[SPARK-16958] [SQL] Reuse subqueries within the same query

## What changes were proposed in this pull request?

There could be multiple subqueries that generate same results, we could re-use the result instead of running it multiple times.

This PR also cleanup up how we run subqueries.

For SQL query
```sql
select id,(select avg(id) from t) from t where id > (select avg(id) from t)
```
The explain is
```
== Physical Plan ==
*Project [id#15L, Subquery subquery29 AS scalarsubquery()#35]
:  +- Subquery subquery29
:     +- *HashAggregate(keys=[], functions=[avg(id#15L)])
:        +- Exchange SinglePartition
:           +- *HashAggregate(keys=[], functions=[partial_avg(id#15L)])
:              +- *Range (0, 1000, splits=4)
+- *Filter (cast(id#15L as double) > Subquery subquery29)
   :  +- Subquery subquery29
   :     +- *HashAggregate(keys=[], functions=[avg(id#15L)])
   :        +- Exchange SinglePartition
   :           +- *HashAggregate(keys=[], functions=[partial_avg(id#15L)])
   :              +- *Range (0, 1000, splits=4)
   +- *Range (0, 1000, splits=4)
```
The visualized plan:

![reuse-subquery](https://cloud.githubusercontent.com/assets/40902/17573229/e578d93c-5f0d-11e6-8a3c-0150d81d3aed.png)

## How was this patch tested?

Existing tests.

Author: Davies Liu <[email protected]>

Closes #14548 from davies/subq.

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/subquery.scala

@@ -102,6 +102,13 @@ case class PredicateSubquery(
   override def nullable: Boolean = nullAware
   override def plan: LogicalPlan = SubqueryAlias(toString, query)
   override def withNewPlan(plan: LogicalPlan): PredicateSubquery = copy(query = plan)
+  override def semanticEquals(o: Expression): Boolean = o match {
+    case p: PredicateSubquery =>
+      query.sameResult(p.query) && nullAware == p.nullAware &&
+        children.length == p.children.length &&
+        children.zip(p.children).forall(p => p._1.semanticEquals(p._2))
+    case _ => false
+  }
   override def toString: String = s"predicate-subquery#${exprId.id} $conditionString"
 }
 

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/trees/TreeNode.scala

@@ -538,9 +538,9 @@ abstract class TreeNode[BaseType <: TreeNode[BaseType]] extends Product {
 
     if (innerChildren.nonEmpty) {
       innerChildren.init.foreach(_.generateTreeString(
-        depth + 2, lastChildren :+ false :+ false, builder, verbose))
+        depth + 2, lastChildren :+ children.isEmpty :+ false, builder, verbose))
       innerChildren.last.generateTreeString(
-        depth + 2, lastChildren :+ false :+ true, builder, verbose)
+        depth + 2, lastChildren :+ children.isEmpty :+ true, builder, verbose)
     }
 
     if (children.nonEmpty) {

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

@@ -101,7 +101,8 @@ class QueryExecution(val sparkSession: SparkSession, val logical: LogicalPlan) {
     PlanSubqueries(sparkSession),
     EnsureRequirements(sparkSession.sessionState.conf),
     CollapseCodegenStages(sparkSession.sessionState.conf),
-    ReuseExchange(sparkSession.sessionState.conf))
+    ReuseExchange(sparkSession.sessionState.conf),
+    ReuseSubquery(sparkSession.sessionState.conf))
 
   protected def stringOrError[A](f: => A): String =
     try f.toString catch { case e: Throwable => e.toString }

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

@@ -142,21 +142,18 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
    * This list is populated by [[prepareSubqueries]], which is called in [[prepare]].
    */
   @transient
-  private val subqueryResults = new ArrayBuffer[(ScalarSubquery, Future[Array[InternalRow]])]
+  private val runningSubqueries = new ArrayBuffer[ExecSubqueryExpression]
 
