zhuzhurk commented on code in PR #24771: URL: https://github.com/apache/flink/pull/24771#discussion_r1609784148
########## flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/adaptivebatch/AdaptiveBatchScheduler.java: ########## @@ -224,18 +250,153 @@ private SpeculativeExecutionHandler createSpeculativeExecutionHandler( protected void startSchedulingInternal() { speculativeExecutionHandler.init( getExecutionGraph(), getMainThreadExecutor(), jobManagerJobMetricGroup); + jobRecoveryHandler.initialize( + log, + getExecutionGraph(), + shuffleMaster, + getMainThreadExecutor(), + failoverStrategy, + this::failJob, + this::resetVerticesInRecovering, + this::updateResultPartitionBytesMetrics, + this::initializeJobVertex, + this::updateTopology); + + if (jobRecoveryHandler.needRecover()) { + getMainThreadExecutor() + .schedule( + () -> + jobRecoveryHandler.startRecovering( + this::onRecoveringFinished, this::onRecoveringFailed), + previousWorkerRecoveryTimeout.toMillis(), + TimeUnit.MILLISECONDS); + } else { + tryComputeSourceParallelismThenRunAsync( + (Void value, Throwable throwable) -> { + if (getExecutionGraph().getState() == JobStatus.CREATED) { + initializeVerticesIfPossible(); + super.startSchedulingInternal(); + } + }); + } + } + + @Override + protected void maybeRestartTasks(final FailureHandlingResult failureHandlingResult) { + FailureHandlingResult wrappedResult = failureHandlingResult; + if (failureHandlingResult.canRestart()) { + Set<ExecutionVertexID> originalNeedToRestartVertices = + failureHandlingResult.getVerticesToRestart(); + + Set<JobVertexID> extraNeedToRestartJobVertices = + originalNeedToRestartVertices.stream() + .map(ExecutionVertexID::getJobVertexId) + .filter(requiredRestartJobVertices::contains) + .collect(Collectors.toSet()); + + requiredRestartJobVertices.removeAll(extraNeedToRestartJobVertices); + + Set<ExecutionVertexID> needToRestartVertices = + extraNeedToRestartJobVertices.stream() + .flatMap( + jobVertexId -> { + ExecutionJobVertex jobVertex = + getExecutionJobVertex(jobVertexId); + return Arrays.stream(jobVertex.getTaskVertices()) + .map(ExecutionVertex::getID); + }) + .collect(Collectors.toSet()); + needToRestartVertices.addAll(originalNeedToRestartVertices); + + wrappedResult = + FailureHandlingResult.restartable( + failureHandlingResult.getFailedExecution().orElse(null), + failureHandlingResult.getError(), + failureHandlingResult.getTimestamp(), + failureHandlingResult.getFailureLabels(), + needToRestartVertices, + failureHandlingResult.getRestartDelayMS(), + failureHandlingResult.isGlobalFailure(), + failureHandlingResult.isRootCause()); + } + + super.maybeRestartTasks(wrappedResult); + } + + @VisibleForTesting + boolean isRecovering() { + return jobRecoveryHandler.isRecovering(); + } + + @Override + public boolean updateTaskExecutionState(final TaskExecutionStateTransition taskExecutionState) { + boolean success = super.updateTaskExecutionState(taskExecutionState); + + if (success + && taskExecutionState.getExecutionState() == ExecutionState.FINISHED + && !isRecovering()) { + final ExecutionVertexID executionVertexId = + taskExecutionState.getID().getExecutionVertexId(); + jobRecoveryHandler.notifyExecutionFinished(executionVertexId, taskExecutionState); + } + return success; + } + + @Override + protected void resetForNewExecutions(Collection<ExecutionVertexID> vertices) { + super.resetForNewExecutions(vertices); + if (!isRecovering()) { + jobRecoveryHandler.notifyExecutionVertexReset(vertices); + } + } + + private void initializeJobVertex( + ExecutionJobVertex jobVertex, + int parallelism, + Map<IntermediateDataSetID, JobVertexInputInfo> jobVertexInputInfos, + long createTimestamp) + throws JobException { + if (!jobVertex.isParallelismDecided()) { + changeJobVertexParallelism(jobVertex, parallelism); + } else { + checkState(parallelism == jobVertex.getParallelism()); + } + checkState(canInitialize(jobVertex)); + getExecutionGraph().initializeJobVertex(jobVertex, createTimestamp, jobVertexInputInfos); + if (!isRecovering()) { + jobRecoveryHandler.notifyExecutionJobVertexInitialization( + jobVertex.getJobVertex().getID(), parallelism, jobVertexInputInfos); + } + } + + private void resetVerticesInRecovering(Set<ExecutionVertexID> verticesToReset) + throws Exception { + for (ExecutionVertexID executionVertexID : verticesToReset) { + notifyCoordinatorsAboutTaskFailure( + getExecutionVertex(executionVertexID).getCurrentExecutionAttempt(), null); + } + resetForNewExecutions(verticesToReset); + restoreState(verticesToReset, false); + } + private void onRecoveringFinished(Set<JobVertexID> requiredRestartJobVertices) { Review Comment: Is is possible to avoid depending on the handler to offer this `requiredRestartJobVertices`, but instead continue the scheduling with the current topology status? -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org