[SPARK-47094][SQL] SPJ : Dynamically rebalance number of buckets when they are not equal

szehon-ho · szehon-ho · commit c81c039a554c · 2024-02-26T16:37:57.000-08:00
### What changes were proposed in this pull request?
-- Allow SPJ between 'compatible' bucket funtions
-- Add a mechanism to define 'reducible' functions, one function whose output can be 'reduced' to another for all inputs.

  ### Why are the changes needed?
-- SPJ currently applies only if the partition transform expressions on both sides are identifical.

  ### Does this PR introduce _any_ user-facing change?
No

  ### How was this patch tested?
Added new tests in KeyGroupedPartitioningSuite

  ### Was this patch authored or co-authored using generative AI tooling?
No
diff --git a/sql/catalyst/src/main/java/org/apache/spark/sql/connector/catalog/functions/Reducer.java b/sql/catalyst/src/main/java/org/apache/spark/sql/connector/catalog/functions/Reducer.java
@@ -0,0 +1,16 @@
+package org.apache.spark.sql.connector.catalog.functions;
+
+import org.apache.spark.annotation.Evolving;
+
+/**
+ * A 'reducer' for output of user-defined functions.
+ *
+ * A user_defined function f_source(x) is 'reducible' on another user_defined function f_target(x),
+ * if there exists a 'reducer' r(x) such that r(f_source(x)) = f_target(x) for all input x.
+ * @param <T> function output type
+ * @since 4.0.0
+ */
+@Evolving
+public interface Reducer<T> {
+    T reduce(T arg1);
+}
diff --git a/sql/catalyst/src/main/java/org/apache/spark/sql/connector/catalog/functions/ReducibleFunction.java b/sql/catalyst/src/main/java/org/apache/spark/sql/connector/catalog/functions/ReducibleFunction.java
@@ -0,0 +1,25 @@
+package org.apache.spark.sql.connector.catalog.functions;
+
+import org.apache.spark.annotation.Evolving;
+import scala.Option;
+
+/**
+ * Base class for user-defined functions that can be 'reduced' on another function.
+ *
+ * A function f_source(x) is 'reducible' on another function f_target(x) if
+ * there exists a reducer function r(x) such that r(f_source(x)) = f_target(x) for all input x.
+ *
+ * @since 4.0.0
+ */
+@Evolving
+public interface ReducibleFunction<T, A> extends ScalarFunction<T> {
+
+    /**
+     * If this function is 'reducible' on another function, return the {@link Reducer} function.
+     * @param other other function
+     * @param thisArgument argument for this function instance
+     * @param otherArgument argument for other function instance
+     * @return a reduction function if it is reducible, none if not
+     */
+    Option<Reducer<A>> reducer(ReducibleFunction<?, ?> other, Option<?> thisArgument, Option<?> otherArgument);
+}
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/TransformExpression.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/TransformExpression.scala
@@ -17,7 +17,7 @@
 
 package org.apache.spark.sql.catalyst.expressions
 
-import org.apache.spark.sql.connector.catalog.functions.BoundFunction
+import org.apache.spark.sql.connector.catalog.functions.{BoundFunction, ReducibleFunction}
 import org.apache.spark.sql.types.DataType
 
 /**
@@ -54,6 +54,31 @@ case class TransformExpression(
       false
   }
 
+  /**
+   * Whether this [[TransformExpression]]'s function is compatible with the `other`
+   * [[TransformExpression]]'s function.
+   *
+   * This is true if both are instances of [[ReducibleFunction]] and there exists a [[Reducer]] r(x)
+   * such that r(t1(x)) = t2(x), or r(t2(x)) = t1(x), for all input x.
+   *
+   * @param other the transform expression to compare to
+   * @return true if compatible, false if not
+   */
+  def isCompatible(other: TransformExpression): Boolean = {
+    if (isSameFunction(other)) {
+      true
+    } else {
+      (function, other.function) match {
+        case (f: ReducibleFunction[Any, Any] @unchecked,
+          o: ReducibleFunction[Any, Any] @unchecked) =>
+          val reducer = f.reducer(o, numBucketsOpt, other.numBucketsOpt)
+          val otherReducer = o.reducer(f, other.numBucketsOpt, numBucketsOpt)
+          reducer.isDefined || otherReducer.isDefined
+        case _ => false
+      }
+    }
+  }
+
   override def dataType: DataType = function.resultType()
 
   override protected def withNewChildrenInternal(newChildren: IndexedSeq[Expression]): Expression =
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala
@@ -24,6 +24,7 @@ import org.apache.spark.{SparkException, SparkUnsupportedOperationException}
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.util.InternalRowComparableWrapper
+import org.apache.spark.sql.connector.catalog.functions.{Reducer, ReducibleFunction}
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.types.{DataType, IntegerType}
 