   /**
    * Finds scalar subquery expressions in this plan node and starts evaluating them.
-   * The list of subqueries are added to [[subqueryResults]].
    */
   protected def prepareSubqueries(): Unit = {
-    val allSubqueries = expressions.flatMap(_.collect {case e: ScalarSubquery => e})
-    allSubqueries.asInstanceOf[Seq[ScalarSubquery]].foreach { e =>
-      val futureResult = Future {
-        // Each subquery should return only one row (and one column). We take two here and throws
-        // an exception later if the number of rows is greater than one.
-        e.executedPlan.executeTake(2)
-      }(SparkPlan.subqueryExecutionContext)
-      subqueryResults += e -> futureResult
+    expressions.foreach {
+      _.collect {
+        case e: ExecSubqueryExpression =>
+          e.plan.prepare()
+          runningSubqueries += e
+      }
     }
   }
 
@@ -165,21 +162,10 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
    */
   protected def waitForSubqueries(): Unit = synchronized {
     // fill in the result of subqueries
-    subqueryResults.foreach { case (e, futureResult) =>
-      val rows = ThreadUtils.awaitResult(futureResult, Duration.Inf)
-      if (rows.length > 1) {
-        sys.error(s"more than one row returned by a subquery used as an expression:\n${e.plan}")
-      }
-      if (rows.length == 1) {
-        assert(rows(0).numFields == 1,
-          s"Expects 1 field, but got ${rows(0).numFields}; something went wrong in analysis")
-        e.updateResult(rows(0).get(0, e.dataType))
-      } else {
-        // If there is no rows returned, the result should be null.
-        e.updateResult(null)
-      }
+    runningSubqueries.foreach { sub =>
+      sub.updateResult()
     }
-    subqueryResults.clear()
+    runningSubqueries.clear()
   }
 
   /**

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

@@ -17,13 +17,19 @@
 
 package org.apache.spark.sql.execution
 
+import scala.concurrent.{ExecutionContext, Future}
+import scala.concurrent.duration.Duration
+
+import org.apache.spark.SparkException
 import org.apache.spark.rdd.{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.types.{LongType, StructField, StructType}
+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}
 
 /** Physical plan for Project. */
@@ -502,15 +508,64 @@ case class OutputFakerExec(output: Seq[Attribute], child: SparkPlan) extends Spa
 
 /**
  * Physical plan for a subquery.
- *
- * This is used to generate tree string for SparkScalarSubquery.
  */
 case class SubqueryExec(name: String, child: SparkPlan) extends UnaryExecNode {
+
+  override lazy val metrics = Map(
+    "dataSize" -> SQLMetrics.createMetric(sparkContext, "data size (bytes)"),
+    "collectTime" -> SQLMetrics.createMetric(sparkContext, "time to collect (ms)"))
+
   override def output: Seq[Attribute] = child.output
   override def outputPartitioning: Partitioning = child.outputPartitioning
   override def outputOrdering: Seq[SortOrder] = child.outputOrdering
 
+  override def sameResult(o: SparkPlan): Boolean = o match {
+    case s: SubqueryExec => child.sameResult(s.child)
+    case _ => false
+  }
+
+  @transient
+  private lazy val relationFuture: Future[Array[InternalRow]] = {
+    // relationFuture is used in "doExecute". Therefore we can get the execution id correctly here.
+    val executionId = sparkContext.getLocalProperty(SQLExecution.EXECUTION_ID_KEY)
+    Future {
+      // This will run in another thread. Set the execution id so that we can connect these jobs
+      // with the correct execution.
+      SQLExecution.withExecutionId(sparkContext, executionId) {
+        val beforeCollect = System.nanoTime()
+        // Note that we use .executeCollect() because we don't want to convert data to Scala types
+        val rows: Array[InternalRow] = child.executeCollect()
+        val beforeBuild = System.nanoTime()
+        longMetric("collectTime") += (beforeBuild - beforeCollect) / 1000000
+        val dataSize = rows.map(_.asInstanceOf[UnsafeRow].getSizeInBytes.toLong).sum
+        longMetric("dataSize") += dataSize
+
+        // There are some cases we don't care about the metrics and call `SparkPlan.doExecute`
+        // directly without setting an execution id. We should be tolerant to it.
+        if (executionId != null) {
+          sparkContext.listenerBus.post(SparkListenerDriverAccumUpdates(
+            executionId.toLong, metrics.values.map(m => m.id -> m.value).toSeq))
+        }
+
+        rows
+      }
+    }(SubqueryExec.executionContext)
+  }
+
+  protected override def doPrepare(): Unit = {
+    relationFuture
+  }
+
   protected override def doExecute(): RDD[InternalRow] = {
-    throw new UnsupportedOperationException
+    child.execute()
   }
+
+  override def executeCollect(): Array[InternalRow] = {
+    ThreadUtils.awaitResult(relationFuture, Duration.Inf)
+  }
+}
+
+object SubqueryExec {
+  private[execution] val executionContext = ExecutionContext.fromExecutorService(
+    ThreadUtils.newDaemonCachedThreadPool("subquery", 16))
 }

