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

Example 1 with RootExceptionHistoryEntry

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

the class DefaultSchedulerTest method testExceptionHistoryConcurrentRestart.

@Test
public void testExceptionHistoryConcurrentRestart() throws Exception {
    final JobGraph jobGraph = singleJobVertexJobGraph(2);
    final TaskManagerLocation taskManagerLocation = new LocalTaskManagerLocation();
    final TestingLogicalSlotBuilder logicalSlotBuilder = new TestingLogicalSlotBuilder();
    logicalSlotBuilder.setTaskManagerLocation(taskManagerLocation);
    executionSlotAllocatorFactory = new TestExecutionSlotAllocatorFactory(logicalSlotBuilder);
    final ReorganizableManuallyTriggeredScheduledExecutor delayExecutor = new ReorganizableManuallyTriggeredScheduledExecutor();
    final TestFailoverStrategyFactory failoverStrategyFactory = new TestFailoverStrategyFactory();
    final DefaultScheduler scheduler = createScheduler(jobGraph, ComponentMainThreadExecutorServiceAdapter.forMainThread(), new PipelinedRegionSchedulingStrategy.Factory(), failoverStrategyFactory, delayExecutor);
    scheduler.startScheduling();
    final ExecutionVertex executionVertex0 = Iterables.get(scheduler.getExecutionGraph().getAllExecutionVertices(), 0);
    final ExecutionVertex executionVertex1 = Iterables.get(scheduler.getExecutionGraph().getAllExecutionVertices(), 1);
    // single-ExecutionVertex failure
    final RuntimeException exception0 = new RuntimeException("failure #0");
    failoverStrategyFactory.setTasksToRestart(executionVertex0.getID());
    final long updateStateTriggeringRestartTimestamp0 = initiateFailure(scheduler, executionVertex0.getCurrentExecutionAttempt().getAttemptId(), exception0);
    // multi-ExecutionVertex failure
    final RuntimeException exception1 = new RuntimeException("failure #1");
    failoverStrategyFactory.setTasksToRestart(executionVertex1.getID(), executionVertex0.getID());
    final long updateStateTriggeringRestartTimestamp1 = initiateFailure(scheduler, executionVertex1.getCurrentExecutionAttempt().getAttemptId(), exception1);
    // there might be a race condition with the delayExecutor if the tasks are scheduled quite
    // close to each other which we want to simulate here
    Collections.reverse(delayExecutor.getCollectedScheduledTasks());
    delayExecutor.triggerNonPeriodicScheduledTasks();
    assertThat(scheduler.getExceptionHistory(), IsIterableWithSize.iterableWithSize(2));
    final Iterator<RootExceptionHistoryEntry> actualExceptionHistory = scheduler.getExceptionHistory().iterator();
    final RootExceptionHistoryEntry entry0 = actualExceptionHistory.next();
    assertThat(entry0, is(ExceptionHistoryEntryMatcher.matchesFailure(exception0, updateStateTriggeringRestartTimestamp0, executionVertex0.getTaskNameWithSubtaskIndex(), executionVertex0.getCurrentAssignedResourceLocation())));
    assertThat(entry0.getConcurrentExceptions(), IsIterableContainingInOrder.contains(ExceptionHistoryEntryMatcher.matchesFailure(exception1, updateStateTriggeringRestartTimestamp1, executionVertex1.getTaskNameWithSubtaskIndex(), executionVertex1.getCurrentAssignedResourceLocation())));
    final RootExceptionHistoryEntry entry1 = actualExceptionHistory.next();
    assertThat(entry1, is(ExceptionHistoryEntryMatcher.matchesFailure(exception1, updateStateTriggeringRestartTimestamp1, executionVertex1.getTaskNameWithSubtaskIndex(), executionVertex1.getCurrentAssignedResourceLocation())));
    assertThat(entry1.getConcurrentExceptions(), IsEmptyIterable.emptyIterable());
}

Example 2 with RootExceptionHistoryEntry

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

the class AdaptiveSchedulerTest method testExceptionHistoryWithTaskFailure.

@Test
public void testExceptionHistoryWithTaskFailure() throws Exception {
    final Exception expectedException = new Exception("Expected Local Exception");
    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);
    assertThat(actualExceptionHistory).hasSize(1);
    final RootExceptionHistoryEntry failure = actualExceptionHistory.iterator().next();
    assertThat(failure.getException().deserializeError(classLoader)).isEqualTo(expectedException);
}

Example 3 with RootExceptionHistoryEntry

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

the class AdaptiveSchedulerTest method testExceptionHistoryWithTaskConcurrentGlobalFailure.

@Test
public void testExceptionHistoryWithTaskConcurrentGlobalFailure() throws Exception {
    final Exception expectedException1 = new Exception("Expected Global Exception 1");
    final Exception expectedException2 = new Exception("Expected Global Exception 2");
    BiConsumer<AdaptiveScheduler, List<ExecutionAttemptID>> testLogic = (scheduler, attemptIds) -> {
        scheduler.handleGlobalFailure(expectedException1);
        scheduler.handleGlobalFailure(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());
    assertThat(foundExceptions).containsExactly(expectedException2);
}

Example 4 with RootExceptionHistoryEntry

use of org.apache.flink.runtime.scheduler.exceptionhistory.RootExceptionHistoryEntry 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 5 with RootExceptionHistoryEntry

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

the class DefaultJobMasterServiceProcessTest method testInitializationFailureSetsExceptionHistoryProperly.

@Test
public void testInitializationFailureSetsExceptionHistoryProperly() throws ExecutionException, InterruptedException {
    final CompletableFuture<JobMasterService> jobMasterServiceFuture = new CompletableFuture<>();
    DefaultJobMasterServiceProcess serviceProcess = createTestInstance(jobMasterServiceFuture);
    final RuntimeException originalCause = new RuntimeException("Expected RuntimeException");
    long beforeFailureTimestamp = System.currentTimeMillis();
    jobMasterServiceFuture.completeExceptionally(originalCause);
    long afterFailureTimestamp = System.currentTimeMillis();
    final RootExceptionHistoryEntry entry = Iterables.getOnlyElement(serviceProcess.getResultFuture().get().getExecutionGraphInfo().getExceptionHistory());
    assertInitializationException(entry.getException(), originalCause, entry.getTimestamp(), beforeFailureTimestamp, afterFailureTimestamp);
    assertThat(entry.isGlobal()).isTrue();
}