@@ -635,6 +636,22 @@ trait ShuffleSpec {
    */
   def createPartitioning(clustering: Seq[Expression]): Partitioning =
     throw SparkUnsupportedOperationException()
+
+  /**
+   * Return a set of [[Reducer]] for the partition expressions of this shuffle spec,
+   * on the partition expressions of another shuffle spec.
+   * <p>
+   * A [[Reducer]] exists for a partition expression function of this shuffle spec if it is
+   * 'reducible' on the corresponding partition expression function of the other shuffle spec.
+   * <p>
+   * If a value is returned, there must be one Option[[Reducer]] per partition expression.
+   * A None value in the set indicates that the particular partition expression is not reducible
+   * on the corresponding expression on the other shuffle spec.
+   * <p>
+   * Returning none also indicates that none of the partition expressions can be reduced on the
+   * corresponding expression on the other shuffle spec.
+   */
+  def reducers(spec: ShuffleSpec): Option[Seq[Option[Reducer[Any]]]] = None
 }
 
 case object SinglePartitionShuffleSpec extends ShuffleSpec {
@@ -829,20 +846,60 @@ case class KeyGroupedShuffleSpec(
     }
   }
 
+  override def canCreatePartitioning: Boolean = SQLConf.get.v2BucketingShuffleEnabled &&
+    // Only support partition expressions are AttributeReference for now
+    partitioning.expressions.forall(_.isInstanceOf[AttributeReference])
+
+  override def createPartitioning(clustering: Seq[Expression]): Partitioning = {
+    KeyGroupedPartitioning(clustering, partitioning.numPartitions, partitioning.partitionValues)
+  }
+
+  override def reducers(other: ShuffleSpec): Option[Seq[Option[Reducer[Any]]]] = {
+    other match {
+      case otherSpec: KeyGroupedShuffleSpec =>
+        val results = partitioning.expressions.zip(otherSpec.partitioning.expressions).map {
+          case (e1: TransformExpression, e2: TransformExpression)
+            if e1.function.isInstanceOf[ReducibleFunction[Any, Any]@unchecked]
+              && e2.function.isInstanceOf[ReducibleFunction[Any, Any]@unchecked] =>
+            e1.function.asInstanceOf[ReducibleFunction[Any, Any]].reducer(
+              e2.function.asInstanceOf[ReducibleFunction[Any, Any]],
+              e1.numBucketsOpt.map(a => a.asInstanceOf[Any]),
+              e2.numBucketsOpt.map(a => a.asInstanceOf[Any]))
+          case (_, _) => None
+        }
+
+        // optimize to not return a value, if none of the partition expressions need reducing
+        if (results.forall(p => p.isEmpty)) None else Some(results)
+      case _ => None
+    }
+  }
+
   private def isExpressionCompatible(left: Expression, right: Expression): Boolean =
     (left, right) match {
       case (_: LeafExpression, _: LeafExpression) => true
       case (left: TransformExpression, right: TransformExpression) =>
-        left.isSameFunction(right)
+        if (SQLConf.get.v2BucketingPushPartValuesEnabled &&
+          !SQLConf.get.v2BucketingPartiallyClusteredDistributionEnabled &&
+          SQLConf.get.v2BucketingAllowCompatibleTransforms) {
+          left.isCompatible(right)
+        } else {
+          left.isSameFunction(right)
+        }
       case _ => false
     }
+}
 
-  override def canCreatePartitioning: Boolean = SQLConf.get.v2BucketingShuffleEnabled &&
-    // Only support partition expressions are AttributeReference for now
-    partitioning.expressions.forall(_.isInstanceOf[AttributeReference])
-
-  override def createPartitioning(clustering: Seq[Expression]): Partitioning = {
-    KeyGroupedPartitioning(clustering, partitioning.numPartitions, partitioning.partitionValues)
+object KeyGroupedShuffleSpec {
+  def reducePartitionValue(row: InternalRow,
+                           expressions: Seq[Expression],
+                           reducers: Seq[Option[Reducer[Any]]]):
+    InternalRowComparableWrapper = {
+    val partitionVals = row.toSeq(expressions.map(_.dataType))
+    val reducedRow = partitionVals.zip(reducers).map{
+      case (v, Some(reducer)) => reducer.reduce(v)
+      case (v, _) => v
+    }.toArray
+    InternalRowComparableWrapper(new GenericInternalRow(reducedRow), expressions)
   }
 }
 
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
@@ -1537,6 +1537,18 @@ object SQLConf {
       .booleanConf
       .createWithDefault(false)
 
