitions
    * are requested. If a larger number of partitions is requested, it will stay at the current

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala

@@ -390,11 +390,11 @@ abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
           f, key, lObj, rObj, lGroup, rGroup, lAttr, rAttr, oAttr,
           planLater(left), planLater(right)) :: Nil
 
-      case logical.Repartition(numPartitions, shuffle, child) =>
+      case logical.Repartition(numPartitions, shuffle, child, coalescer) =>
         if (shuffle) {
           ShuffleExchange(RoundRobinPartitioning(numPartitions), planLater(child)) :: Nil
         } else {
-          execution.CoalesceExec(numPartitions, planLater(child)) :: Nil
+          execution.CoalesceExec(numPartitions, planLater(child), coalescer) :: Nil
         }
       case logical.Sort(sortExprs, global, child) =>
         execution.SortExec(sortExprs, global, planLater(child)) :: Nil

sql/core/src/main/scala/org/apache/spark/sql/execution/basicPhysicalOperators.scala

@@ -20,14 +20,13 @@ package org.apache.spark.sql.execution
 import scala.concurrent.{ExecutionContext, Future}
 import scala.concurrent.duration.Duration
 
-import org.apache.spark.{InterruptibleIterator, SparkException, TaskContext}
-import org.apache.spark.rdd.{EmptyRDD, PartitionwiseSampledRDD, RDD}
+import org.apache.spark.{InterruptibleIterator, TaskContext}
+import org.apache.spark.rdd.{EmptyRDD, PartitionCoalescer, PartitionwiseSampledRDD, RDD}
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.expressions.codegen.{CodegenContext, ExprCode, ExpressionCanonicalizer}
 import org.apache.spark.sql.catalyst.plans.physical._
 import org.apache.spark.sql.execution.metric.SQLMetrics
-import org.apache.spark.sql.execution.ui.SparkListenerDriverAccumUpdates
 import org.apache.spark.sql.types.LongType
 import org.apache.spark.util.ThreadUtils
 import org.apache.spark.util.random.{BernoulliCellSampler, PoissonSampler}
@@ -561,7 +560,7 @@ case class UnionExec(children: Seq[SparkPlan]) extends SparkPlan {
  * Physical plan for returning a new RDD that has exactly `numPartitions` partitions.
  * Similar to coalesce defined on an [[RDD]], this operation results in a narrow dependency, e.g.
  * if you go from 1000 partitions to 100 partitions, there will not be a shuffle, instead each of
- * the 100 new partitions will claim 10 of the current partitions.  If a larger number of partitions
+ * the 100 new partitions will claim 10 of the current partitions. If a larger number of partitions
  * is requested, it will stay at the current number of partitions.
  *
  * However, if you're doing a drastic coalesce, e.g. to numPartitions = 1,
@@ -570,8 +569,16 @@ case class UnionExec(children: Seq[SparkPlan]) extends SparkPlan {
  * you see ShuffleExchange. This will add a shuffle step, but means the
  * current upstream partitions will be executed in parallel (per whatever
  * the current partitioning is).
+ *
+ * If you want to define how to coalesce partitions, you can set a custom strategy
+ * to coalesce partitions in `coalescer`.
+ *
+ * @param numPartitions Number of partitions this coalescer tries to reduce partitions into
+ * @param child the SparkPlan
+ * @param coalescer Optional coalescer that an user specifies
  */
-case class CoalesceExec(numPartitions: Int, child: SparkPlan) extends UnaryExecNode {
+case class CoalesceExec(numPartitions: Int, child: SparkPlan, coalescer: Option[PartitionCoalescer])
+  extends UnaryExecNode {
   override def output: Seq[Attribute] = child.output
 
   override def outputPartitioning: Partitioning = {
@@ -580,7 +587,7 @@ case class CoalesceExec(numPartitions: Int, child: SparkPlan) extends UnaryExecN
   }
 
   protected override def doExecute(): RDD[InternalRow] = {
-    child.execute().coalesce(numPartitions, shuffle = false)
+    child.execute().coalesce(numPartitions, shuffle = false, coalescer)
   }
 }
 

sql/core/src/test/scala/org/apache/spark/sql/DatasetSuite.scala

@@ -20,10 +20,13 @@ package org.apache.spark.sql
 import java.io.{Externalizable, ObjectInput, ObjectOutput}
 import java.sql.{Date, Timestamp}
 
