spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From GitBox <...@apache.org>
Subject [GitHub] [spark] c21 commented on a change in pull request #29655: [SPARK-32806][SQL] SortMergeJoin with partial hash distribution can be optimized to remove shuffle
Date Sat, 12 Sep 2020 20:52:34 GMT

c21 commented on a change in pull request #29655:
URL: https://github.com/apache/spark/pull/29655#discussion_r487378141



##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),
+        rSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            rChild,
+            rChildOutputPartitioning: HashPartitioning),
+          _),
+        false) if isPartialHashDistribution(
+          s.leftKeys, lChildOutputPartitioning, s.rightKeys, rChildOutputPartitioning) =>
+        // Remove ShuffleExchangeExec.
+        s.copy(left = lSort.copy(child = lChild), right = rSort.copy(child = rChild))
+      case other => other
+    }
+  }
+
+  /*
+   * Returns true if both HashPartitioning have the same number of partitions and
+   * their partitioning expressions are a subset of their respective join keys.
+   */
+  private def isPartialHashDistribution(
+      leftKeys: Seq[Expression],
+      leftPartitioning: HashPartitioning,
+      rightKeys: Seq[Expression],
+      rightPartitioning: HashPartitioning): Boolean = {
+    val mapping = leftKeyToRightKeyMapping(leftKeys, rightKeys)
+    (leftPartitioning.numPartitions == rightPartitioning.numPartitions) &&
+      leftPartitioning.expressions.zip(rightPartitioning.expressions)
+        .forall {
+          case (le, re) => mapping.get(le.canonicalized)
+            .map(_.exists(_.semanticEquals(re)))
+            .getOrElse(false)
+        }

Review comment:
       sorry if I miss anything, but I feel this might not be correct. We should make sure
the `leftPartitioning.expressions` and `rightPartitioning.expressions` has same size, and
the order of expressions matters, right?
   
   `expressions` size is different, so we should not remove shuffle:
   ```
   t1 has 1024 buckets on column (a)
   t2 has 1024 buckets on columns (a, b)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND t1.b = t2.b
   ```
   
   `expressions` size is same, but order is wrong, so we should not remove shuffle:
   
   ```
   t1 has 1024 buckets on column (a, b)
   t2 has 1024 buckets on columns (b, a)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND AND t1.a = t2.b AND t1.b = t2.a AND t1.b = t2.b
   ```

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),

Review comment:
       nit: why we can't just pattern matching `ShuffleExchangeExec(_, leftChild, _)` here?
It seems to be looking simpler to me.

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),
+        rSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            rChild,
+            rChildOutputPartitioning: HashPartitioning),
+          _),
+        false) if isPartialHashDistribution(
+          s.leftKeys, lChildOutputPartitioning, s.rightKeys, rChildOutputPartitioning) =>
+        // Remove ShuffleExchangeExec.
+        s.copy(left = lSort.copy(child = lChild), right = rSort.copy(child = rChild))
+      case other => other
+    }
+  }
+
+  /*
+   * Returns true if both HashPartitioning have the same number of partitions and
+   * their partitioning expressions are a subset of their respective join keys.
+   */
+  private def isPartialHashDistribution(
+      leftKeys: Seq[Expression],
+      leftPartitioning: HashPartitioning,
+      rightKeys: Seq[Expression],
+      rightPartitioning: HashPartitioning): Boolean = {
+    val mapping = leftKeyToRightKeyMapping(leftKeys, rightKeys)
+    (leftPartitioning.numPartitions == rightPartitioning.numPartitions) &&
+      leftPartitioning.expressions.zip(rightPartitioning.expressions)
+        .forall {
+          case (le, re) => mapping.get(le.canonicalized)
+            .map(_.exists(_.semanticEquals(re)))
+            .getOrElse(false)
+        }

Review comment:
       sorry if I miss anything, but I feel this might not be correct. We should make sure
the `leftPartitioning.expressions` and `rightPartitioning.expressions` has same size, and
the order of expressions matters, right?
   
