[SPARK-13306] [SQL] uncorrelated scalar subquery

sql/catalyst/src/main/antlr3/org/apache/spark/sql/catalyst/parser/ExpressionParser.g

@@ -205,6 +205,8 @@ atomExpression
     | whenExpression
     | (functionName LPAREN) => function
     | tableOrColumn
+    | (LPAREN KW_SELECT) => subQueryExpression
+      -> ^(TOK_SUBQUERY_EXPR ^(TOK_SUBQUERY_OP) subQueryExpression)
     | LPAREN! expression RPAREN!
     ;
 

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystQl.scala

@@ -667,6 +667,8 @@ https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation%2C+Cube%2C
           UnresolvedAttribute(nameParts :+ cleanIdentifier(attr))
         case other => UnresolvedExtractValue(other, Literal(cleanIdentifier(attr)))
       }
+    case Token("TOK_SUBQUERY_EXPR", Token("TOK_SUBQUERY_OP", Nil) :: subquery :: Nil) =>
+      ScalarSubquery(nodeToPlan(subquery))
 
     /* Stars (*) */
     case Token("TOK_ALLCOLREF", Nil) => UnresolvedStar(None)

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala

@@ -80,6 +80,7 @@ class Analyzer(
       ResolveGenerate ::
       ResolveFunctions ::
       ResolveAliases ::
+      ResolveSubquery ::
       ResolveWindowOrder ::
       ResolveWindowFrame ::
       ResolveNaturalJoin ::
@@ -120,7 +121,14 @@ class Analyzer(
             withAlias.getOrElse(relation)
           }
           substituted.getOrElse(u)
+        case other =>
+          // This can't be done in ResolveSubquery because that does not know the CTE.
+          other transformExpressions {
+            case e: SubqueryExpression =>
+              e.withNewPlan(substituteCTE(e.query, cteRelations))
+          }
       }
+
     }
   }
 
@@ -693,6 +701,30 @@ class Analyzer(
     }
   }
 
+  /**
+   * This rule resolve subqueries inside expressions.
+   *
+   * Note: CTE are handled in CTESubstitution.
+   */
+  object ResolveSubquery extends Rule[LogicalPlan] with PredicateHelper {
+
+    private def hasSubquery(e: Expression): Boolean = {
+      e.find(_.isInstanceOf[SubqueryExpression]).isDefined
+    }
+
+    private def hasSubquery(q: LogicalPlan): Boolean = {
+      q.expressions.exists(hasSubquery)
+    }
+
+    def apply(plan: LogicalPlan): LogicalPlan = plan resolveOperators {
+      case q: LogicalPlan if q.childrenResolved && hasSubquery(q) =>
+        q transformExpressions {
+          case e: SubqueryExpression if !e.query.resolved =>
+            e.withNewPlan(execute(e.query))
+        }
+    }
+  }
+
   /**
    * Turns projections that contain aggregate expressions into aggregations.
    */

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

@@ -0,0 +1,82 @@
+/*
+ * 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.expressions
+
+import org.apache.spark.sql.catalyst.analysis.TypeCheckResult
+import org.apache.spark.sql.catalyst.plans.QueryPlan
+import org.apache.spark.sql.catalyst.plans.logical.{LogicalPlan, Subquery}
+import org.apache.spark.sql.types.DataType
+
+/**
+ * An interface for subquery that is used in expressions.
+ */
+abstract class SubqueryExpression extends LeafExpression {
+
+  /**
+   * The logical plan of the query.
+   */
+  def query: LogicalPlan
+
+  /**
+   * Either a logical plan or a physical plan. The generated tree string (explain output) uses this
+   * field to explain the subquery.
+   */
+  def plan: QueryPlan[_]
+
+  /**
+   * Updates the query with new logical plan.
+   */
+  def withNewPlan(plan: LogicalPlan): SubqueryExpression
+}
+
+/**
+ * A subquery that will return only one row and one column.
+ *
+ * This will be converted into [[execution.ScalarSubquery]] during physical planning.
+ *
+ * Note: `exprId` is used to have unique name in explain string output.
+ */
+case class ScalarSubquery(
+    query: LogicalPlan,
+    exprId: ExprId = NamedExpression.newExprId)
+  extends SubqueryExpression with Unevaluable {
+
+  override def plan: LogicalPlan = Subquery(toString, query)
+
+  override lazy val resolved: Boolean = query.resolved
+
+  override def dataType: DataType = query.schema.fields.head.dataType
+
+  override def checkInputDataTypes(): TypeCheckResult = {
+    if (query.schema.length != 1) {
+      TypeCheckResult.TypeCheckFailure("Scalar subquery must return only one column, but got " +
+        query.schema.length.toString)
+    } else {
+      TypeCheckResult.TypeCheckSuccess
+    }
+  }
+
+  override def foldable: Boolean = false
+  override def nullable: Boolean = true
+
+  override def withNewPlan(plan: LogicalPlan): ScalarSubquery = ScalarSubquery(plan, exprId)
+
+  override def toString: String = s"subquery#${exprId.id}"
+
+  // TODO: support sql()
+}

