[GitHub] [spark] imback82 commented on a change in pull request #29079: [SPARK-32286][SQL] Coalesce bucketed table for shuffled hash join if applicable

[GitBox <gi...@apache.org>]

imback82 commented on a change in pull request #29079:
URL: https://github.com/apache/spark/pull/29079#discussion_r455948548



##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/CoalesceBucketsInJoin.scala
##########
@@ -0,0 +1,175 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.execution.bucketing
+
+import scala.annotation.tailrec
+
+import org.apache.spark.sql.catalyst.catalog.BucketSpec
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.optimizer.{BuildLeft, BuildRight}
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.{BaseJoinExec, ShuffledHashJoinExec, SortMergeJoinExec}
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule coalesces one side of the `SortMergeJoin` and `ShuffledHashJoin`
+ * if the following conditions are met:
+ *   - Two bucketed tables are joined.
+ *   - Join keys match with output partition expressions on their respective sides.
+ *   - The larger bucket number is divisible by the smaller bucket number.
+ *   - COALESCE_BUCKETS_IN_JOIN_ENABLED is set to true.
+ *   - The ratio of the number of buckets is less than the value set in
+ *     COALESCE_BUCKETS_IN_SORT_MERGE_JOIN_MAX_BUCKET_RATIO (`SortMergeJoin`) or,
+ *     COALESCE_BUCKETS_IN_SHUFFLED_HASH_JOIN_MAX_BUCKET_RATIO (`ShuffledHashJoin`).
+ */
+case class CoalesceBucketsInJoin(conf: SQLConf) extends Rule[SparkPlan] {
+  private def updateNumCoalescedBuckets(
+      join: BaseJoinExec,
+      numLeftBuckets: Int,
+      numRightBucket: Int,
+      numCoalescedBuckets: Int): BaseJoinExec = {
+    if (numCoalescedBuckets != numLeftBuckets) {
+      val leftCoalescedChild = join.left transformUp {
+        case f: FileSourceScanExec =>
+          f.copy(optionalNumCoalescedBuckets = Some(numCoalescedBuckets))
+      }
+      join match {
+        case j: SortMergeJoinExec => j.copy(left = leftCoalescedChild)
+        case j: ShuffledHashJoinExec => j.copy(left = leftCoalescedChild)
+      }
+    } else {
+      val rightCoalescedChild = join.right transformUp {
+        case f: FileSourceScanExec =>
+          f.copy(optionalNumCoalescedBuckets = Some(numCoalescedBuckets))
+      }
+      join match {
+        case j: SortMergeJoinExec => j.copy(right = rightCoalescedChild)
+        case j: ShuffledHashJoinExec => j.copy(right = rightCoalescedChild)
+      }
+    }
+  }
+
+  private def isCoalesceSHJStreamSide(
+      join: ShuffledHashJoinExec,
+      numLeftBuckets: Int,
+      numRightBucket: Int,
+      numCoalescedBuckets: Int): Boolean = {
+    if (numCoalescedBuckets == numLeftBuckets) {
+      join.buildSide != BuildRight
+    } else {
+      join.buildSide != BuildLeft
+    }
+  }
+
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.coalesceBucketsInJoinEnabled) {
+      return plan
+    }
+
+    plan transform {
+      case ExtractJoinWithBuckets(join, numLeftBuckets, numRightBuckets) =>
+        val bucketRatio = math.max(numLeftBuckets, numRightBuckets) /
+          math.min(numLeftBuckets, numRightBuckets)
+        val numCoalescedBuckets = math.min(numLeftBuckets, numRightBuckets)
+        join match {
+          case j: SortMergeJoinExec
+            if bucketRatio <= conf.coalesceBucketsInSortMergeJoinMaxBucketRatio =>
+            updateNumCoalescedBuckets(j, numLeftBuckets, numRightBuckets, numCoalescedBuckets)
+          case j: ShuffledHashJoinExec
+            // Only coalesce the buckets for shuffled hash join stream side,
+            // to avoid OOM for build side.
+            if bucketRatio <= conf.coalesceBucketsInShuffledHashJoinMaxBucketRatio &&
+              isCoalesceSHJStreamSide(j, numLeftBuckets, numRightBuckets, numCoalescedBuckets) =>
+            updateNumCoalescedBuckets(j, numLeftBuckets, numRightBuckets, numCoalescedBuckets)
+          case other => other
+        }
+      case other => other
+    }
+  }
+}
+
+/**
+ * An extractor that extracts `SortMergeJoinExec` and `ShuffledHashJoin`,
+ * where both sides of the join have the bucketed tables,
+ * are consisted of only the scan operation,
+ * and numbers of buckets are not equal but divisible.
+ */
+object ExtractJoinWithBuckets {
+  @tailrec
+  private def isScanOperation(plan: SparkPlan): Boolean = plan match {
+    case f: FilterExec => isScanOperation(f.child)
+    case p: ProjectExec => isScanOperation(p.child)
+    case _: FileSourceScanExec => true
+    case _ => false
+  }
+
+  private def getBucketSpec(plan: SparkPlan): Option[BucketSpec] = {
+    plan.collectFirst {
+      case f: FileSourceScanExec if f.relation.bucketSpec.nonEmpty &&
+          f.optionalNumCoalescedBuckets.isEmpty =>
+        f.relation.bucketSpec.get
+    }
+  }
+
+  /**
+   * The join keys should match with expressions for output partitioning. Note that
+   * the ordering does not matter because it will be handled in `EnsureRequirements`.
+   */
+  private def satisfiesOutputPartitioning(
+      keys: Seq[Expression],
+      partitioning: Partitioning): Boolean = {
+    partitioning match {
+      case HashPartitioning(exprs, _) if exprs.length == keys.length =>
+        exprs.forall(e => keys.exists(_.semanticEquals(e)))
+      case _ => false
+    }
+  }
+
+  private def isApplicable(j: BaseJoinExec): Boolean = {
+    (j.isInstanceOf[SortMergeJoinExec] ||
+      j.isInstanceOf[ShuffledHashJoinExec]) &&
+      isScanOperation(j.left) &&
+      isScanOperation(j.right) &&
+      satisfiesOutputPartitioning(j.leftKeys, j.left.outputPartitioning) &&
+      satisfiesOutputPartitioning(j.rightKeys, j.right.outputPartitioning)
+  }
+
+  private def isDivisible(numBuckets1: Int, numBuckets2: Int): Boolean = {
+    val (small, large) = (math.min(numBuckets1, numBuckets2), math.max(numBuckets1, numBuckets2))
+    // A bucket can be coalesced only if the bigger number of buckets is divisible by the smaller
+    // number of buckets because bucket id is calculated by modding the total number of buckets.
+    numBuckets1 != numBuckets2 && large % small == 0
+  }
+
+  def unapply(plan: SparkPlan): Option[(BaseJoinExec, Int, Int)] = {
+    plan match {
+      case s: BaseJoinExec if isApplicable(s) =>

Review comment:
       nit: Use `j` for join?




----------------------------------------------------------------
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:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org