GJL commented on a change in pull request #8309: [FLINK-12229] [runtime] Implement LazyFromSourcesScheduling Strategy URL: https://github.com/apache/flink/pull/8309#discussion_r282783770
########## File path: flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/strategy/LazyFromSourcesSchedulingStrategy.java ########## @@ -0,0 +1,296 @@ +/* + * 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.flink.runtime.scheduler.strategy; + +import org.apache.flink.api.common.InputDependencyConstraint; +import org.apache.flink.runtime.execution.ExecutionState; +import org.apache.flink.runtime.io.network.partition.ResultPartitionID; +import org.apache.flink.runtime.jobgraph.IntermediateDataSet; +import org.apache.flink.runtime.jobgraph.IntermediateDataSetID; +import org.apache.flink.runtime.jobgraph.JobGraph; +import org.apache.flink.runtime.scheduler.DeploymentOption; +import org.apache.flink.runtime.scheduler.ExecutionVertexDeploymentOption; +import org.apache.flink.runtime.scheduler.SchedulerOperations; + +import java.util.ArrayList; +import java.util.Collection; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.Set; + +import static org.apache.flink.runtime.scheduler.strategy.SchedulingResultPartition.ResultPartitionState.DONE; +import static org.apache.flink.runtime.scheduler.strategy.SchedulingResultPartition.ResultPartitionState.PRODUCING; +import static org.apache.flink.util.Preconditions.checkNotNull; + +/** + * {@link SchedulingStrategy} instance for batch job which schedule vertices when input data are ready. + */ +public class LazyFromSourcesSchedulingStrategy implements SchedulingStrategy { + + private final SchedulerOperations schedulerOperations; + + private final SchedulingTopology schedulingTopology; + + private final Map<ExecutionVertexID, DeploymentOption> deploymentOptions; + + private final Map<IntermediateDataSetID, SchedulingIntermediateDataSet> intermediateDataSets; + + public LazyFromSourcesSchedulingStrategy( + SchedulerOperations schedulerOperations, + SchedulingTopology schedulingTopology) { + this.schedulerOperations = checkNotNull(schedulerOperations); + this.schedulingTopology = checkNotNull(schedulingTopology); + this.intermediateDataSets = new HashMap<>(); + this.deploymentOptions = new HashMap<>(); + } + + @Override + public void startScheduling() { + List<ExecutionVertexDeploymentOption> executionVertexDeploymentOptions = new ArrayList<>(); + DeploymentOption updateOption = new DeploymentOption(true); + DeploymentOption nonUpdateOption = new DeploymentOption(false); + + for (SchedulingExecutionVertex schedulingVertex : schedulingTopology.getVertices()) { + DeploymentOption option = nonUpdateOption; + for (SchedulingResultPartition srp : schedulingVertex.getProducedResultPartitions()) { + SchedulingIntermediateDataSet sid = intermediateDataSets.computeIfAbsent(srp.getResultId(), + (key) -> new SchedulingIntermediateDataSet()); + sid.addSchedulingResultPartition(srp); + if (srp.getPartitionType().isPipelined()) { + option = updateOption; + } + } + deploymentOptions.put(schedulingVertex.getId(), option); + + if (schedulingVertex.getConsumedResultPartitions().isEmpty()) { + // schedule vertices without consumed result partition + executionVertexDeploymentOptions.add( + new ExecutionVertexDeploymentOption(schedulingVertex.getId(), option)); + } + } + + schedulerOperations.allocateSlotsAndDeploy(executionVertexDeploymentOptions); + } + + @Override + public void restartTasks(Set<ExecutionVertexID> verticesToRestart) { + // increase counter of the dataset first + for (ExecutionVertexID executionVertexId : verticesToRestart) { + final SchedulingExecutionVertex schedulingVertex = schedulingTopology.getVertex(executionVertexId) + .orElseThrow(() -> new IllegalStateException("can not find scheduling vertex for " + executionVertexId)); + + for (SchedulingResultPartition srp : schedulingVertex.getProducedResultPartitions()) { + if (srp.getPartitionType().isBlocking() && DONE.equals(srp.getState())) { + SchedulingIntermediateDataSet intermediateDataSet = intermediateDataSets.get(srp.getResultId()); + if (intermediateDataSet == null) { + throw new IllegalStateException("can not find scheduling intermediate dataset for " + + srp.getResultId()); + } + intermediateDataSet.incrementNumberOfRunningProducers(); + } + } + } + + List<ExecutionVertexDeploymentOption> executionVertexDeploymentOptions = new ArrayList<>(verticesToRestart.size()); + for (ExecutionVertexID executionVertexId : verticesToRestart) { + final SchedulingExecutionVertex schedulingVertex = schedulingTopology.getVertex(executionVertexId) + .orElseThrow(() -> new IllegalStateException("can not find scheduling vertex for " + executionVertexId)); + final InputDependencyConstraint constraint = schedulingTopology + .getInputDependencyConstraint(schedulingVertex.getId().getJobVertexId()) + .orElseThrow(() -> new IllegalStateException("can not find input dependency constraint for " + + executionVertexId)); + + Collection<SchedulingResultPartition> resultPartitions = schedulingVertex.getConsumedResultPartitions(); + boolean allConsumable = resultPartitions.isEmpty() || + (resultPartitions.iterator().next().getPartitionType().isBlocking() && InputDependencyConstraint.ALL.equals(constraint)); + boolean shouldSchedule; + if (allConsumable) { + shouldSchedule = resultPartitions.stream().allMatch(this::partitionConsumable); + } else { + shouldSchedule = resultPartitions.stream().anyMatch(this::partitionConsumable); + } + + if (shouldSchedule) { + executionVertexDeploymentOptions.add( + new ExecutionVertexDeploymentOption(executionVertexId, deploymentOptions.get(executionVertexId))); + } + } + schedulerOperations.allocateSlotsAndDeploy(executionVertexDeploymentOptions); + } + + @Override + public void onExecutionStateChange(ExecutionVertexID executionVertexId, ExecutionState executionState) { + final SchedulingExecutionVertex schedulingVertex = schedulingTopology.getVertex(executionVertexId) + .orElseThrow(() -> new IllegalStateException("can not find scheduling vertex for " + executionVertexId)); + + if (ExecutionState.FINISHED.equals(executionState)) { + List<SchedulingResultPartition> finishedPartitions = null; + for (SchedulingResultPartition srp : schedulingVertex.getProducedResultPartitions()) { + if (srp.getPartitionType().isBlocking()) { + SchedulingIntermediateDataSet intermediateDataSet = intermediateDataSets.get(srp.getResultId()); + if (intermediateDataSet == null) { + throw new IllegalStateException("can not find scheduling intermediate dataset for " + + srp.getResultId()); + } + if (intermediateDataSet.decrementNumberOfRunningProducers()) { + if (finishedPartitions == null) { + finishedPartitions = new ArrayList<>(); + } + finishedPartitions.addAll(intermediateDataSet.getSchedulingResultPartition()); + } + } + } + + if (finishedPartitions != null) { + List<SchedulingExecutionVertex> toScheduleConsumers = new ArrayList<>(); + for (SchedulingResultPartition finishedPartition : finishedPartitions) { + toScheduleConsumers.addAll(finishedPartition.getConsumers()); + } + scheduleBlockingConsumers(toScheduleConsumers); + } + } + } + + @Override + public void onPartitionConsumable(ExecutionVertexID executionVertexId, ResultPartitionID resultPartitionId) { + final SchedulingExecutionVertex producerVertex = schedulingTopology.getVertex(executionVertexId) + .orElseThrow(() -> new IllegalStateException("can not find scheduling vertex for " + executionVertexId)); + final SchedulingResultPartition resultPartition = schedulingTopology + .getResultPartition(resultPartitionId.getPartitionId()) + .orElseThrow(() -> new IllegalStateException("can not find scheduling result partition for " + + resultPartitionId)); + + if (!producerVertex.getProducedResultPartitions().contains(resultPartition)) { + throw new IllegalStateException("partition " + resultPartitionId + + " is not the produced partition of " + executionVertexId); + } + + List<ExecutionVertexDeploymentOption> executionVertexDeploymentOptions = new ArrayList<>(); + for (SchedulingExecutionVertex schedulingVertex : resultPartition.getConsumers()) { + if (ExecutionState.CREATED.equals(schedulingVertex.getState()) && + resultPartition.getPartitionType().isPipelined() && Review comment: What is the importance of this condition? I think the logic would be still correct without it. ``` resultPartition.getPartitionType().isPipelined() ``` ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
