Search in sources :

Example 6 with StandaloneCheckpointIDCounter

use of org.apache.flink.runtime.checkpoint.StandaloneCheckpointIDCounter in project flink by apache.

the class AdaptiveSchedulerTest method runExceptionHistoryTests.

private Iterable<RootExceptionHistoryEntry> runExceptionHistoryTests(BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic, Consumer<AdaptiveSchedulerBuilder> setupScheduler, Consumer<JobGraph> setupJobGraph) throws Exception {
    final int numAvailableSlots = 4;
    final JobGraph jobGraph = createJobGraph();
    setupJobGraph.accept(jobGraph);
    RunFailedJobListener listener = new RunFailedJobListener();
    List<ExecutionAttemptID> cancelledTasks = new ArrayList<>();
    final CompletedCheckpointStore completedCheckpointStore = new StandaloneCompletedCheckpointStore(1);
    final CheckpointIDCounter checkpointIDCounter = new StandaloneCheckpointIDCounter();
    final CheckpointsCleaner checkpointCleaner = new CheckpointsCleaner();
    TestingCheckpointRecoveryFactory checkpointRecoveryFactory = new TestingCheckpointRecoveryFactory(completedCheckpointStore, checkpointIDCounter);
    final DefaultDeclarativeSlotPool declarativeSlotPool = createDeclarativeSlotPool(jobGraph.getJobID());
    final Configuration configuration = new Configuration();
    configuration.set(JobManagerOptions.RESOURCE_WAIT_TIMEOUT, Duration.ofMillis(1L));
    AdaptiveSchedulerBuilder builder = new AdaptiveSchedulerBuilder(jobGraph, singleThreadMainThreadExecutor).setJobMasterConfiguration(configuration).setDeclarativeSlotPool(declarativeSlotPool).setCheckpointRecoveryFactory(checkpointRecoveryFactory).setCheckpointCleaner(checkpointCleaner).setJobStatusListener(listener);
    setupScheduler.accept(builder);
    final AdaptiveScheduler scheduler = builder.build();
    final SubmissionBufferingTaskManagerGateway taskManagerGateway = new SubmissionBufferingTaskManagerGateway(numAvailableSlots);
    taskManagerGateway.setCancelConsumer(cancelledTasks::add);
    singleThreadMainThreadExecutor.execute(() -> {
        scheduler.startScheduling();
        offerSlots(declarativeSlotPool, createSlotOffersForResourceRequirements(ResourceCounter.withResource(ResourceProfile.UNKNOWN, numAvailableSlots)), taskManagerGateway);
    });
    listener.waitForRunning();
    CompletableFuture<Iterable<ArchivedExecutionVertex>> vertexFuture = new CompletableFuture<>();
    singleThreadMainThreadExecutor.execute(() -> vertexFuture.complete(scheduler.requestJob().getArchivedExecutionGraph().getAllExecutionVertices()));
    final Iterable<ArchivedExecutionVertex> executionVertices = vertexFuture.get();
    final List<ExecutionAttemptID> attemptIds = IterableUtils.toStream(executionVertices).map(ArchivedExecutionVertex::getCurrentExecutionAttempt).map(ArchivedExecution::getAttemptId).collect(Collectors.toList());
    CompletableFuture<Void> runTestLogicFuture = CompletableFuture.runAsync(() -> testLogic.accept(scheduler, attemptIds), singleThreadMainThreadExecutor);
    runTestLogicFuture.get();
    Consumer<ExecutionAttemptID> canceller = attemptId -> scheduler.updateTaskExecutionState(new TaskExecutionStateTransition(new TaskExecutionState(attemptId, ExecutionState.CANCELED, null)));
    CompletableFuture<Void> cancelFuture = CompletableFuture.runAsync(() -> cancelledTasks.forEach(canceller), singleThreadMainThreadExecutor);
    cancelFuture.get();
    listener.waitForTerminal();
    return scheduler.requestJob().getExceptionHistory();
}