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

@@ -17,42 +17,78 @@
 
 package org.apache.spark.sql.execution
 
+import scala.collection.mutable
+import scala.collection.mutable.ArrayBuffer
+
 import org.apache.spark.sql.SparkSession
-import org.apache.spark.sql.catalyst.expressions
-import org.apache.spark.sql.catalyst.InternalRow
-import org.apache.spark.sql.catalyst.expressions.{Expression, ExprId, Literal, SubqueryExpression}
+import org.apache.spark.sql.catalyst.{expressions, InternalRow}
+import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.expressions.codegen.{CodegenContext, ExprCode}
 import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
 import org.apache.spark.sql.catalyst.rules.Rule
-import org.apache.spark.sql.types.DataType
+import org.apache.spark.sql.internal.SQLConf
+import org.apache.spark.sql.types.{BooleanType, DataType, StructType}
+
+/**
+ * The base class for subquery that is used in SparkPlan.
+ */
+trait ExecSubqueryExpression extends SubqueryExpression {
+
+  val executedPlan: SubqueryExec
+  def withExecutedPlan(plan: SubqueryExec): ExecSubqueryExpression
+
+  // does not have logical plan
+  override def query: LogicalPlan = throw new UnsupportedOperationException
+  override def withNewPlan(plan: LogicalPlan): SubqueryExpression =
+    throw new UnsupportedOperationException
+
+  override def plan: SparkPlan = executedPlan
+
+  /**
+   * Fill the expression with collected result from executed plan.
+   */
+  def updateResult(): Unit
+}
 
 /**
  * A subquery that will return only one row and one column.
  *
  * This is the physical copy of ScalarSubquery to be used inside SparkPlan.
  */
 case class ScalarSubquery(
-    executedPlan: SparkPlan,
+    executedPlan: SubqueryExec,
     exprId: ExprId)
-  extends SubqueryExpression {
-
-  override def query: LogicalPlan = throw new UnsupportedOperationException
-  override def withNewPlan(plan: LogicalPlan): SubqueryExpression = {
-    throw new UnsupportedOperationException
-  }
-  override def plan: SparkPlan = SubqueryExec(simpleString, executedPlan)
+  extends ExecSubqueryExpression {
 
   override def dataType: DataType = executedPlan.schema.fields.head.dataType
   override def children: Seq[Expression] = Nil
   override def nullable: Boolean = true
-  override def toString: String = s"subquery#${exprId.id}"
+  override def toString: String = executedPlan.simpleString
+
+  def withExecutedPlan(plan: SubqueryExec): ExecSubqueryExpression = copy(executedPlan = plan)
+
+  override def semanticEquals(other: Expression): Boolean = other match {
+    case s: ScalarSubquery => executedPlan.sameResult(executedPlan)
+    case _ => false
+  }
 
   // the first column in first row from `query`.
   @volatile private var result: Any = null
   @volatile private var updated: Boolean = false
 
-  def updateResult(v: Any): Unit = {
-    result = v
+  def updateResult(): Unit = {
+    val rows = plan.executeCollect()
+    if (rows.length > 1) {
+      sys.error(s"more than one row returned by a subquery used as an expression:\n${plan}")
+    }
+    if (rows.length == 1) {
+      assert(rows(0).numFields == 1,
+        s"Expects 1 field, but got ${rows(0).numFields}; something went wrong in analysis")
+      result = rows(0).get(0, dataType)
+    } else {
+      // If there is no rows returned, the result should be null.
+      result = null
+    }
     updated = true
   }
 
