ulysses-you commented on code in PR #7570: URL: https://github.com/apache/kyuubi/pull/7570#discussion_r3595076754
########## extensions/spark/kyuubi-extension-spark-4-2/src/main/scala/org/apache/spark/sql/FinalStageResourceManager.scala: ########## @@ -0,0 +1,289 @@ +/* + * 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 + +import scala.annotation.tailrec +import scala.collection.mutable +import scala.collection.mutable.ArrayBuffer + +import org.apache.spark.{ExecutorAllocationClient, MapOutputTrackerMaster, SparkContext, SparkEnv} +import org.apache.spark.internal.Logging +import org.apache.spark.resource.ResourceProfile +import org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.adaptive._ +import org.apache.spark.sql.execution.columnar.InMemoryTableScanExec +import org.apache.spark.sql.execution.command.DataWritingCommandExec +import org.apache.spark.sql.execution.datasources.WriteFilesExec +import org.apache.spark.sql.execution.datasources.v2.V2TableWriteExec +import org.apache.spark.sql.execution.exchange.{ENSURE_REQUIREMENTS, ShuffleExchangeExec} + +import org.apache.kyuubi.sql.{KyuubiSQLConf, WriteUtils} + +/** + * This rule assumes the final write stage has less cores requirement than previous, otherwise + * this rule would take no effect. + * + * It provide a feature: + * 1. Kill redundant executors before running final write stage + */ +case class FinalStageResourceManager(session: SparkSession) + extends Rule[SparkPlan] with FinalRebalanceStageHelper { + override def apply(plan: SparkPlan): SparkPlan = { + if (!conf.getConf(KyuubiSQLConf.FINAL_WRITE_STAGE_EAGERLY_KILL_EXECUTORS_ENABLED)) { + return plan + } + + if (!WriteUtils.isWrite(session, plan)) { + return plan + } + + val sc = session.sparkContext + val dra = sc.getConf.getBoolean("spark.dynamicAllocation.enabled", false) + val coresPerExecutor = sc.getConf.getInt("spark.executor.cores", 1) + val minExecutors = sc.getConf.getInt("spark.dynamicAllocation.minExecutors", 0) + val maxExecutors = sc.getConf.getInt("spark.dynamicAllocation.maxExecutors", Int.MaxValue) + val factor = conf.getConf(KyuubiSQLConf.FINAL_WRITE_STAGE_PARTITION_FACTOR) + val hasImprovementRoom = maxExecutors - 1 > minExecutors * factor + // Fast fail if: + // 1. DRA off + // 2. only work with yarn and k8s + // 3. maxExecutors is not bigger than minExecutors * factor + if (!dra || !sc.schedulerBackend.isInstanceOf[CoarseGrainedSchedulerBackend] || + !hasImprovementRoom) { + return plan + } + + val stageOpt = findFinalRebalanceStage(plan) + if (stageOpt.isEmpty) { + return plan + } + + // It's not safe to kill executors if this plan contains table cache. + // If the executor loses then the rdd would re-compute those partition. + if (hasTableCache(plan) && + conf.getConf(KyuubiSQLConf.FINAL_WRITE_STAGE_SKIP_KILLING_EXECUTORS_FOR_TABLE_CACHE)) { + return plan + } + + // TODO: move this to query stage optimizer when updating Spark to 3.5.x + // Since we are in `prepareQueryStage`, the AQE shuffle read has not been applied. + // So we need to apply it by self. + val shuffleRead = queryStageOptimizerRules.foldLeft(stageOpt.get.asInstanceOf[SparkPlan]) { + case (latest, rule) => rule.apply(latest) + } + val (targetCores, stage) = shuffleRead match { + case AQEShuffleReadExec(stage: ShuffleQueryStageExec, partitionSpecs) => + (partitionSpecs.length, stage) + case stage: ShuffleQueryStageExec => + // we can still kill executors if no AQE shuffle read, e.g., `.repartition(2)` + (stage.shuffle.numPartitions, stage) + case _ => + // it should never happen in current Spark, but to be safe do nothing if happens + logWarning("BUG, Please report to Apache Kyuubi community") + return plan + } + // The condition whether inject custom resource profile: + // - target executors < active executors + // - active executors - target executors > min executors + val numActiveExecutors = sc.getExecutorIds().length + val targetExecutors = (math.ceil(targetCores.toFloat / coresPerExecutor) * factor).toInt + .max(1) + val hasBenefits = targetExecutors < numActiveExecutors && + (numActiveExecutors - targetExecutors) > minExecutors + logInfo(s"The snapshot of current executors view, " + + s"active executors: $numActiveExecutors, min executor: $minExecutors, " + + s"target executors: $targetExecutors, has benefits: $hasBenefits") + if (hasBenefits) { + val shuffleId = stage.plan.asInstanceOf[ShuffleExchangeExec].shuffleDependency.shuffleId + val numReduce = stage.plan.asInstanceOf[ShuffleExchangeExec].numPartitions + // Now, there is only a final rebalance stage waiting to execute and all tasks of previous + // stage are finished. Kill redundant existed executors eagerly so the tasks of final + // stage can be centralized scheduled. + killExecutors(sc, targetExecutors, shuffleId, numReduce) + } + + plan + } + + /** + * The priority of kill executors follow: + * 1. kill executor who is younger than other (The older the JIT works better) + * 2. kill executor who produces less shuffle data first + */ + private def findExecutorToKill( + sc: SparkContext, + targetExecutors: Int, + shuffleId: Int, + numReduce: Int): Seq[String] = { + val tracker = SparkEnv.get.mapOutputTracker.asInstanceOf[MapOutputTrackerMaster] + val shuffleStatusOpt = tracker.shuffleStatuses.get(shuffleId) + if (shuffleStatusOpt.isEmpty) { + return Seq.empty + } + val shuffleStatus = shuffleStatusOpt.get + val executorToBlockSize = new mutable.HashMap[String, Long] + shuffleStatus.withMapStatuses { mapStatus => + mapStatus.foreach { status => + var i = 0 + var sum = 0L + while (i < numReduce) { + sum += status.getSizeForBlock(i) + i += 1 + } + executorToBlockSize.getOrElseUpdate(status.location.executorId, sum) + } + } + + val backend = sc.schedulerBackend.asInstanceOf[CoarseGrainedSchedulerBackend] + val executorsWithRegistrationTs = backend.getExecutorsWithRegistrationTs() + val existedExecutors = executorsWithRegistrationTs.keys.toSet + val expectedNumExecutorToKill = existedExecutors.size - targetExecutors + if (expectedNumExecutorToKill < 1) { + return Seq.empty + } + + val executorIdsToKill = new ArrayBuffer[String]() + // We first kill executor who does not hold shuffle block. It would happen because + // the last stage is running fast and finished in a short time. The existed executors are + // from previous stages that have not been killed by DRA, so we can not find it by tracking + // shuffle status. + // We should evict executors by their alive time first and retain all of executors which + // have better locality for shuffle block. + executorsWithRegistrationTs.toSeq.sortBy(_._2).foreach { case (id, _) => Review Comment: it is a valid suggestion -- 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]