Also used : Arrays(java.util.Arrays) ResourceRequirement(org.apache.flink.runtime.slots.ResourceRequirement) TaskNotRunningException(org.apache.flink.runtime.operators.coordination.TaskNotRunningException) ArchivedExecution(org.apache.flink.runtime.executiongraph.ArchivedExecution) TestingSlotAllocator(org.apache.flink.runtime.scheduler.adaptive.allocator.TestingSlotAllocator) CheckpointException(org.apache.flink.runtime.checkpoint.CheckpointException) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID) TestingFatalErrorHandler(org.apache.flink.runtime.util.TestingFatalErrorHandler) MetricRegistry(org.apache.flink.runtime.metrics.MetricRegistry) Duration(java.time.Duration) ClassRule(org.junit.ClassRule) TestingCheckpointRecoveryFactory(org.apache.flink.runtime.checkpoint.TestingCheckpointRecoveryFactory) ManuallyTriggeredComponentMainThreadExecutor(org.apache.flink.runtime.concurrent.ManuallyTriggeredComponentMainThreadExecutor) BlockingQueue(java.util.concurrent.BlockingQueue) MetricOptions(org.apache.flink.configuration.MetricOptions) JobManagerOptions(org.apache.flink.configuration.JobManagerOptions) Executors(java.util.concurrent.Executors) MetricNames(org.apache.flink.runtime.metrics.MetricNames) ArrayBlockingQueue(java.util.concurrent.ArrayBlockingQueue) VertexParallelismStore(org.apache.flink.runtime.scheduler.VertexParallelismStore) Time(org.apache.flink.api.common.time.Time) RootExceptionHistoryEntry(org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry) FlinkException(org.apache.flink.util.FlinkException) ComponentMainThreadExecutor(org.apache.flink.runtime.concurrent.ComponentMainThreadExecutor) UpTimeGauge(org.apache.flink.runtime.executiongraph.metrics.UpTimeGauge) ExceptionHistoryEntry(org.apache.flink.runtime.scheduler.exceptionhistory.ExceptionHistoryEntry) LocalTaskManagerLocation(org.apache.flink.runtime.taskmanager.LocalTaskManagerLocation) ResourceCounter(org.apache.flink.runtime.util.ResourceCounter) JobStatus(org.apache.flink.api.common.JobStatus) DefaultAllocatedSlotPool(org.apache.flink.runtime.jobmaster.slotpool.DefaultAllocatedSlotPool) ArrayList(java.util.ArrayList) SchedulerNG(org.apache.flink.runtime.scheduler.SchedulerNG) DownTimeGauge(org.apache.flink.runtime.executiongraph.metrics.DownTimeGauge) PartitionProducerDisposedException(org.apache.flink.runtime.jobmanager.PartitionProducerDisposedException) Gauge(org.apache.flink.metrics.Gauge) ScheduledExecutorService(java.util.concurrent.ScheduledExecutorService) BiConsumer(java.util.function.BiConsumer) FixedDelayRestartBackoffTimeStrategy(org.apache.flink.runtime.executiongraph.failover.flip1.FixedDelayRestartBackoffTimeStrategy) Nullable(javax.annotation.Nullable) ArchivedExecutionVertex(org.apache.flink.runtime.executiongraph.ArchivedExecutionVertex) TestExecutorResource(org.apache.flink.testutils.executor.TestExecutorResource) ExecutionState(org.apache.flink.runtime.execution.ExecutionState) CheckpointsCleaner(org.apache.flink.runtime.checkpoint.CheckpointsCleaner) TestOperatorEvent(org.apache.flink.runtime.operators.coordination.TestOperatorEvent) Test(org.junit.Test) IOException(java.io.IOException) IterableUtils(org.apache.flink.util.IterableUtils) SimpleAckingTaskManagerGateway(org.apache.flink.runtime.executiongraph.utils.SimpleAckingTaskManagerGateway) ExecutionException(java.util.concurrent.ExecutionException) TaskExecutionStateTransition(org.apache.flink.runtime.executiongraph.TaskExecutionStateTransition) JobID(org.apache.flink.api.common.JobID) NoRestartBackoffTimeStrategy(org.apache.flink.runtime.executiongraph.failover.flip1.NoRestartBackoffTimeStrategy) SlotPoolTestUtils.offerSlots(org.apache.flink.runtime.jobmaster.slotpool.SlotPoolTestUtils.offerSlots) JobManagerMetricGroup(org.apache.flink.runtime.metrics.groups.JobManagerMetricGroup) ComponentMainThreadExecutorServiceAdapter(org.apache.flink.runtime.concurrent.ComponentMainThreadExecutorServiceAdapter) Assertions.assertThat(org.assertj.core.api.Assertions.assertThat) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) CheckpointCoordinatorConfiguration(org.apache.flink.runtime.jobgraph.tasks.CheckpointCoordinatorConfiguration) SuppressRestartsException(org.apache.flink.runtime.execution.SuppressRestartsException) ArchivedExecutionGraphTest(org.apache.flink.runtime.executiongraph.ArchivedExecutionGraphTest) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) TestLogger(org.apache.flink.util.TestLogger) JobCheckpointingSettings(org.apache.flink.runtime.jobgraph.tasks.JobCheckpointingSettings) CheckpointIDCounter(org.apache.flink.runtime.checkpoint.CheckpointIDCounter) ArchivedExecutionJobVertex(org.apache.flink.runtime.executiongraph.ArchivedExecutionJobVertex) TestingCompletedCheckpointStore(org.apache.flink.runtime.checkpoint.TestingCompletedCheckpointStore) TestRestartBackoffTimeStrategy(org.apache.flink.runtime.executiongraph.failover.flip1.TestRestartBackoffTimeStrategy) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) Collectors(java.util.stream.Collectors) ResourceProfile(org.apache.flink.runtime.clusterframework.types.ResourceProfile) DefaultDeclarativeSlotPool(org.apache.flink.runtime.jobmaster.slotpool.DefaultDeclarativeSlotPool) List(java.util.List) StandaloneCheckpointIDCounter(org.apache.flink.runtime.checkpoint.StandaloneCheckpointIDCounter) CoordinationRequest(org.apache.flink.runtime.operators.coordination.CoordinationRequest) OperatorID(org.apache.flink.runtime.jobgraph.OperatorID) Optional(java.util.Optional) SchedulerExecutionMode(org.apache.flink.configuration.SchedulerExecutionMode) KeyGroupRangeAssignment(org.apache.flink.runtime.state.KeyGroupRangeAssignment) OneShotLatch(org.apache.flink.core.testutils.OneShotLatch) SchedulerBase(org.apache.flink.runtime.scheduler.SchedulerBase) SavepointFormatType(org.apache.flink.core.execution.SavepointFormatType) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) CompletableFuture(java.util.concurrent.CompletableFuture) VertexParallelismInformation(org.apache.flink.runtime.scheduler.VertexParallelismInformation) TaskDeploymentDescriptor(org.apache.flink.runtime.deployment.TaskDeploymentDescriptor) TestingCheckpointIDCounter(org.apache.flink.runtime.checkpoint.TestingCheckpointIDCounter) TestingMetricRegistry(org.apache.flink.runtime.metrics.util.TestingMetricRegistry) Nonnull(javax.annotation.Nonnull) StandaloneCompletedCheckpointStore(org.apache.flink.runtime.checkpoint.StandaloneCompletedCheckpointStore) DefaultSchedulerTest(org.apache.flink.runtime.scheduler.DefaultSchedulerTest) JobGraphTestUtils.streamingJobGraph(org.apache.flink.runtime.jobgraph.JobGraphTestUtils.streamingJobGraph) ArchivedExecutionGraph(org.apache.flink.runtime.executiongraph.ArchivedExecutionGraph) ArchivedExecutionGraphBuilder(org.apache.flink.runtime.rest.handler.legacy.utils.ArchivedExecutionGraphBuilder) Logger(org.slf4j.Logger) Configuration(org.apache.flink.configuration.Configuration) CompletedCheckpointStore(org.apache.flink.runtime.checkpoint.CompletedCheckpointStore) JobStatusListener(org.apache.flink.runtime.executiongraph.JobStatusListener) DefaultDeclarativeSlotPoolTest.createSlotOffersForResourceRequirements(org