@@ -67,6 +103,51 @@ case class ScalarSubquery(
   }
 }
 
+/**
+ * A subquery that will check the value of `child` whether is in the result of a query or not.
+ */
+case class InSubquery(
+    child: Expression,
+    executedPlan: SubqueryExec,
+    exprId: ExprId,
+    private var result: Array[Any] = null,
+    private var updated: Boolean = false) extends ExecSubqueryExpression {
+
+  override def dataType: DataType = BooleanType
+  override def children: Seq[Expression] = child :: Nil
+  override def nullable: Boolean = child.nullable
+  override def toString: String = s"$child IN ${executedPlan.name}"
+
+  def withExecutedPlan(plan: SubqueryExec): ExecSubqueryExpression = copy(executedPlan = plan)
+
+  override def semanticEquals(other: Expression): Boolean = other match {
+    case in: InSubquery => child.semanticEquals(in.child) &&
+      executedPlan.sameResult(in.executedPlan)
+    case _ => false
+  }
+
+  def updateResult(): Unit = {
+    val rows = plan.executeCollect()
+    result = rows.map(_.get(0, child.dataType)).asInstanceOf[Array[Any]]
+    updated = true
+  }
+
+  override def eval(input: InternalRow): Any = {
+    require(updated, s"$this has not finished")
+    val v = child.eval(input)
+    if (v == null) {
+      null
+    } else {
+      result.contains(v)
+    }
+  }
+
+  override def doGenCode(ctx: CodegenContext, ev: ExprCode): ExprCode = {
+    require(updated, s"$this has not finished")
+    InSet(child, result.toSet).doGenCode(ctx, ev)
+  }
+}
+
 /**
  * Plans scalar subqueries from that are present in the given [[SparkPlan]].
  */
@@ -75,7 +156,39 @@ case class PlanSubqueries(sparkSession: SparkSession) extends Rule[SparkPlan] {
     plan.transformAllExpressions {
       case subquery: expressions.ScalarSubquery =>
         val executedPlan = new QueryExecution(sparkSession, subquery.plan).executedPlan
-        ScalarSubquery(executedPlan, subquery.exprId)
+        ScalarSubquery(
+          SubqueryExec(s"subquery${subquery.exprId.id}", executedPlan),
+          subquery.exprId)
+      case expressions.PredicateSubquery(plan, Seq(e: Expression), _, exprId) =>
+        val executedPlan = new QueryExecution(sparkSession, plan).executedPlan
+        InSubquery(e, SubqueryExec(s"subquery${exprId.id}", executedPlan), exprId)
+    }
+  }
+}
+
+
+/**
+ * Find out duplicated exchanges in the spark plan, then use the same exchange for all the
+ * references.
+ */
+case class ReuseSubquery(conf: SQLConf) extends Rule[SparkPlan] {
+
+  def apply(plan: SparkPlan): SparkPlan = {
+    if (!conf.exchangeReuseEnabled) {
+      return plan
+    }
+    // Build a hash map using schema of exchanges to avoid O(N*N) sameResult calls.
+    val subqueries = mutable.HashMap[StructType, ArrayBuffer[SubqueryExec]]()
+    plan transformAllExpressions {
+      case sub: ExecSubqueryExpression =>
+        val sameSchema = subqueries.getOrElseUpdate(sub.plan.schema, ArrayBuffer[SubqueryExec]())
+        val sameResult = sameSchema.find(_.sameResult(sub.plan))
+        if (sameResult.isDefined) {
+          sub.withExecutedPlan(sameResult.get)
+        } else {
+          sameSchema += sub.executedPlan
+          sub
+        }
     }
   }
 }

sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala

@@ -99,7 +99,11 @@ object SparkPlanGraph {
       case "Subquery" if subgraph != null =>
         // Subquery should not be included in WholeStageCodegen
         buildSparkPlanGraphNode(planInfo, nodeIdGenerator, nodes, edges, parent, null, exchanges)
-      case "ReusedExchange" =>
+      case "Subquery" if exchanges.contains(planInfo) =>
+        // Point to the re-used subquery
+        val node = exchanges(planInfo)
+        edges += SparkPlanGraphEdge(node.id, parent.id)
+      case "ReusedExchange" if exchanges.contains(planInfo.children.head) =>
         // Point to the re-used exchange
         val node = exchanges(planInfo.children.head)
         edges += SparkPlanGraphEdge(node.id, parent.id)
@@ -115,7 +119,7 @@ object SparkPlanGraph {
         } else {
           subgraph.nodes += node
         }
-        if (name.contains("Exchange")) {
+        if (name.contains("Exchange") || name == "Subquery") {
           exchanges += planInfo -> node
         }