   `expressions` size is different, so we should not remove shuffle:
   ```
   t1 has 1024 buckets on column (a)
   t2 has 1024 buckets on columns (a, b)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND t1.b = t2.b
   ```
   
   `expressions` size is same, but order is wrong, so we should not remove shuffle:
   
   ```
   t1 has 1024 buckets on column (a, b)
   t2 has 1024 buckets on columns (b, a)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND AND t1.a = t2.b AND t1.b = t2.a AND t1.b = t2.b
   ```

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),

Review comment:
       nit: why we can't just pattern matching `ShuffleExchangeExec(_, leftChild, _)` here?
It seems to be looking simpler to me.

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),
+        rSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            rChild,
+            rChildOutputPartitioning: HashPartitioning),
+          _),
+        false) if isPartialHashDistribution(
+          s.leftKeys, lChildOutputPartitioning, s.rightKeys, rChildOutputPartitioning) =>
+        // Remove ShuffleExchangeExec.
+        s.copy(left = lSort.copy(child = lChild), right = rSort.copy(child = rChild))
+      case other => other
+    }
+  }
+
+  /*
+   * Returns true if both HashPartitioning have the same number of partitions and
+   * their partitioning expressions are a subset of their respective join keys.
+   */
+  private def isPartialHashDistribution(
+      leftKeys: Seq[Expression],
+      leftPartitioning: HashPartitioning,
+      rightKeys: Seq[Expression],
+      rightPartitioning: HashPartitioning): Boolean = {
+    val mapping = leftKeyToRightKeyMapping(leftKeys, rightKeys)
+    (leftPartitioning.numPartitions == rightPartitioning.numPartitions) &&
+      leftPartitioning.expressions.zip(rightPartitioning.expressions)
+        .forall {
+          case (le, re) => mapping.get(le.canonicalized)
+            .map(_.exists(_.semanticEquals(re)))
+            .getOrElse(false)
+        }

Review comment:
       sorry if I miss anything, but I feel this might not be correct. We should make sure
the `leftPartitioning.expressions` and `rightPartitioning.expressions` has same size, and
the order of expressions matters, right?
   
   `expressions` size is different, so we should not remove shuffle:
   ```
   t1 has 1024 buckets on column (a)
   t2 has 1024 buckets on columns (a, b)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND t1.b = t2.b
   ```
   
   `expressions` size is same, but order is wrong, so we should not remove shuffle:
   
   ```
   t1 has 1024 buckets on column (a, b)
   t2 has 1024 buckets on columns (b, a)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND AND t1.a = t2.b AND t1.b = t2.a AND t1.b = t2.b
   ```

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),

Review comment:
       nit: why we can't just pattern matching `ShuffleExchangeExec(_, leftChild, _)` here?
It seems to be looking simpler to me.

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),
+        rSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            rChild,
+            rChildOutputPartitioning: HashPartitioning),
+          _),
+        false) if isPartialHashDistribution(
+          s.leftKeys, lChildOutputPartitioning, s.rightKeys, rChildOutputPartitioning) =>
+        // Remove ShuffleExchangeExec.
+        s.copy(left = lSort.copy(child = lChild), right = rSort.copy(child = rChild))
+      case other => other
+    }
+  }
+
+  /*
+   * Returns true if both HashPartitioning have the same number of partitions and
+   * their partitioning expressions are a subset of their respective join keys.
+   */
+  private def isPartialHashDistribution(
+      leftKeys: Seq[Expression],
+      leftPartitioning: HashPartitioning,
+      rightKeys: Seq[Expression],
+      rightPartitioning: HashPartitioning): Boolean = {
+    val mapping = leftKeyToRightKeyMapping(leftKeys, rightKeys)
+    (leftPartitioning.numPartitions == rightPartitioning.numPartitions) &&
+      leftPartitioning.expressions.zip(rightPartitioning.expressions)
+        .forall {
+          case (le, re) => mapping.get(le.canonicalized)
+            .map(_.exists(_.semanticEquals(re)))
+            .getOrElse(false)
+        }

Review comment:
       sorry if I miss anything, but I feel this might not be correct. We should make sure
the `leftPartitioning.expressions` and `rightPartitioning.expressions` has same size, and
the order of expressions matters, right?
   
