Pajaraja commented on code in PR #49955: URL: https://github.com/apache/spark/pull/49955#discussion_r1981043784
########## sql/core/src/main/scala/org/apache/spark/sql/execution/RecursiveCTEExecution.scala: ########## @@ -0,0 +1,229 @@ +/* + * 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.execution + +import scala.collection.mutable + +import org.apache.spark.SparkException +import org.apache.spark.rdd.{EmptyRDD, RDD} +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.expressions.{Alias, Attribute, Literal} +import org.apache.spark.sql.catalyst.plans.QueryPlan +import org.apache.spark.sql.catalyst.plans.logical.{Filter, LocalLimit, LogicalPlan, Project, Union, UnionLoopRef} +import org.apache.spark.sql.classic.Dataset +import org.apache.spark.sql.execution.metric.SQLMetrics +import org.apache.spark.sql.internal.SQLConf + + +/** + * The physical node for recursion. Currently only UNION ALL case is supported. + * For the details about the execution, look at the comment above doExecute function. + * + * A simple recursive query: + * {{{ + * WITH RECURSIVE t(n) AS ( + * SELECT 1 + * UNION ALL + * SELECT n+1 FROM t WHERE n < 5) + * SELECT * FROM t; + * }}} + * Corresponding logical plan for the recursive query above: + * {{{ + * WithCTE + * :- CTERelationDef 0, false + * : +- SubqueryAlias t + * : +- Project [1#0 AS n#3] + * : +- UnionLoop 0 + * : :- Project [1 AS 1#0] + * : : +- OneRowRelation + * : +- Project [(n#1 + 1) AS (n + 1)#2] + * : +- Filter (n#1 < 5) + * : +- SubqueryAlias t + * : +- Project [1#0 AS n#1] + * : +- UnionLoopRef 0, [1#0], false + * +- Project [n#3] + * +- SubqueryAlias t + * +- CTERelationRef 0, true, [n#3], false, false + * }}} + * + * @param loopId This is id of the CTERelationDef containing the recursive query. Its value is + * first passed down to UnionLoop when creating it, and then to UnionLoopExec in + * SparkStrategies. + * @param anchor The logical plan of the initial element of the loop. + * @param recursion The logical plan that describes the recursion with an [[UnionLoopRef]] node. + * CTERelationRef, which is marked as recursive, gets substituted with + * [[UnionLoopRef]] in ResolveWithCTE. + * Both anchor and recursion are marked with @transient annotation, so that they + * are not serialized. + * @param output The output attributes of this loop. + * @param limit If defined, the total number of rows output by this operator will be bounded by + * limit. + * Its value is pushed down to UnionLoop in Optimizer in case Limit node is present + * in the logical plan and then transferred to UnionLoopExec in SparkStrategies. + * Note here: limit can be applied in the main query calling the recursive CTE, and not + * inside the recursive term of recursive CTE. + */ +case class UnionLoopExec( + loopId: Long, + @transient anchor: LogicalPlan, + @transient recursion: LogicalPlan, + override val output: Seq[Attribute], + localLimit: Option[Int] = None, + globalLimit: Option[Int] = None) extends LeafExecNode { + + override def innerChildren: Seq[QueryPlan[_]] = Seq(anchor, recursion) + + private val numPartitions: Int = conf.defaultNumShufflePartitions + + override lazy val metrics = Map( + "numOutputRows" -> SQLMetrics.createMetric(sparkContext, "number of output rows"), + "numRecursiveLoops" -> SQLMetrics.createMetric(sparkContext, "number of recursive loops")) + + private val simpleRecursion = { + recursion match { + case Project(_, Filter(_, Project(_, UnionLoopRef(_, _, _)))) => + true + case Filter(_, Project(_, UnionLoopRef(_, _, _))) => + true + case Project(_, Filter(_, UnionLoopRef(_, _, _))) => + true + case Project(_, UnionLoopRef(_, _, _)) => + true + case _ => + false + } + } + /** + * This function executes the plan (optionally with appended limit node) and caches the result, + * with the caching mode specified in config. + */ + private def executeAndCacheAndCount( + plan: LogicalPlan, currentLimit: Int) = { + // In case limit is defined, we create a (local) limit node above the plan and execute + // the newly created plan. + val planOrLimitedPlan = if (globalLimit.isDefined || localLimit.isDefined) { + LocalLimit(Literal(currentLimit), plan) + } else { + plan + } + val df = Dataset.ofRows(session, planOrLimitedPlan) + val newDF = { + if (!simpleRecursion) { + df.repartition(numPartitions) + } else { + df + } + } + val count = newDF.count() + (newDF, count) + } + + /** + * In the first iteration, anchor term is executed. + * Then, in each following iteration, the UnionLoopRef node is substituted with the plan from the + * previous iteration, and such plan is executed. + * After every iteration, the dataframe is repartitioned. + * The recursion stops when the generated dataframe is empty, or either the limit or + * the specified maximum depth from the config is reached. + */ + override protected def doExecute(): RDD[InternalRow] = { + val executionId = sparkContext.getLocalProperty(SQLExecution.EXECUTION_ID_KEY) + val numOutputRows = longMetric("numOutputRows") + val numRecursiveLoops = longMetric("numRecursiveLoops") + val levelLimit = conf.getConf(SQLConf.CTE_RECURSION_LEVEL_LIMIT) + + // currentLimit is initialized from the limit argument, and in each step it is decreased by + // the number of rows generated in that step. + // If limit is not passed down, currentLimit is set to be zero and won't be considered in the + // condition of while loop down (limit.isEmpty will be true). + var globalLimitNum = globalLimit.getOrElse(0) + var localLimitNum = localLimit.getOrElse(0) + var currentLimit = Math.max(globalLimitNum, localLimitNum * numPartitions) Review Comment: Is this true even though I apply repartition at the end? Not really sure how to proceed here, because it seems that if I don't apply repartition at the end, we get one partition for each iteration, and then the localLimit can never apply. I don't really understand how the partitions interact with the iterations. Would love to discuss more about this local limit, and see whether we can come to a way to patch it. -- 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