.apache.flink.runtime.jobmaster.slotpool.DefaultDeclarativeSlotPoolTest.createSlotOffersForResourceRequirements) TimeUnit(java.util.concurrent.TimeUnit) Consumer(java.util.function.Consumer) ExecutionAttemptID(org.apache.flink.runtime.executiongraph.ExecutionAttemptID) ExecutionGraphTestUtils.createNoOpVertex(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.createNoOpVertex) SchedulerTestingUtils.enableCheckpointing(org.apache.flink.runtime.scheduler.SchedulerTestingUtils.enableCheckpointing) TaskExecutionState(org.apache.flink.runtime.taskmanager.TaskExecutionState) TemporaryFolder(org.junit.rules.TemporaryFolder) CheckpointCoordinatorConfiguration(org.apache.flink.runtime.jobgraph.tasks.CheckpointCoordinatorConfiguration) Configuration(org.apache.flink.configuration.Configuration) TaskExecutionStateTransition(org.apache.flink.runtime.executiongraph.TaskExecutionStateTransition) ArrayList(java.util.ArrayList) TestingCheckpointRecoveryFactory(org.apache.flink.runtime.checkpoint.TestingCheckpointRecoveryFactory) CompletableFuture(java.util.concurrent.CompletableFuture) CheckpointsCleaner(org.apache.flink.runtime.checkpoint.CheckpointsCleaner) CheckpointIDCounter(org.apache.flink.runtime.checkpoint.CheckpointIDCounter) StandaloneCheckpointIDCounter(org.apache.flink.runtime.checkpoint.StandaloneCheckpointIDCounter) TestingCheckpointIDCounter(org.apache.flink.runtime.checkpoint.TestingCheckpointIDCounter) ExecutionAttemptID(org.apache.flink.runtime.executiongraph.ExecutionAttemptID) ArchivedExecutionVertex(org.apache.flink.runtime.executiongraph.ArchivedExecutionVertex) TaskExecutionState(org.apache.flink.runtime.taskmanager.TaskExecutionState) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobGraphTestUtils.streamingJobGraph(org.apache.flink.runtime.jobgraph.JobGraphTestUtils.streamingJobGraph) StandaloneCompletedCheckpointStore(org.apache.flink.runtime.checkpoint.StandaloneCompletedCheckpointStore) DefaultDeclarativeSlotPool(org.apache.flink.runtime.jobmaster.slotpool.DefaultDeclarativeSlotPool) TestingCompletedCheckpointStore(org.apache.flink.runtime.checkpoint.TestingCompletedCheckpointStore) StandaloneCompletedCheckpointStore(org.apache.flink.runtime.checkpoint.StandaloneCompletedCheckpointStore) CompletedCheckpointStore(org.apache.flink.runtime.checkpoint.CompletedCheckpointStore) StandaloneCheckpointIDCounter(org.apache.flink.runtime.checkpoint.StandaloneCheckpointIDCounter)

Aggregations

StandaloneCheckpointIDCounter (org.apache.flink.runtime.checkpoint.StandaloneCheckpointIDCounter)6 CheckpointsCleaner (org.apache.flink.runtime.checkpoint.CheckpointsCleaner)5 StandaloneCompletedCheckpointStore (org.apache.flink.runtime.checkpoint.StandaloneCompletedCheckpointStore)5 JobGraph (org.apache.flink.runtime.jobgraph.JobGraph)5 Test (org.junit.Test)5 CheckpointIDCounter (org.apache.flink.runtime.checkpoint.CheckpointIDCounter)4 IOException (java.io.IOException)3 CompletableFuture (java.util.concurrent.CompletableFuture)3 JobStatus (org.apache.flink.api.common.JobStatus)3 Configuration (org.apache.flink.configuration.Configuration)3 CompletedCheckpointStore (org.apache.flink.runtime.checkpoint.CompletedCheckpointStore)3 Duration (java.time.Duration)2 ArrayList (java.util.ArrayList)2 Arrays (java.util.Arrays)2 List (java.util.List)2 Optional (java.util.Optional)2 ArrayBlockingQueue (java.util.concurrent.ArrayBlockingQueue)2 BlockingQueue (java.util.concurrent.BlockingQueue)2 ExecutionException (java.util.concurrent.ExecutionException)2 Executors (java.util.concurrent.Executors)2