+  val V2_BUCKETING_ALLOW_COMPATIBLE_TRANSFORMS =
+    buildConf("spark.sql.sources.v2.bucketing.allow.enabled")
+      .doc("Whether to allow storage-partition join in the case where the partition transforms" +
+        "are compatible but not identical.  This config requires both " +
+        s"${V2_BUCKETING_ENABLED.key} and ${V2_BUCKETING_PUSH_PART_VALUES_ENABLED.key} to be " +
+        s"enabled and ${V2_BUCKETING_PARTIALLY_CLUSTERED_DISTRIBUTION_ENABLED.key} " +
+        "to be disabled."
+      )
+      .version("4.0.0")
+      .booleanConf
+      .createWithDefault(false)
+
   val BUCKETING_MAX_BUCKETS = buildConf("spark.sql.sources.bucketing.maxBuckets")
     .doc("The maximum number of buckets allowed.")
     .version("2.4.0")
@@ -5201,6 +5213,9 @@ class SQLConf extends Serializable with Logging with SqlApiConf {
   def v2BucketingAllowJoinKeysSubsetOfPartitionKeys: Boolean =
     getConf(SQLConf.V2_BUCKETING_ALLOW_JOIN_KEYS_SUBSET_OF_PARTITION_KEYS)
 
+  def v2BucketingAllowCompatibleTransforms: Boolean =
+    getConf(SQLConf.V2_BUCKETING_ALLOW_COMPATIBLE_TRANSFORMS)
+
   def dataFrameSelfJoinAutoResolveAmbiguity: Boolean =
     getConf(DATAFRAME_SELF_JOIN_AUTO_RESOLVE_AMBIGUITY)
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/BatchScanExec.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/BatchScanExec.scala
@@ -24,9 +24,10 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.plans.QueryPlan
-import org.apache.spark.sql.catalyst.plans.physical.{KeyGroupedPartitioning, Partitioning, SinglePartition}
+import org.apache.spark.sql.catalyst.plans.physical.{KeyGroupedPartitioning, KeyGroupedShuffleSpec, Partitioning, SinglePartition}
 import org.apache.spark.sql.catalyst.util.{truncatedString, InternalRowComparableWrapper}
 import org.apache.spark.sql.connector.catalog.Table
+import org.apache.spark.sql.connector.catalog.functions.Reducer
 import org.apache.spark.sql.connector.read._
 import org.apache.spark.util.ArrayImplicits._
 
@@ -164,6 +165,18 @@ case class BatchScanExec(
               (groupedParts, expressions)
           }
 
+          // Also re-group the partitions if we are reducing compatible partition expressions
+          val finalGroupedPartitions = spjParams.reducers match {
+            case Some(reducers) =>
+              val result = groupedPartitions.groupBy { case (row, _) =>
+                KeyGroupedShuffleSpec.reducePartitionValue(row, partExpressions, reducers)
+              }.map { case (wrapper, splits) => (wrapper.row, splits.flatMap(_._2)) }.toSeq
+              val rowOrdering = RowOrdering.createNaturalAscendingOrdering(
+                expressions.map(_.dataType))
+              result.sorted(rowOrdering.on((t: (InternalRow, _)) => t._1))
+            case _ => groupedPartitions
+          }
+
           // When partially clustered, the input partitions are not grouped by partition
           // values. Here we'll need to check `commonPartitionValues` and decide how to group
           // and replicate splits within a partition.
@@ -174,7 +187,7 @@ case class BatchScanExec(
                 .get
                 .map(t => (InternalRowComparableWrapper(t._1, partExpressions), t._2))
                 .toMap
-            val nestGroupedPartitions = groupedPartitions.map { case (partValue, splits) =>
+            val nestGroupedPartitions = finalGroupedPartitions.map { case (partValue, splits) =>
               // `commonPartValuesMap` should contain the part value since it's the super set.
               val numSplits = commonPartValuesMap
                   .get(InternalRowComparableWrapper(partValue, partExpressions))
@@ -207,7 +220,7 @@ case class BatchScanExec(
           } else {
             // either `commonPartitionValues` is not defined, or it is defined but
             // `applyPartialClustering` is false.
-            val partitionMapping = groupedPartitions.map { case (partValue, splits) =>
+            val partitionMapping = finalGroupedPartitions.map { case (partValue, splits) =>
               InternalRowComparableWrapper(partValue, partExpressions) -> splits
             }.toMap
 
