[SPARK-20636] Add new optimization rule to transpose adjacent Window expressions.

## What changes were proposed in this pull request?

Add new optimization rule to eliminate unnecessary shuffling by flipping adjacent Window expressions.

## How was this patch tested?

Tested with unit tests, integration tests, and manual tests.

Closes #17899 from ptkool/adjacent_window_optimization.

Authored-by: ptkool <[email protected]>
Signed-off-by: gatorsmile <[email protected]>

Author: ptkool

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

@@ -734,6 +734,28 @@ object CollapseWindow extends Rule[LogicalPlan] {
   }
 }
 
+/**
+ * Transpose Adjacent Window Expressions.
+ * - If the partition spec of the parent Window expression is compatible with the partition spec
+ *   of the child window expression, transpose them.
+ */
+object TransposeWindow extends Rule[LogicalPlan] {
+  private def compatibleParititions(ps1 : Seq[Expression], ps2: Seq[Expression]): Boolean = {
+    ps1.length < ps2.length && ps2.take(ps1.length).permutations.exists(ps1.zip(_).forall {
+      case (l, r) => l.semanticEquals(r)
+    })
+  }
+
+  def apply(plan: LogicalPlan): LogicalPlan = plan transformUp {
+    case w1 @ Window(we1, ps1, os1, w2 @ Window(we2, ps2, os2, grandChild))
+        if w1.references.intersect(w2.windowOutputSet).isEmpty &&
+           w1.expressions.forall(_.deterministic) &&
+           w2.expressions.forall(_.deterministic) &&
+           compatibleParititions(ps1, ps2) =>
+      Project(w1.output, Window(we2, ps2, os2, Window(we1, ps1, os1, grandChild)))
+  }
+}
+
 /**
  * Generate a list of additional filters from an operator's existing constraint but remove those
  * that are either already part of the operator's condition or are part of the operator's child

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/TransposeWindowSuite.scala

@@ -0,0 +1,114 @@
+/*
+ * 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.catalyst.optimizer
+
+import org.apache.spark.sql.catalyst.dsl.expressions._
+import org.apache.spark.sql.catalyst.dsl.plans._
+import org.apache.spark.sql.catalyst.expressions.Rand
+import org.apache.spark.sql.catalyst.plans.PlanTest
+import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan}
+import org.apache.spark.sql.catalyst.rules.RuleExecutor
+
+class TransposeWindowSuite extends PlanTest {
+  object Optimize extends RuleExecutor[LogicalPlan] {
+    val batches =
+      Batch("CollapseProject", FixedPoint(100), CollapseProject, RemoveRedundantProject) ::
+      Batch("FlipWindow", Once, CollapseWindow, TransposeWindow) :: Nil
+  }
+
+  val testRelation = LocalRelation('a.string, 'b.string, 'c.int, 'd.string)
+
+  val a = testRelation.output(0)
+  val b = testRelation.output(1)
+  val c = testRelation.output(2)
+  val d = testRelation.output(3)
+
+  val partitionSpec1 = Seq(a)
+  val partitionSpec2 = Seq(a, b)
+  val partitionSpec3 = Seq(d)
+  val partitionSpec4 = Seq(b, a, d)
+
+  val orderSpec1 = Seq(d.asc)
+  val orderSpec2 = Seq(d.desc)
+
+  test("transpose two adjacent windows with compatible partitions") {
+    val query = testRelation
+      .window(Seq(sum(c).as('sum_a_2)), partitionSpec2, orderSpec2)
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec1, orderSpec1)
+
+    val analyzed = query.analyze
+    val optimized = Optimize.execute(analyzed)
+
+    val correctAnswer = testRelation
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec1, orderSpec1)
+      .window(Seq(sum(c).as('sum_a_2)), partitionSpec2, orderSpec2)
+      .select('a, 'b, 'c, 'd, 'sum_a_2, 'sum_a_1)
+
+    comparePlans(optimized, correctAnswer.analyze)
+  }
+
+  test("transpose two adjacent windows with differently ordered compatible partitions") {
+    val query = testRelation
+      .window(Seq(sum(c).as('sum_a_2)), partitionSpec4, Seq.empty)
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec2, Seq.empty)
+
+    val analyzed = query.analyze
+    val optimized = Optimize.execute(analyzed)
+
+    val correctAnswer = testRelation
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec2, Seq.empty)
+      .window(Seq(sum(c).as('sum_a_2)), partitionSpec4, Seq.empty)
+      .select('a, 'b, 'c, 'd, 'sum_a_2, 'sum_a_1)
+
+    comparePlans(optimized, correctAnswer.analyze)
+  }
+
+  test("don't transpose two adjacent windows with incompatible partitions") {
+    val query = testRelation
+      .window(Seq(sum(c).as('sum_a_2)), partitionSpec3, Seq.empty)
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec1, Seq.empty)
+
+    val analyzed = query.analyze
+    val optimized = Optimize.execute(analyzed)
+
+    comparePlans(optimized, analyzed)
+  }
+
+  test("don't transpose two adjacent windows with intersection of partition and output set") {
+    val query = testRelation
+      .window(Seq(('a + 'b).as('e), sum(c).as('sum_a_2)), partitionSpec3, Seq.empty)
+      .window(Seq(sum(c).as('sum_a_1)), Seq(a, 'e), Seq.empty)
+
+    val analyzed = query.analyze
+    val optimized = Optimize.execute(analyzed)
+
+    comparePlans(optimized, analyzed)
+  }
+
+  test("don't transpose two adjacent windows with non-deterministic expressions") {
+    val query = testRelation
+      .window(Seq(Rand(0).as('e), sum(c).as('sum_a_2)), partitionSpec3, Seq.empty)
+      .window(Seq(sum(c).as('sum_a_1)), partitionSpec1, Seq.empty)
+
+    val analyzed = query.analyze
+    val optimized = Optimize.execute(analyzed)
+
+    comparePlans(optimized, analyzed)
+  }
+
+}