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

@@ -88,7 +88,19 @@ abstract class Optimizer extends RuleExecutor[LogicalPlan] {
     Batch("Decimal Optimizations", FixedPoint(100),
       DecimalAggregates) ::
     Batch("LocalRelation", FixedPoint(100),
-      ConvertToLocalRelation) :: Nil
+      ConvertToLocalRelation) ::
+    Batch("Subquery", Once,
+      OptimizeSubqueries) :: Nil
+  }
+
+  /**
+   * Optimize all the subqueries inside expression.
+   */
+  object OptimizeSubqueries extends Rule[LogicalPlan] {
+    def apply(plan: LogicalPlan): LogicalPlan = plan transformAllExpressions {
+      case subquery: SubqueryExpression =>
+        subquery.withNewPlan(Optimizer.this.execute(subquery.query))
+    }
   }
 }
 

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala

@@ -18,6 +18,7 @@
 package org.apache.spark.sql.catalyst.plans
 
 import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.catalyst.plans.logical.Subquery
 import org.apache.spark.sql.catalyst.trees.TreeNode
 import org.apache.spark.sql.types.{DataType, StructType}
 
@@ -226,4 +227,29 @@ abstract class QueryPlan[PlanType <: TreeNode[PlanType]] extends TreeNode[PlanTy
   protected def statePrefix = if (missingInput.nonEmpty && children.nonEmpty) "!" else ""
 
   override def simpleString: String = statePrefix + super.simpleString
+
+  override def generateTreeString(
+      depth: Int, lastChildren: Seq[Boolean], builder: StringBuilder): StringBuilder = {
+    if (depth > 0) {
+      lastChildren.init.foreach { isLast =>
+        val prefixFragment = if (isLast) "   " else ":  "
+        builder.append(prefixFragment)
+      }
+
+      val branch = if (lastChildren.last) "+- " else ":- "
+      builder.append(branch)
+    }
+
+    builder.append(simpleString)
+    builder.append("\n")
+
+    val allSubqueries = expressions.flatMap(_.collect {case e: SubqueryExpression => e})
+    val allChildren = children ++ allSubqueries.map(e => e.plan)
+    if (allChildren.nonEmpty) {
+      allChildren.init.foreach(_.generateTreeString(depth + 1, lastChildren :+ false, builder))
+      allChildren.last.generateTreeString(depth + 1, lastChildren :+ true, builder)
+    }
+
+    builder
+  }
 }

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/CatalystQlSuite.scala

@@ -22,6 +22,7 @@ import org.apache.spark.sql.catalyst.analysis._
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.plans.PlanTest
 import org.apache.spark.sql.catalyst.plans.logical._
+import org.apache.spark.sql.types.BooleanType
 import org.apache.spark.unsafe.types.CalendarInterval
 
 class CatalystQlSuite extends PlanTest {
@@ -201,4 +202,10 @@ class CatalystQlSuite extends PlanTest {
     parser.parsePlan("select sum(product + 1) over (partition by (product + (1)) order by 2) " +
       "from windowData")
   }
+
+  test("subquery") {
+    parser.parsePlan("select (select max(b) from s) ss from t")
+    parser.parsePlan("select * from t where a = (select b from s)")
+    parser.parsePlan("select * from t group by g having a > (select b from s)")
+  }
 }

