Github user chenghao-intel commented on a diff in the pull request:
https://github.com/apache/spark/pull/3249#discussion_r25552386
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
---
@@ -422,6 +424,108 @@ class Analyzer(catalog: Catalog,
Generate(g, join = false, outer = false, None, child)
}
}
+
+ /**
+ * Transforms the query which has subquery expressions in where clause
to left semi join.
+ * select T1.x from T1 where T1.x in (select T2.y from T2) transformed to
+ * select T1.x from T1 left semi join T2 on T1.x = T2.y.
+ */
+ object SubQueryExpressions extends Rule[LogicalPlan] {
+
+ def apply(plan: LogicalPlan): LogicalPlan = plan transform {
+ case p: LogicalPlan if !p.childrenResolved => p
+ case filter @ Filter(conditions, child) =>
+ val subqueryExprs = conditions.collect {
+ case In(exp, Seq(SubqueryExpression(subquery))) => (exp,
subquery)
+ }
+ // Replace subqueries with a dummy true literal since they are
evaluated separately now.
+ val transformedConds = conditions.transform {
+ case In(_, Seq(SubqueryExpression(_))) => Literal(true)
+ }
+ subqueryExprs match {
+ case Seq() => filter // No subqueries.
+ case Seq((exp, subquery)) =>
+ createLeftSemiJoin(
+ child,
+ exp,
+ subquery,
+ transformedConds)
+ case _ =>
+ throw new TreeNodeException(filter, "Only one SubQuery
expression is supported.")
+ }
+ }
+
+ /**
+ * Create LeftSemi join with parent query to the subquery which is
mentioned in 'IN' predicate
+ * And combine the subquery conditions and parent query conditions.
+ */
+ def createLeftSemiJoin(left: LogicalPlan,
+ value: Expression,
+ subquery: LogicalPlan,
+ parentConds: Expression) : LogicalPlan = {
+ val (transformedPlan, subqueryConds) =
transformAndGetConditions(value, subquery)
+ // Add both parent query conditions and subquery conditions as join
conditions
+ val allPredicates = And(parentConds, subqueryConds)
+ Join(left, transformedPlan, LeftSemi, Some(allPredicates))
+ }
+
+ /**
+ * Transform the subquery LogicalPlan and add the expressions which
are used as filters to the
+ * projection. And also return filter conditions used in subquery
+ */
+ def transformAndGetConditions(value: Expression,
+ subquery: LogicalPlan): (LogicalPlan, Expression) = {
+ val expr = new scala.collection.mutable.ArrayBuffer[Expression]()
+ // TODO : we only decorelate subqueries in very specific cases like
the cases mentioned above
+ // in documentation. The more complex queries like using of
subqueries inside subqueries can
+ // be supported in future.
+ val transformedPlan = subquery transform {
+ case project @ Project(projectList, f @ Filter(condition, child))
=>
+ // Don't support more than one item in select list of subquery
+ if(projectList.size > 1) {
+ throw new TreeNodeException(
+ project,
+ "SubQuery can contain only one item in Select List")
+ }
+ val resolvedChild = ResolveRelations(child)
+ // Add the expressions to the projections which are used as
filters in subquery
+ val toBeAddedExprs = f.references.filter{a =>
+ resolvedChild.resolve(a.name, resolver) != None &&
!project.outputSet.contains(a)}
+ val nameToExprMap = collection.mutable.Map[String, Alias]()
+ // Create aliases for all projection expressions.
+ val witAliases = (projectList ++
toBeAddedExprs).zipWithIndex.map {
+ case (exp, index) =>
+ nameToExprMap.put(exp.name, Alias(exp, s"sqc$index")())
+ Alias(exp, s"sqc$index")()
+ }
+ // Replace the condition column names with alias names.
+ val transformedConds = condition.transform {
+ case a: Attribute if resolvedChild.resolve(a.name, resolver)
!= None =>
+ nameToExprMap.get(a.name).get.toAttribute
+ }
+ // Join the first projection column of subquery to the main
query and add as condition
+ // TODO : We can avoid if the parent condition already has this
condition.
+ expr += EqualTo(value, witAliases(0).toAttribute)
+ expr += transformedConds
--- End diff --
https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala#L544
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