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

@@ -30,6 +30,7 @@ import org.apache.spark.sql.types._
  * Window function testing for DataFrame API.
  */
 class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
+
   import testImplicits._
 
   test("reuse window partitionBy") {
@@ -72,9 +73,9 @@ class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
         cume_dist().over(Window.partitionBy("value").orderBy("key")),
         percent_rank().over(Window.partitionBy("value").orderBy("key"))),
       Row(1, 1, 1, 1.0d, 1, 1, 1, 1, 1, 1, 1.0d, 0.0d) ::
-      Row(1, 1, 1, 1.0d, 1, 1, 1, 1, 1, 1, 1.0d / 3.0d, 0.0d) ::
-      Row(2, 2, 1, 5.0d / 3.0d, 3, 5, 1, 2, 2, 2, 1.0d, 0.5d) ::
-      Row(2, 2, 1, 5.0d / 3.0d, 3, 5, 2, 3, 2, 2, 1.0d, 0.5d) :: Nil)
+        Row(1, 1, 1, 1.0d, 1, 1, 1, 1, 1, 1, 1.0d / 3.0d, 0.0d) ::
+        Row(2, 2, 1, 5.0d / 3.0d, 3, 5, 1, 2, 2, 2, 1.0d, 0.5d) ::
+        Row(2, 2, 1, 5.0d / 3.0d, 3, 5, 2, 3, 2, 2, 1.0d, 0.5d) :: Nil)
   }
 
   test("window function should fail if order by clause is not specified") {
@@ -162,12 +163,12 @@ class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
       Seq(
         Row("a", -50.0, 50.0, 50.0, 7.0710678118654755, 7.0710678118654755),
         Row("b", -50.0, 50.0, 50.0, 7.0710678118654755, 7.0710678118654755),
-        Row("c", 0.0, 0.0, 0.0, 0.0, 0.0 ),
-        Row("d", 0.0, 0.0, 0.0, 0.0, 0.0 ),
-        Row("e", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544 ),
-        Row("f", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544 ),
-        Row("g", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544 ),
-        Row("h", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544 ),
+        Row("c", 0.0, 0.0, 0.0, 0.0, 0.0),
+        Row("d", 0.0, 0.0, 0.0, 0.0, 0.0),
+        Row("e", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544),
+        Row("f", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544),
+        Row("g", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544),
+        Row("h", 24.0, 12.0, 12.0, 3.4641016151377544, 3.4641016151377544),
         Row("i", Double.NaN, Double.NaN, Double.NaN, Double.NaN, Double.NaN)))
   }
 
@@ -326,7 +327,7 @@ class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
         var_samp($"value").over(window),
         approx_count_distinct($"value").over(window)),
       Seq.fill(4)(Row("a", 1.0d / 4.0d, 1.0d / 3.0d, 2))
-      ++ Seq.fill(3)(Row("b", 2.0d / 3.0d, 1.0d, 3)))
+        ++ Seq.fill(3)(Row("b", 2.0d / 3.0d, 1.0d, 3)))
   }
 
   test("window function with aggregates") {
@@ -624,7 +625,7 @@ class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
 
   test("SPARK-24575: Window functions inside WHERE and HAVING clauses") {
     def checkAnalysisError(df: => DataFrame): Unit = {
-      val thrownException = the [AnalysisException] thrownBy {
+      val thrownException = the[AnalysisException] thrownBy {
         df.queryExecution.analyzed
       }
       assert(thrownException.message.contains("window functions inside WHERE and HAVING clauses"))
@@ -658,4 +659,26 @@ class DataFrameWindowFunctionsSuite extends QueryTest with SharedSQLContext {
            |GROUP BY a
            |HAVING SUM(b) = 5 AND RANK() OVER(ORDER BY a) = 1""".stripMargin))
   }
+
+  test("window functions in multiple selects") {
+    val df = Seq(
+      ("S1", "P1", 100),
+      ("S1", "P1", 700),
+      ("S2", "P1", 200),
+      ("S2", "P2", 300)
+    ).toDF("sno", "pno", "qty")
+
+    val w1 = Window.partitionBy("sno")
+    val w2 = Window.partitionBy("sno", "pno")
+
+    checkAnswer(
+      df.select($"sno", $"pno", $"qty", sum($"qty").over(w2).alias("sum_qty_2"))
+        .select($"sno", $"pno", $"qty", col("sum_qty_2"), sum("qty").over(w1).alias("sum_qty_1")),
+      Seq(
+        Row("S1", "P1", 100, 800, 800),
+        Row("S1", "P1", 700, 800, 800),
+        Row("S2", "P1", 200, 200, 500),
+        Row("S2", "P2", 300, 300, 500)))
+
+  }
 }