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisErrorSuite.scala

@@ -113,6 +113,17 @@ class AnalysisErrorSuite extends AnalysisTest {
 
   val dateLit = Literal.create(null, DateType)
 
+  errorTest(
+    "scalar subquery with 2 columns",
+     testRelation.select(
+       (ScalarSubquery(testRelation.select('a, dateLit.as('b))) + Literal(1)).as('a)),
+     "Scalar subquery must return only one column, but got 2" :: Nil)
+
+  errorTest(
+    "scalar subquery with no column",
+    testRelation.select(ScalarSubquery(LocalRelation()).as('a)),
+    "Scalar subquery must return only one column, but got 0" :: Nil)
+
   errorTest(
     "single invalid type, single arg",
     testRelation.select(TestFunction(dateLit :: Nil, IntegerType :: Nil).as('a)),

sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala

@@ -884,6 +884,7 @@ class SQLContext private[sql](
   @transient
   protected[sql] val prepareForExecution = new RuleExecutor[SparkPlan] {
     val batches = Seq(
+      Batch("Subquery", Once, PlanSubqueries(self)),
       Batch("Add exchange", Once, EnsureRequirements(self)),
       Batch("Whole stage codegen", Once, CollapseCodegenStages(self))
     )

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

@@ -20,6 +20,8 @@ package org.apache.spark.sql.execution
 import java.util.concurrent.atomic.AtomicBoolean
 
 import scala.collection.mutable.ArrayBuffer
+import scala.concurrent.{Await, ExecutionContext, Future}
+import scala.concurrent.duration._
 
 import org.apache.spark.Logging
 import org.apache.spark.rdd.{RDD, RDDOperationScope}
@@ -31,6 +33,7 @@ import org.apache.spark.sql.catalyst.plans.QueryPlan
 import org.apache.spark.sql.catalyst.plans.physical._
 import org.apache.spark.sql.execution.metric.{LongSQLMetric, SQLMetric}
 import org.apache.spark.sql.types.DataType
+import org.apache.spark.util.ThreadUtils
 
 /**
  * The base class for physical operators.
@@ -122,7 +125,34 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
   final def prepare(): Unit = {
     if (prepareCalled.compareAndSet(false, true)) {
       doPrepare()
+
+      // collect all the subqueries and submit jobs to execute them in background
+      val queryResults = ArrayBuffer[(ScalarSubquery, Future[Array[InternalRow]])]()
+      val allSubqueries = expressions.flatMap(_.collect {case e: ScalarSubquery => e})
+      allSubqueries.foreach { e =>
+        val futureResult = Future {
+          e.plan.executeTake(2)
+        }(SparkPlan.subqueryExecutionContext)
+        queryResults += e -> futureResult
+      }
+
       children.foreach(_.prepare())
+
+      // fill in the result of subqueries
+      queryResults.foreach {
+        case (e, futureResult) =>
+          val rows = Await.result(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, "Analyzer should make sure this only returns one column")
+            e.updateResult(rows(0).get(0, e.dataType))
+          } else {
+            // There is no rows returned, the result should be null.
+            e.updateResult(null)
+          }
+      }
     }
   }
 
@@ -231,6 +261,11 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
   }
 }
 
+object SparkPlan {
+  private[execution] val subqueryExecutionContext = ExecutionContext.fromExecutorService(
+    ThreadUtils.newDaemonCachedThreadPool("subquery", 16))
+}
+
 private[sql] trait LeafNode extends SparkPlan {
   override def children: Seq[SparkPlan] = Nil
   override def producedAttributes: AttributeSet = outputSet

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

@@ -343,3 +343,18 @@ case class OutputFaker(output: Seq[Attribute], child: SparkPlan) extends SparkPl
 
   protected override def doExecute(): RDD[InternalRow] = child.execute()
 }
+
+/**
+ * A plan as subquery.
+ *
+ * This is used to generate tree string for SparkScalarSubquery.
+ */
+case class Subquery(name: String, child: SparkPlan) extends UnaryNode {
+  override def output: Seq[Attribute] = child.output
+  override def outputPartitioning: Partitioning = child.outputPartitioning
+  override def outputOrdering: Seq[SortOrder] = child.outputOrdering
+
+  protected override def doExecute(): RDD[InternalRow] = {
+    child.execute()
+  }
+}