Examples with RootExceptionHistoryEntry org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry used on opensource projects

Example 16 with RootExceptionHistoryEntry

use of org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry in project flink by apache.

the class AdaptiveSchedulerTest method testExceptionHistoryWithTaskConcurrentFailure.

@Test
public void testExceptionHistoryWithTaskConcurrentFailure() throws Exception {
    final Exception expectedException1 = new Exception("Expected Local Exception 1");
    final Exception expectedException2 = new Exception("Expected Local Exception 2");
    BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic = (scheduler, attemptIds) -> {
        final ExecutionAttemptID attemptId = attemptIds.remove(0);
        final ExecutionAttemptID attemptId2 = attemptIds.remove(0);
        scheduler.updateTaskExecutionState(new TaskExecutionStateTransition(new TaskExecutionState(attemptId, ExecutionState.FAILED, expectedException1)));
        scheduler.updateTaskExecutionState(new TaskExecutionStateTransition(new TaskExecutionState(attemptId2, ExecutionState.FAILED, expectedException2)));
    };
    final Iterable<RootExceptionHistoryEntry> entries = runExceptionHistoryTests(testLogic);
    assertThat(entries).hasSize(1);
    final RootExceptionHistoryEntry failure = entries.iterator().next();
    assertThat(failure.getException().deserializeError(classLoader)).isEqualTo(expectedException1);
    final Iterable<ExceptionHistoryEntry> concurrentExceptions = failure.getConcurrentExceptions();
    final List<Throwable> foundExceptions = IterableUtils.toStream(concurrentExceptions).map(ExceptionHistoryEntry::getException).map(exception -> exception.deserializeError(classLoader)).collect(Collectors.toList());
    // In the future, concurrent local failures should be stored.
    assertThat(foundExceptions).isEmpty();
}

Example 17 with RootExceptionHistoryEntry

use of org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry 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();
}

Example 18 with RootExceptionHistoryEntry

use of org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry in project flink by apache.

the class AdaptiveSchedulerTest method testExceptionHistoryWithGlobalFailure.

@Test
public void testExceptionHistoryWithGlobalFailure() throws Exception {
    final Exception expectedException = new Exception("Expected Global Exception");
    BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic = (scheduler, attemptIds) -> scheduler.handleGlobalFailure(expectedException);
    final Iterable<RootExceptionHistoryEntry> actualExceptionHistory = runExceptionHistoryTests(testLogic);
    assertThat(actualExceptionHistory).hasSize(1);
    final RootExceptionHistoryEntry failure = actualExceptionHistory.iterator().next();
    assertThat(failure.getTaskManagerLocation()).isNull();
    assertThat(failure.getFailingTaskName()).isNull();
    assertThat(failure.getException().deserializeError(classLoader)).isEqualTo(expectedException);
}

Example 19 with RootExceptionHistoryEntry

use of org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry in project flink by apache.

the class AdaptiveSchedulerTest method testExceptionHistoryWithTaskFailureWithRestart.

@Test
public void testExceptionHistoryWithTaskFailureWithRestart() throws Exception {
    final Exception expectedException = new Exception("Expected Local Exception");
    Consumer<AdaptiveSchedulerBuilder> setupScheduler = builder -> builder.setRestartBackoffTimeStrategy(new FixedDelayRestartBackoffTimeStrategy.FixedDelayRestartBackoffTimeStrategyFactory(1, 100).create());
    BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic = (scheduler, attemptIds) -> {
        final ExecutionAttemptID attemptId = attemptIds.get(1);
        scheduler.updateTaskExecutionState(new TaskExecutionStateTransition(new TaskExecutionState(attemptId, ExecutionState.FAILED, expectedException)));
    };
    final Iterable<RootExceptionHistoryEntry> actualExceptionHistory = runExceptionHistoryTests(testLogic, setupScheduler);
    assertThat(actualExceptionHistory).hasSize(1);
    final RootExceptionHistoryEntry failure = actualExceptionHistory.iterator().next();
    assertThat(failure.getException().deserializeError(classLoader)).isEqualTo(expectedException);
}