KevinyhZou commented on code in PR #8558: URL: https://github.com/apache/incubator-gluten/pull/8558#discussion_r1984502581
########## backends-clickhouse/src/main/scala/org/apache/gluten/extension/CollapseNestedExpressions.scala: ########## @@ -0,0 +1,179 @@ +/* + * 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.gluten.extension + +import org.apache.gluten.execution.{FilterExecTransformer, ProjectExecTransformer} +import org.apache.gluten.expression.{CollapsedExpressionMappings, ExpressionMappings} + +import org.apache.spark.sql.SparkSession +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.SparkPlan +import org.apache.spark.sql.types.DataType + +/** + * Collapse nested expressions for optimization, to reduce expression calls. Now support `and`, + * `or`. e.g. select ... and(and(a=1, b=2), c=3) => select ... and(a=1, b=2, c=3). + */ +case class CollapseNestedExpressions(spark: SparkSession) extends Rule[SparkPlan] { + + override def apply(plan: SparkPlan): SparkPlan = { + if (canBeOptimized(plan)) { + visitPlan(plan) + } else { + plan + } + } + + private def canBeOptimized(plan: SparkPlan): Boolean = plan match { + case p: ProjectExecTransformer => + var res = p.projectList.exists(c => c.isInstanceOf[And] || c.isInstanceOf[Or]) + if (res) { + return false + } + res = p.projectList.exists(c => canBeOptimized(c)) + if (!res) { + res = p.children.exists(c => canBeOptimized(c)) + } + res + case f: FilterExecTransformer => + var res = canBeOptimized(f.condition) + if (!res) { + res = canBeOptimized(f.child) + } + res + case _ => plan.children.exists(c => canBeOptimized(c)) + } + + private def canBeOptimized(expr: Expression): Boolean = { + var exprCall = expr + expr match { + case a: Alias => exprCall = a.child + case _ => + } + val exprName = getExpressionName(exprCall) + exprName match { + case None => + exprCall match { + case _: LeafExpression => false + case _ => exprCall.children.exists(c => canBeOptimized(c)) + } + case Some(f) => + CollapsedExpressionMappings.supported(f) + } + } + + private def getExpressionName(expr: Expression): Option[String] = expr match { + case _: And => ExpressionMappings.expressionsMap.get(classOf[And]) + case _: Or => ExpressionMappings.expressionsMap.get(classOf[Or]) + case _ => Option.empty[String] + } + + private def visitPlan(plan: SparkPlan): SparkPlan = plan match { + case p: ProjectExecTransformer => + var newProjectList = Seq.empty[NamedExpression] + p.projectList.foreach { + case a: Alias => + val newAlias = Alias(optimize(a.child), a.name)(a.exprId) + newProjectList :+= newAlias + case p => + newProjectList :+= p + } + val newChild = visitPlan(p.child) + ProjectExecTransformer(newProjectList, newChild) + case f: FilterExecTransformer => + val newCondition = optimize(f.condition) + val newChild = visitPlan(f.child) + FilterExecTransformer(newCondition, newChild) + case _ => + val newChildren = plan.children.map(p => visitPlan(p)) + plan.withNewChildren(newChildren) + } + + private def optimize(expr: Expression): Expression = { + var resultExpr = expr + var name = getExpressionName(expr) + var children = Seq.empty[Expression] + var nestedFunctions = 0 + var dataType = null.asInstanceOf[DataType] + + def f(e: Expression, parent: Option[Expression] = Option.empty[Expression]): Unit = { + parent match { + case None => + name = getExpressionName(e) + dataType = e.dataType + case _ => + } + e match { + case a: And if canBeOptimized(a) => + parent match { + case Some(_: And) | None => + f(a.left, Option.apply(a)) + f(a.right, Option.apply(a)) + nestedFunctions += 1 + case _ => + children +:= optimize(a) + } + case o: Or if canBeOptimized(o) => + parent match { + case Some(_: Or) | None => + f(o.left, parent = Option.apply(o)) + f(o.right, parent = Option.apply(o)) + nestedFunctions += 1 + case _ => + children +:= optimize(o) + } + case _ => + if (parent.nonEmpty) { + children +:= optimize(e) + } else { + nestedFunctions = 0 + children = Seq.empty[Expression] + val exprNewChildren = e.children.map(p => optimize(p)) + resultExpr = e.withNewChildren(exprNewChildren) + } + } + } + f(expr) + if ((nestedFunctions > 1 && name.isDefined) || collapsedExpressionExists(children)) { + + /** + * To offload the collapsed nested functions to clickhouse backend, we should convert it to + * ScalaUDF at first, and then pass the ScalaUDF to clickhouse backend. e.g. And(And(a=1, + * b=2),c=3) can be optimized to And(a=1, b=2, c=3)，but And(a=1, b=2, c=3) can not be + * supported by spark `And` function, so we need to convert it to ScalaUDF, with name is + * `And`, and have 3 arguments, when pass the `ScalaUDF(#and(a=1,b=2,c=3))` to clickhouse Review Comment: I have use a `CHCollapsedExperession` to replace the `ScalaUDF` to pass the optimized plan to clickhouse backend. please help to review the code , thanks . @PHILO-HE -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