   `expressions` size is different, so we should not remove shuffle:
   ```
   t1 has 1024 buckets on column (a)
   t2 has 1024 buckets on columns (a, b)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND t1.b = t2.b
   ```
   
   `expressions` size is same, but order is wrong, so we should not remove shuffle:
   
   ```
   t1 has 1024 buckets on column (a, b)
   t2 has 1024 buckets on columns (b, a)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND AND t1.a = t2.b AND t1.b = t2.a AND t1.b = t2.b
   ```

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),

Review comment:
       nit: why we can't just pattern matching `ShuffleExchangeExec(_, leftChild, _)` here?
It seems to be looking simpler to me.

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),
+        rSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            rChild,
+            rChildOutputPartitioning: HashPartitioning),
+          _),
+        false) if isPartialHashDistribution(
+          s.leftKeys, lChildOutputPartitioning, s.rightKeys, rChildOutputPartitioning) =>
+        // Remove ShuffleExchangeExec.
+        s.copy(left = lSort.copy(child = lChild), right = rSort.copy(child = rChild))
+      case other => other
+    }
+  }
+
+  /*
+   * Returns true if both HashPartitioning have the same number of partitions and
+   * their partitioning expressions are a subset of their respective join keys.
+   */
+  private def isPartialHashDistribution(
+      leftKeys: Seq[Expression],
+      leftPartitioning: HashPartitioning,
+      rightKeys: Seq[Expression],
+      rightPartitioning: HashPartitioning): Boolean = {
+    val mapping = leftKeyToRightKeyMapping(leftKeys, rightKeys)
+    (leftPartitioning.numPartitions == rightPartitioning.numPartitions) &&
+      leftPartitioning.expressions.zip(rightPartitioning.expressions)
+        .forall {
+          case (le, re) => mapping.get(le.canonicalized)
+            .map(_.exists(_.semanticEquals(re)))
+            .getOrElse(false)
+        }

Review comment:
       sorry if I miss anything, but I feel this might not be correct. We should make sure
the `leftPartitioning.expressions` and `rightPartitioning.expressions` has same size, and
the order of expressions matters, right?
   
   `expressions` size is different, so we should not remove shuffle:
   ```
   t1 has 1024 buckets on column (a)
   t2 has 1024 buckets on columns (a, b)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND t1.b = t2.b
   ```
   
   `expressions` size is same, but order is wrong, so we should not remove shuffle:
   
   ```
   t1 has 1024 buckets on column (a, b)
   t2 has 1024 buckets on columns (b, a)
   
   SELECT *
   FROM t1
   JOIN t2
   ON t1.a = t2.a AND AND t1.a = t2.b AND t1.b = t2.a AND t1.b = t2.b
   ```

##########
File path: sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/OptimizeSortMergeJoinWithPartialHashDistribution.scala
##########
@@ -0,0 +1,115 @@
+/*
+ * 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.exchange
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.catalyst.plans.physical.{HashPartitioning, Partitioning}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.execution.{SortExec, SparkPlan}
+import org.apache.spark.sql.execution.joins.SortMergeJoinExec
+import org.apache.spark.sql.internal.SQLConf
+
+/**
+ * This rule removes shuffle for the sort merge join if the following conditions are met:
+ * - The child of ShuffleExchangeExec has HashPartitioning with the same number of partitions
+ *   as the other side of join.
+ * - The child of ShuffleExchangeExec has output partitioning which has the subset of
+ *   join keys on the respective join side.
+ *
+ * If the above conditions are met, shuffle can be eliminated for the sort merge join
+ * because rows are sorted before join logic is applied.
+ */
+case class OptimizeSortMergeJoinWithPartialHashDistribution(conf: SQLConf) extends Rule[SparkPlan]
{
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.optimizeSortMergeJoinWithPartialHashDistribution) {
+      return plan
+    }
+
+    plan.transformUp {
+      case s @ SortMergeJoinExec(_, _, _, _,
+        lSort @ SortExec(_, _,
+          ExtractShuffleExchangeExecChild(
+            lChild,
+            lChildOutputPartitioning: HashPartitioning),
+          _),

Review comment:
       nit: why we can't just pattern matching `ShuffleExchangeExec(_, leftChild, _)` here?
It seems to be looking simpler to me.




----------------------------------------------------------------
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


Mime
View raw message