Examples with TestingCheckpointRecoveryFactory org.apache.flink.runtime.checkpoint.TestingCheckpointRecoveryFactory used on opensource projects

Example 1 with TestingCheckpointRecoveryFactory

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

the class DefaultSchedulerTest method doTestCheckpointCleanerIsClosedAfterCheckpointServices.

/**
 * Visible for re-use in {@link
 * org.apache.flink.runtime.scheduler.adaptive.AdaptiveSchedulerTest}.
 */
public static void doTestCheckpointCleanerIsClosedAfterCheckpointServices(BiFunction<CheckpointRecoveryFactory, CheckpointsCleaner, SchedulerNG> schedulerFactory, ScheduledExecutorService executorService) throws Exception {
    final CountDownLatch checkpointServicesShutdownBlocked = new CountDownLatch(1);
    final CountDownLatch cleanerClosed = new CountDownLatch(1);
    final CompletedCheckpointStore completedCheckpointStore = new StandaloneCompletedCheckpointStore(1) {

        @Override
        public void shutdown(JobStatus jobStatus, CheckpointsCleaner checkpointsCleaner) throws Exception {
            checkpointServicesShutdownBlocked.await();
            super.shutdown(jobStatus, checkpointsCleaner);
        }
    };
    final CheckpointIDCounter checkpointIDCounter = new StandaloneCheckpointIDCounter() {

        @Override
        public void shutdown(JobStatus jobStatus) throws Exception {
            checkpointServicesShutdownBlocked.await();
            super.shutdown(jobStatus);
        }
    };
    final CheckpointsCleaner checkpointsCleaner = new CheckpointsCleaner() {

        @Override
        public synchronized CompletableFuture<Void> closeAsync() {
            cleanerClosed.countDown();
            return super.closeAsync();
        }
    };
    final SchedulerNG scheduler = schedulerFactory.apply(new TestingCheckpointRecoveryFactory(completedCheckpointStore, checkpointIDCounter), checkpointsCleaner);
    final CompletableFuture<Void> schedulerClosed = new CompletableFuture<>();
    final CountDownLatch schedulerClosing = new CountDownLatch(1);
    executorService.submit(() -> {
        scheduler.closeAsync().thenRun(() -> schedulerClosed.complete(null));
        schedulerClosing.countDown();
    });
    // Wait for scheduler to start closing.
    schedulerClosing.await();
    assertFalse("CheckpointCleaner should not close before checkpoint services.", cleanerClosed.await(10, TimeUnit.MILLISECONDS));
    checkpointServicesShutdownBlocked.countDown();
    cleanerClosed.await();
    schedulerClosed.get();
}

Example 2 with TestingCheckpointRecoveryFactory

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

the class AdaptiveSchedulerTest method testExceptionHistoryWithTaskFailureFromStopWithSavepoint.

@Test
public void testExceptionHistoryWithTaskFailureFromStopWithSavepoint() throws Exception {
    final Exception expectedException = new Exception("Expected Local Exception");
    Consumer<JobGraph> setupJobGraph = jobGraph -> jobGraph.setSnapshotSettings(new JobCheckpointingSettings(CheckpointCoordinatorConfiguration.builder().build(), null));
    final CompletedCheckpointStore completedCheckpointStore = new StandaloneCompletedCheckpointStore(1);
    final CheckpointIDCounter checkpointIDCounter = new StandaloneCheckpointIDCounter();
    final CheckpointsCleaner checkpointCleaner = new CheckpointsCleaner();
    TestingCheckpointRecoveryFactory checkpointRecoveryFactory = new TestingCheckpointRecoveryFactory(completedCheckpointStore, checkpointIDCounter);
    Consumer<AdaptiveSchedulerBuilder> setupScheduler = builder -> builder.setCheckpointRecoveryFactory(checkpointRecoveryFactory).setCheckpointCleaner(checkpointCleaner);
    BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic = (scheduler, attemptIds) -> {
        final ExecutionAttemptID attemptId = attemptIds.get(1);
        scheduler.stopWithSavepoint("file:///tmp/target", true, SavepointFormatType.CANONICAL);
        scheduler.updateTaskExecutionState(new TaskExecutionStateTransition(new TaskExecutionState(attemptId, ExecutionState.FAILED, expectedException)));
    };
    final Iterable<RootExceptionHistoryEntry> actualExceptionHistory = runExceptionHistoryTests(testLogic, setupScheduler, setupJobGraph);
    assertThat(actualExceptionHistory).hasSize(1);
    final RootExceptionHistoryEntry failure = actualExceptionHistory.iterator().next();
    assertThat(failure.getException().deserializeError(classLoader)).isEqualTo(expectedException);
}

Example 3 with TestingCheckpointRecoveryFactory

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

the class AdaptiveSchedulerTest method testCloseShutsDownCheckpointingComponents.

@Test
public void testCloseShutsDownCheckpointingComponents() throws Exception {
    final CompletableFuture<JobStatus> completedCheckpointStoreShutdownFuture = new CompletableFuture<>();
    final CompletedCheckpointStore completedCheckpointStore = TestingCompletedCheckpointStore.createStoreWithShutdownCheckAndNoCompletedCheckpoints(completedCheckpointStoreShutdownFuture);
    final CompletableFuture<JobStatus> checkpointIdCounterShutdownFuture = new CompletableFuture<>();
    final CheckpointIDCounter checkpointIdCounter = TestingCheckpointIDCounter.createStoreWithShutdownCheckAndNoStartAction(checkpointIdCounterShutdownFuture);
    final JobGraph jobGraph = createJobGraph();
    // checkpointing components are only created if checkpointing is enabled
    jobGraph.setSnapshotSettings(new JobCheckpointingSettings(CheckpointCoordinatorConfiguration.builder().build(), null));
    final AdaptiveScheduler scheduler = new AdaptiveSchedulerBuilder(jobGraph, singleThreadMainThreadExecutor).setCheckpointRecoveryFactory(new TestingCheckpointRecoveryFactory(completedCheckpointStore, checkpointIdCounter)).build();
    singleThreadMainThreadExecutor.execute(() -> {
        scheduler.startScheduling();
        // transition into the FAILED state
        scheduler.handleGlobalFailure(new FlinkException("Test exception"));
        scheduler.closeAsync();
    });
    assertThat(completedCheckpointStoreShutdownFuture.get()).isEqualTo(JobStatus.FAILED);
    assertThat(checkpointIdCounterShutdownFuture.get()).isEqualTo(JobStatus.FAILED);
}

Example 4 with TestingCheckpointRecoveryFactory

use of org.apache.flink.runtime.checkpoint.TestingCheckpointRecoveryFactory 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();
}