You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by "peter-toth (via GitHub)" <gi...@apache.org> on 2023/08/01 15:10:01 UTC
[GitHub] [spark] peter-toth commented on a diff in pull request #42223: [SPARK-44571][SQL] Eliminate the Join by combine multiple Aggregates
peter-toth commented on code in PR #42223:
URL: https://github.com/apache/spark/pull/42223#discussion_r1280788640
##########
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/EliminateJoinByCombineAggregate.scala:
##########
@@ -0,0 +1,196 @@
+/*
+ * 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 scala.collection.mutable.ArrayBuffer
+
+import org.apache.spark.sql.catalyst.expressions.{Alias, Attribute, Expression, NamedExpression, Or}
+import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression
+import org.apache.spark.sql.catalyst.plans.{Cross, FullOuter, Inner, JoinType, LeftOuter, RightOuter}
+import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, Filter, Join, LeafNode, LogicalPlan, Project, SerializeFromObject}
+import org.apache.spark.sql.catalyst.rules.Rule
+import org.apache.spark.sql.catalyst.trees.TreePattern.{AGGREGATE, JOIN}
+
+/**
+ * This rule eliminates the [[Join]] if all the join side are [[Aggregate]]s by combine these
+ * [[Aggregate]]s. This rule also support the nested [[Join]], as long as all the join sides for
+ * every [[Join]] are [[Aggregate]]s.
+ *
+ * Note: The [[Aggregate]]s to be merged must not exists filter.
+ */
+object EliminateJoinByCombineAggregate extends Rule[LogicalPlan] {
+
+ private def isSupportedJoinType(joinType: JoinType): Boolean =
+ Seq(Inner, Cross, LeftOuter, RightOuter, FullOuter).contains(joinType)
+
+ // Collect all the Aggregates from both side of single or nested Join.
+ private def collectAggregate(plan: LogicalPlan, aggregates: ArrayBuffer[Aggregate]): Boolean = {
+ var flag = true
+ if (plan.containsAnyPattern(JOIN, AGGREGATE)) {
+ plan match {
+ case Join(left: Aggregate, right: Aggregate, _, None, _)
+ if left.groupingExpressions.isEmpty && right.groupingExpressions.isEmpty &&
+ left.aggregateExpressions.forall(filterNotDefined) &&
+ right.aggregateExpressions.forall(filterNotDefined) =>
+ aggregates += left
+ aggregates += right
+ case Join(left @ Join(_, _, joinType, None, _), right: Aggregate, _, None, _)
+ if isSupportedJoinType(joinType) && right.groupingExpressions.isEmpty &&
+ right.aggregateExpressions.forall(filterNotDefined) =>
+ flag = collectAggregate(left, aggregates)
+ aggregates += right
+ case Join(left: Aggregate, right @ Join(_, _, joinType, None, _), _, None, _)
+ if isSupportedJoinType(joinType) && left.groupingExpressions.isEmpty &&
+ left.aggregateExpressions.forall(filterNotDefined) =>
+ aggregates += left
+ flag = collectAggregate(right, aggregates)
+ // The side of Join is neither Aggregate nor Join.
+ case _ => flag = false
+ }
+ }
+
+ flag
+ }
+
+ // TODO Support aggregate expression with filter clause.
+ private def filterNotDefined(ne: NamedExpression): Boolean = {
+ ne match {
+ case Alias(ae: AggregateExpression, _) => ae.filter.isEmpty
+ case ae: AggregateExpression => ae.filter.isEmpty
+ }
+ }
+
+ // Merge the multiple Aggregates.
+ private def mergePlan(
+ left: LogicalPlan,
+ right: LogicalPlan): Option[(LogicalPlan, Map[Expression, Attribute], Seq[Expression])] = {
+ (left, right) match {
+ case (la: Aggregate, ra: Aggregate) =>
+ val mergedChildPlan = mergePlan(la.child, ra.child)
+ mergedChildPlan.map { case (newChild, outputMap, filters) =>
+ val rightAggregateExprs = ra.aggregateExpressions.map { ne =>
+ ne.transform {
+ case attr: Attribute =>
+ outputMap.getOrElse(attr.canonicalized, attr)
+ }.asInstanceOf[NamedExpression]
+ }
+
+ val mergedAggregateExprs = if (filters.length == 2) {
+ la.aggregateExpressions.map { ne =>
+ ne.transform {
+ case ae @ AggregateExpression(_, _, _, None, _) =>
+ ae.copy(filter = Some(filters.head))
+ }.asInstanceOf[NamedExpression]
+ } ++ rightAggregateExprs.map { ne =>
+ ne.transform {
+ case ae @ AggregateExpression(_, _, _, None, _) =>
+ ae.copy(filter = Some(filters.last))
+ }.asInstanceOf[NamedExpression]
+ }
+ } else {
+ la.aggregateExpressions ++ rightAggregateExprs
+ }
+
+ (Aggregate(Seq.empty, mergedAggregateExprs, newChild), Map.empty, Seq.empty)
+ }
+ case (lp: Project, rp: Project) =>
+ val mergedInfo = mergePlan(lp.child, rp.child)
+ val mergedProjectList = ArrayBuffer[NamedExpression](lp.projectList: _*)
+
+ mergedInfo.map { case (newChild, outputMap, filters) =>
+ val allFilterReferences = filters.flatMap(_.references)
+ val newOutputMap = (rp.projectList ++ allFilterReferences).map { ne =>
+ val mapped = ne.transform {
+ case attr: Attribute =>
+ outputMap.getOrElse(attr.canonicalized, attr)
+ }.asInstanceOf[NamedExpression]
+
+ val withoutAlias = mapped match {
+ case Alias(child, _) => child
+ case e => e
+ }
+
+ val outputAttr = mergedProjectList.find {
+ case Alias(child, _) => child semanticEquals withoutAlias
+ case e => e semanticEquals withoutAlias
+ }.getOrElse {
+ mergedProjectList += mapped
+ mapped
+ }.toAttribute
+ ne.toAttribute.canonicalized -> outputAttr
+ }.toMap
+
+ (Project(mergedProjectList.toSeq, newChild), newOutputMap, filters)
+ }
+ // TODO support only one side contains Filter
+ case (lf: Filter, rf: Filter) =>
+ val mergedInfo = mergePlan(lf.child, rf.child)
+ mergedInfo.map { case (newChild, outputMap, _) =>
Review Comment:
I mean this case branch seems to behave incorrectly when you have adjacent filters in both plans.
E.g. if you have:
```
SELECT avg(a)
FROM (
SELECT * FROM t WHERE c1 = 1
) t
WHERE c2 = 1
```
and
```
SELECT sum(b)
FROM (
SELECT * FROM t WHERE c1 = 2
) t
WHERE c2 = 2
```
and `PushDownPredicates` is disabled then you are in that situation. When you are merging the parent `Filter` nodes you shouldn`t ignore the propagated filters from child `Filter` nodes.
Actually https://github.com/apache/spark/pull/37630 handles that case and many others and that's why I think splitting this PR into 2 parts would be better:
- This rule would be still useful if you used the existing `MergeScalarSubqueries.tryMergePlans()` to merge plans. It couldn't deal with Q28 immediately, but could deal with simpler aggregates without differing filters.
- https://github.com/apache/spark/pull/37630 could be a good base for implementing the filter merging feature. We could simplify that PR and remove the tricky physical plan checking part and introduce a simple boolean flag to enable/disable the filter merging feature like this PR does. In this case both Q9 and Q28 could benefit from the new feature...
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
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