@@ -224,7 +237,6 @@ case class BatchScanExec(
 
         case _ => filteredPartitions
       }
-
       new DataSourceRDD(
         sparkContext, finalPartitions, readerFactory, supportsColumnar, customMetrics)
     }
@@ -259,6 +271,7 @@ case class StoragePartitionJoinParams(
     keyGroupedPartitioning: Option[Seq[Expression]] = None,
     joinKeyPositions: Option[Seq[Int]] = None,
     commonPartitionValues: Option[Seq[(InternalRow, Int)]] = None,
+    reducers: Option[Seq[Option[Reducer[Any]]]] = None,
     applyPartialClustering: Boolean = false,
     replicatePartitions: Boolean = false) {
   override def equals(other: Any): Boolean = other match {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/EnsureRequirements.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/EnsureRequirements.scala
@@ -26,6 +26,7 @@ import org.apache.spark.sql.catalyst.plans._
 import org.apache.spark.sql.catalyst.plans.physical._
 import org.apache.spark.sql.catalyst.rules.Rule
 import org.apache.spark.sql.catalyst.util.InternalRowComparableWrapper
+import org.apache.spark.sql.connector.catalog.functions.Reducer
 import org.apache.spark.sql.execution._
 import org.apache.spark.sql.execution.datasources.v2.BatchScanExec
 import org.apache.spark.sql.execution.joins.{ShuffledHashJoinExec, SortMergeJoinExec}
@@ -505,11 +506,28 @@ case class EnsureRequirements(
           }
         }
 
-        // Now we need to push-down the common partition key to the scan in each child
-        newLeft = populatePartitionValues(left, mergedPartValues, leftSpec.joinKeyPositions,
-          applyPartialClustering, replicateLeftSide)
-        newRight = populatePartitionValues(right, mergedPartValues, rightSpec.joinKeyPositions,
-          applyPartialClustering, replicateRightSide)
+        // in case of compatible but not identical partition expressions, we apply 'reduce'
+        // transforms to group one side's partitions as well as the common partition values
+        val leftReducers = leftSpec.reducers(rightSpec)
+        val rightReducers = rightSpec.reducers(leftSpec)
+
+        if (leftReducers.isDefined || rightReducers.isDefined) {
+          mergedPartValues = reduceCommonPartValues(mergedPartValues,
+            leftSpec.partitioning.expressions,
+            leftReducers)
+          mergedPartValues = reduceCommonPartValues(mergedPartValues,
+            rightSpec.partitioning.expressions,
+            rightReducers)
+          val rowOrdering = RowOrdering
+            .createNaturalAscendingOrdering(partitionExprs.map(_.dataType))
+          mergedPartValues = mergedPartValues.sorted(rowOrdering.on((t: (InternalRow, _)) => t._1))
+        }
+
+        // Now we need to push-down the common partition information to the scan in each child
+        newLeft = populateCommonPartitionInfo(left, mergedPartValues, leftSpec.joinKeyPositions,
+          leftReducers, applyPartialClustering, replicateLeftSide)
+        newRight = populateCommonPartitionInfo(right, mergedPartValues, rightSpec.joinKeyPositions,
+          rightReducers, applyPartialClustering, replicateRightSide)
       }
     }
 
@@ -527,25 +545,38 @@ case class EnsureRequirements(
         joinType == LeftAnti || joinType == LeftOuter
   }
 
-  // Populate the common partition values down to the scan nodes
-  private def populatePartitionValues(
+  // Populate the common partition information down to the scan nodes
+  private def populateCommonPartitionInfo(
       plan: SparkPlan,
       values: Seq[(InternalRow, Int)],
       joinKeyPositions: Option[Seq[Int]],
+      reducers: Option[Seq[Option[Reducer[Any]]]],
       applyPartialClustering: Boolean,
       replicatePartitions: Boolean): SparkPlan = plan match {
     case scan: BatchScanExec =>
       scan.copy(
         spjParams = scan.spjParams.copy(
           commonPartitionValues = Some(values),
           joinKeyPositions = joinKeyPositions,
+          reducers = reducers,
           applyPartialClustering = applyPartialClustering,
           replicatePartitions = replicatePartitions
         )
       )
     case node =>
-      node.mapChildren(child => populatePartitionValues(
-        child, values, joinKeyPositions, applyPartialClustering, replicatePartitions))
+      node.mapChildren(child => populateCommonPartitionInfo(
+        child, values, joinKeyPositions, reducers, applyPartialClustering, replicatePartitions))
+  }
+
+  private def reduceCommonPartValues(commonPartValues: Seq[(InternalRow, Int)],
+                                     expressions: Seq[Expression],
+                                     reducers: Option[Seq[Option[Reducer[Any]]]]) = {
+    reducers match {
+      case Some(reducers) => commonPartValues.groupBy { case (row, _) =>
+        KeyGroupedShuffleSpec.reducePartitionValue(row, expressions, reducers)
+      }.map{ case(wrapper, splits) => (wrapper.row, splits.map(_._2).sum) }.toSeq
+      case _ => commonPartValues
+    }
   }
 
   /**
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/connector/KeyGroupedPartitioningSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/connector/KeyGroupedPartitioningSuite.scala
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/connector/catalog/functions/transformFunctions.scala b/sql/core/src/test/scala/org/apache/spark/sql/connector/catalog/functions/transformFunctions.scala
