peter-toth commented on code in PR #42223: URL: https://github.com/apache/spark/pull/42223#discussion_r1283307075
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/CombineJoinedAggregates.scala: ########## @@ -0,0 +1,132 @@ +/* + * 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, And, Attribute, AttributeMap, 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: this rule doesn't following cases: + * 1. One of the to be merged two [[Aggregate]]s with child [[Filter]] and the other one is not. Review Comment: > I think a difference here is this PR only merges Aggregates with this pattern > > ``` > Aggregate > Filter > TableScan > ``` > > That said, `that data "flows through" the whole plan and gets filtered up in the aggregate node only` is not an issue as the table scan output goes directly to Aggregate and gets filtered. That's true, merging logic in this PR is simpler (e.g. can't handle joins). But please note that currently this PR accepts different patterns as well. Infinite number of `Projects` and `Filter` nodes are accepted between the `Aggregate` and `Scan` nodes as long as both queries contains them. Although that's a bit atificial example as optimizer rules gets rid of them in most cases... (@beliefer, this is actually connected to https://github.com/apache/spark/pull/42223#discussion_r1279051263 that still doesn't seem correct.) But I'm Ok with accepting only this simple pattern like aggregates for merging. > I think it's a bit hard to estimate the data size of "overlapped scans". The only safe case is when one filter is always `true`. For `p1 = 1` and `p2 = 1`, if we don't know the correlation of column `p1` and `p2`, we don't know the overlapped scan size. I can't agree more. > Another direction is we stop merging aggregates if the filter is costly enough. We can define that a filter expression is cheap if it only contains simple comparison checks and the expression tree size is smaller than a threshold. Hm, or we could also avoid double evaluation by aliasing conditions like: ``` Aggregate sum(a) FILTER (WHERE condition1), sum(b) FILTER (WHERE condition2) Filter condition1 OR condition2 Project <condition 1> as condition1, <condition 2> as condition2 TableScan ``` Actually I already do a very similar thing when I propagate up a filter condition through a `Project`: https://github.com/apache/spark/pull/37630/files#diff-3d3aa853c51c01216a7f7307219544a48e8c14fabe1817850e58739208bc406aR575-R577 -- 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-unsubscr...@spark.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org
