|
| 1 | +/* |
| 2 | + * Licensed to the Apache Software Foundation (ASF) under one or more |
| 3 | + * contributor license agreements. See the NOTICE file distributed with |
| 4 | + * this work for additional information regarding copyright ownership. |
| 5 | + * The ASF licenses this file to You under the Apache License, Version 2.0 |
| 6 | + * (the "License"); you may not use this file except in compliance with |
| 7 | + * the License. You may obtain a copy of the License at |
| 8 | + * |
| 9 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | + * |
| 11 | + * Unless required by applicable law or agreed to in writing, software |
| 12 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | + * See the License for the specific language governing permissions and |
| 15 | + * limitations under the License. |
| 16 | + */ |
| 17 | + |
| 18 | +package org.apache.spark.sql.catalyst.optimizer |
| 19 | + |
| 20 | +import scala.collection.mutable.ArrayBuffer |
| 21 | + |
| 22 | +import org.apache.spark.sql.catalyst.expressions.{Alias, And, Attribute, AttributeMap, Expression, NamedExpression, Or} |
| 23 | +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression |
| 24 | +import org.apache.spark.sql.catalyst.plans.{Cross, FullOuter, Inner, JoinType, LeftOuter, RightOuter} |
| 25 | +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, Filter, Join, LeafNode, LogicalPlan, Project, SerializeFromObject} |
| 26 | +import org.apache.spark.sql.catalyst.rules.Rule |
| 27 | +import org.apache.spark.sql.catalyst.trees.TreePattern.{AGGREGATE, ARRAY_CONTAINS, ARRAYS_OVERLAP, AT_LEAST_N_NON_NULLS, BLOOM_FILTER, DYNAMIC_PRUNING_EXPRESSION, DYNAMIC_PRUNING_SUBQUERY, EXISTS_SUBQUERY, HIGH_ORDER_FUNCTION, IN, IN_SUBQUERY, INSET, INVOKE, JOIN, JSON_TO_STRUCT, LIKE_FAMLIY, PYTHON_UDF, REGEXP_EXTRACT_FAMILY, REGEXP_REPLACE, SCALA_UDF, STRING_PREDICATE} |
| 28 | + |
| 29 | +/** |
| 30 | + * This rule eliminates the [[Join]] if all the join side are [[Aggregate]]s by combine these |
| 31 | + * [[Aggregate]]s. This rule also support the nested [[Join]], as long as all the join sides for |
| 32 | + * every [[Join]] are [[Aggregate]]s. |
| 33 | + * |
| 34 | + * Note: this rule doesn't support following cases: |
| 35 | + * 1. The [[Aggregate]]s to be merged if at least one of them does not have a predicate or |
| 36 | + * has low predicate selectivity. |
| 37 | + * 2. The upstream node of these [[Aggregate]]s to be merged exists [[Join]]. |
| 38 | + */ |
| 39 | +object CombineJoinedAggregates extends Rule[LogicalPlan] with MergeScalarSubqueriesHelper { |
| 40 | + |
| 41 | + private def isSupportedJoinType(joinType: JoinType): Boolean = |
| 42 | + Seq(Inner, Cross, LeftOuter, RightOuter, FullOuter).contains(joinType) |
| 43 | + |
| 44 | + private def isCheapPredicate(e: Expression): Boolean = { |
| 45 | + !e.containsAnyPattern(PYTHON_UDF, SCALA_UDF, INVOKE, JSON_TO_STRUCT, LIKE_FAMLIY, |
| 46 | + REGEXP_EXTRACT_FAMILY, REGEXP_REPLACE, DYNAMIC_PRUNING_SUBQUERY, DYNAMIC_PRUNING_EXPRESSION, |
| 47 | + HIGH_ORDER_FUNCTION, IN_SUBQUERY, IN, INSET, EXISTS_SUBQUERY, STRING_PREDICATE, |
| 48 | + AT_LEAST_N_NON_NULLS, BLOOM_FILTER, ARRAY_CONTAINS, ARRAYS_OVERLAP) && |
| 49 | + Option(e.apply(conf.maxTreeNodeNumOfPredicate)).isEmpty |
| 50 | + } |
| 51 | + |
| 52 | + /** |
| 53 | + * Try to merge two `Aggregate`s by traverse down recursively. |
| 54 | + * |
| 55 | + * @return The optional tuple as follows: |
| 56 | + * 1. the merged plan |
| 57 | + * 2. the attribute mapping from the old to the merged version |
| 58 | + * 3. optional filters of both plans that need to be propagated and merged in an |
| 59 | + * ancestor `Aggregate` node if possible. |
| 60 | + */ |
| 61 | + private def mergePlan( |
| 62 | + left: LogicalPlan, |
| 63 | + right: LogicalPlan): Option[(LogicalPlan, AttributeMap[Attribute], Seq[Expression])] = { |
| 64 | + (left, right) match { |