+import org.apache.spark.Partition
+import org.apache.spark.rdd._
 import org.apache.spark.sql.catalyst.encoders.{OuterScopes, RowEncoder}
 import org.apache.spark.sql.catalyst.plans.{LeftAnti, LeftSemi}
 import org.apache.spark.sql.catalyst.util.sideBySide
 import org.apache.spark.sql.execution.{LogicalRDD, RDDScanExec}
+import org.apache.spark.sql.execution.datasources.{FilePartition, FileScanRDD}
 import org.apache.spark.sql.execution.exchange.{BroadcastExchangeExec, ShuffleExchange}
 import org.apache.spark.sql.execution.streaming.MemoryStream
 import org.apache.spark.sql.functions._
@@ -119,6 +122,39 @@ class DatasetSuite extends QueryTest with SharedSQLContext {
       data: _*)
   }
 
+  test("coalesce, custom") {
+    withTempPath { path =>
+      val maxSplitSize = 512
+      val testData = (1 to 1000).map(i => ClassData(i.toString, i))
+      testData.toDS().repartition(50).write.format("csv").save(path.toString)
+
+      withSQLConf(
+        SQLConf.FILES_MAX_PARTITION_BYTES.key -> (maxSplitSize / 3).toString,
+        SQLConf.FILES_OPEN_COST_IN_BYTES.key -> "0"
+      ) {
+        val ds = spark.read.format("csv")
+          .schema("a STRING, b INT")
+          .load(path.toString)
+          .as[ClassData]
+
+        val coalescedDataSet =
+          ds.coalesce(4, Some(new DatasetSizeBasedPartitionCoalescer(maxSplitSize)))
+
+        assert(coalescedDataSet.rdd.partitions.length <= 50)
+
+        val expectedPartitionCount = ds.rdd.partitions.size
+        val totalPartitionCount = coalescedDataSet.rdd.partitions.map { p1 =>
+          val splitSizes = p1.asInstanceOf[CoalescedRDDPartition].parents.map { p2 =>
+            p2.asInstanceOf[FilePartition].files.map(_.length).sum
+          }
+          assert(splitSizes.sum <= maxSplitSize)
+          splitSizes.size
+        }.sum
+        assert(totalPartitionCount === expectedPartitionCount)
+      }
+    }
+  }
+
   test("as tuple") {
     val data = Seq(("a", 1), ("b", 2)).toDF("a", "b")
     checkDataset(
@@ -1402,3 +1438,14 @@ case class CircularReferenceClassB(cls: CircularReferenceClassA)
 case class CircularReferenceClassC(ar: Array[CircularReferenceClassC])
 case class CircularReferenceClassD(map: Map[String, CircularReferenceClassE])
 case class CircularReferenceClassE(id: String, list: List[CircularReferenceClassD])
+
+class DatasetSizeBasedPartitionCoalescer(maxSize: Int) extends SizeBasedCoalescer(maxSize) {
+
+  override def getPartitions(parent: RDD[_]): Array[Partition] = {
+    parent.firstParent.partitions
+  }
+
+  override def getPartitionSize(partition: Partition): Long = {
+    partition.asInstanceOf[FilePartition].files.map(x => x.length - x.start).sum
+  }
+}

sql/core/src/test/scala/org/apache/spark/sql/execution/PlannerSuite.scala

@@ -244,7 +244,7 @@ class PlannerSuite extends SharedSQLContext {
     assert(countRepartitions(doubleRepartitioned.queryExecution.logical) === 3)
     assert(countRepartitions(doubleRepartitioned.queryExecution.optimizedPlan) === 2)
     doubleRepartitioned.queryExecution.optimizedPlan match {
-      case Repartition (numPartitions, shuffle, Repartition(_, shuffleChild, _)) =>
+      case Repartition(numPartitions, shuffle, Repartition(_, shuffleChild, _, _), _) =>
         assert(numPartitions === 5)
         assert(shuffle === false)
         assert(shuffleChild === true)