| 65 | + case (la: Aggregate, ra: Aggregate) => |
| 66 | + mergePlan(la.child, ra.child).map { case (newChild, outputMap, filters) => |
| 67 | + val rightAggregateExprs = ra.aggregateExpressions.map(mapAttributes(_, outputMap)) |
| 68 | + |
| 69 | + val mergedAggregateExprs = if (filters.length == 2) { |
| 70 | + Seq( |
| 71 | + (la.aggregateExpressions, filters.head), |
| 72 | + (rightAggregateExprs, filters.last) |
| 73 | + ).flatMap { case (aggregateExpressions, propagatedFilter) => |
| 74 | + aggregateExpressions.map { ne => |
| 75 | + ne.transform { |
| 76 | + case ae @ AggregateExpression(_, _, _, filterOpt, _) => |
| 77 | + val newFilter = filterOpt.map { filter => |
| 78 | + And(propagatedFilter, filter) |
| 79 | + }.orElse(Some(propagatedFilter)) |
| 80 | + ae.copy(filter = newFilter) |
| 81 | + }.asInstanceOf[NamedExpression] |
| 82 | + } |
| 83 | + } |
| 84 | + } else { |
| 85 | + la.aggregateExpressions ++ rightAggregateExprs |
| 86 | + } |
| 87 | + |
| 88 | + (Aggregate(Seq.empty, mergedAggregateExprs, newChild), AttributeMap.empty, Seq.empty) |
| 89 | + } |
| 90 | + case (lp: Project, rp: Project) => |
| 91 | + val mergedProjectList = ArrayBuffer[NamedExpression](lp.projectList: _*) |
| 92 | + |
| 93 | + mergePlan(lp.child, rp.child).map { case (newChild, outputMap, filters) => |
| 94 | + val allFilterReferences = filters.flatMap(_.references) |
| 95 | + val newOutputMap = AttributeMap((rp.projectList ++ allFilterReferences).map { ne => |
| 96 | + val mapped = mapAttributes(ne, outputMap) |
| 97 | + |
| 98 | + val withoutAlias = mapped match { |
| 99 | + case Alias(child, _) => child |
| 100 | + case e => e |
| 101 | + } |
| 102 | + |
| 103 | + val outputAttr = mergedProjectList.find { |
| 104 | + case Alias(child, _) => child semanticEquals withoutAlias |
| 105 | + case e => e semanticEquals withoutAlias |
| 106 | + }.getOrElse { |
| 107 | + mergedProjectList += mapped |
| 108 | + mapped |
| 109 | + }.toAttribute |
| 110 | + ne.toAttribute -> outputAttr |
| 111 | + }) |
| 112 | + |
| 113 | + (Project(mergedProjectList.toSeq, newChild), newOutputMap, filters) |
| 114 | + } |
| 115 | + case (lf: Filter, rf: Filter) |
| 116 | + if isCheapPredicate(lf.condition) && isCheapPredicate(rf.condition) => |
| 117 | + mergePlan(lf.child, rf.child).map { |
| 118 | + case (newChild, outputMap, filters) => |
| 119 | + val mappedRightCondition = mapAttributes(rf.condition, outputMap) |
| 120 | + val (newLeftCondition, newRightCondition) = if (filters.length == 2) { |
| 121 | + (And(lf.condition, filters.head), And(mappedRightCondition, filters.last)) |
| 122 | + } else { |
| 123 | + (lf.condition, mappedRightCondition) |
| 124 | + } |
| 125 | + val newCondition = Or(newLeftCondition, newRightCondition) |
| 126 | + |
| 127 | + (Filter(newCondition, newChild), outputMap, Seq(newLeftCondition, newRightCondition)) |
| 128 | + } |
| 129 | + case (ll: LeafNode, rl: LeafNode) => |
| 130 | + checkIdenticalPlans(rl, ll).map { outputMap => |
| 131 | + (ll, outputMap, Seq.empty) |
| 132 | + } |
| 133 | + case (ls: SerializeFromObject, rs: SerializeFromObject) => |
| 134 | + checkIdenticalPlans(rs, ls).map { outputMap => |
| 135 | + (ls, outputMap, Seq.empty) |
| 136 | + } |
| 137 | + case _ => None |
| 138 | + } |
| 139 | + } |
| 140 | + |
| 141 | + def apply(plan: LogicalPlan): LogicalPlan = { |
| 142 | + if (!conf.combineJoinedAggregatesEnabled) return plan |
| 143 | + |
| 144 | + plan.transformUpWithPruning(_.containsAnyPattern(JOIN, AGGREGATE), ruleId) { |
| 145 | + case j @ Join(left: Aggregate, right: Aggregate, joinType, None, _) |
| 146 | + if isSupportedJoinType(joinType) && |
| 147 | + left.groupingExpressions.isEmpty && right.groupingExpressions.isEmpty => |
| 148 | + val mergedAggregate = mergePlan(left, right) |
| 149 | + mergedAggregate.map(_._1).getOrElse(j) |
| 150 | + } |
| 151 | + } |
| 152 | +} |
0 commit comments