Examples with TaskEventTAFailed org.apache.tez.dag.app.dag.event.TaskEventTAFailed used on opensource projects

Example 6 with TaskEventTAFailed

use of org.apache.tez.dag.app.dag.event.TaskEventTAFailed in project tez by apache.

the class TestTaskImpl method failAttempt.

private void failAttempt(MockTaskAttemptImpl taskAttempt, int index, int expectedIncompleteAttempts) {
    InputReadErrorEvent mockReEvent = InputReadErrorEvent.create("", 0, index);
    TezTaskAttemptID mockDestId = mock(TezTaskAttemptID.class);
    EventMetaData meta = new EventMetaData(EventProducerConsumerType.INPUT, "Vertex", "Edge", mockDestId);
    TezEvent tzEvent = new TezEvent(mockReEvent, meta);
    TaskAttemptEventOutputFailed outputFailedEvent = new TaskAttemptEventOutputFailed(mockDestId, tzEvent, 1);
    taskAttempt.handle(outputFailedEvent);
    TaskEvent tEventFail1 = new TaskEventTAFailed(taskAttempt.getID(), TaskFailureType.NON_FATAL, outputFailedEvent);
    mockTask.handle(tEventFail1);
    assertEquals("Unexpected number of incomplete attempts!", expectedIncompleteAttempts, mockTask.getUncompletedAttemptsCount());
}

Example 7 with TaskEventTAFailed

use of org.apache.tez.dag.app.dag.event.TaskEventTAFailed in project tez by apache.

the class TestTaskImpl method testFailedTaskTransitionWithLaunchedAttempt.

@Test(timeout = 30000)
public void testFailedTaskTransitionWithLaunchedAttempt() throws InterruptedException {
    Configuration newConf = new Configuration(conf);
    newConf.setInt(TezConfiguration.TEZ_AM_TASK_MAX_FAILED_ATTEMPTS, 1);
    Vertex vertex = mock(Vertex.class);
    doReturn(new VertexImpl.VertexConfigImpl(newConf)).when(vertex).getVertexConfig();
    mockTask = new MockTaskImpl(vertexId, partition, eventHandler, conf, taskCommunicatorManagerInterface, clock, taskHeartbeatHandler, appContext, leafVertex, taskResource, containerContext, vertex);
    TezTaskID taskId = getNewTaskID();
    scheduleTaskAttempt(taskId);
    MockTaskAttemptImpl firstMockTaskAttempt = mockTask.getLastAttempt();
    launchTaskAttempt(firstMockTaskAttempt.getID());
    mockTask.handle(createTaskTAAddSpecAttempt(mockTask.getLastAttempt().getID()));
    MockTaskAttemptImpl secondMockTaskAttempt = mockTask.getLastAttempt();
    launchTaskAttempt(secondMockTaskAttempt.getID());
    firstMockTaskAttempt.handle(new TaskAttemptEventSchedule(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), 10, 10));
    secondMockTaskAttempt.handle(new TaskAttemptEventSchedule(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), 10, 10));
    firstMockTaskAttempt.handle(new TaskAttemptEventSubmitted(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), mockContainer.getId()));
    secondMockTaskAttempt.handle(new TaskAttemptEventSubmitted(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), mockContainer.getId()));
    secondMockTaskAttempt.handle(new TaskAttemptEventStartedRemotely(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString())));
    firstMockTaskAttempt.handle(new TaskAttemptEventStartedRemotely(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString())));
    secondMockTaskAttempt.handle(new TaskAttemptEventAttemptFailed(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "test", TaskAttemptTerminationCause.NO_PROGRESS));
    firstMockTaskAttempt.handle(new TaskAttemptEventAttemptFailed(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "test", TaskAttemptTerminationCause.NO_PROGRESS));
    firstMockTaskAttempt.handle(new TaskAttemptEventContainerTerminated(mockContainerId, firstMockTaskAttempt.getID(), "test", TaskAttemptTerminationCause.NO_PROGRESS));
    secondMockTaskAttempt.handle(new TaskAttemptEventContainerTerminated(mockContainerId, secondMockTaskAttempt.getID(), "test", TaskAttemptTerminationCause.NO_PROGRESS));
    mockTask.handle(new TaskEventTAFailed(secondMockTaskAttempt.getID(), TaskFailureType.NON_FATAL, mock(TaskAttemptEvent.class)));
    mockTask.handle(new TaskEventTAFailed(firstMockTaskAttempt.getID(), TaskFailureType.NON_FATAL, mock(TaskAttemptEvent.class)));
    assertTrue("Attempts should have failed!", firstMockTaskAttempt.getInternalState() == TaskAttemptStateInternal.FAILED && secondMockTaskAttempt.getInternalState() == TaskAttemptStateInternal.FAILED);
    assertEquals("Task should have no uncompleted attempts!", 0, mockTask.getUncompletedAttemptsCount());
    assertTrue("Task should have failed!", mockTask.getState() == TaskState.FAILED);
    mockTask.handle(createTaskTAAddSpecAttempt(mockTask.getLastAttempt().getID()));
    MockTaskAttemptImpl thirdMockTaskAttempt = mockTask.getLastAttempt();
    mockTask.handle(createTaskTALauncherEvent(thirdMockTaskAttempt.getID()));
}

Example 8 with TaskEventTAFailed

use of org.apache.tez.dag.app.dag.event.TaskEventTAFailed in project tez by apache.

the class TestVertexImpl method testFailuresMaxPercentExceededSourceTaskAttemptCompletionEvents.

@Test(timeout = 5000)
public void testFailuresMaxPercentExceededSourceTaskAttemptCompletionEvents() throws TezException {
    LOG.info("Testing testFailuresMaxPercentSourceTaskAttemptCompletionEvents");
    // Override the basic setup for this test to inject the specific config setting needed for this test
    useCustomInitializer = false;
    customInitializer = null;
    setupPreDagCreation();
    conf.setFloat(TezConfiguration.TEZ_VERTEX_FAILURES_MAXPERCENT, 50.0f);
    conf.setInt(TezConfiguration.TEZ_AM_TASK_MAX_FAILED_ATTEMPTS, 1);
    dagPlan = createTestDAGPlan();
    setupPostDagCreation();
    initAllVertices(VertexState.INITED);
    VertexImpl v4 = vertices.get("vertex4");
    VertexImpl v5 = vertices.get("vertex5");
    VertexImpl v6 = vertices.get("vertex6");
    startVertex(vertices.get("vertex1"));
    startVertex(vertices.get("vertex2"));
    dispatcher.await();
    LOG.info("Verifying v6 state " + v6.getState());
    Assert.assertEquals(VertexState.RUNNING, v6.getState());
    TezTaskID t1_v4 = TezTaskID.getInstance(v4.getVertexId(), 0);
    TezTaskID t2_v4 = TezTaskID.getInstance(v4.getVertexId(), 1);
    TezTaskID t1_v5 = TezTaskID.getInstance(v5.getVertexId(), 0);
    TezTaskID t2_v5 = TezTaskID.getInstance(v5.getVertexId(), 1);
    TezTaskAttemptID ta1_t1_v4 = TezTaskAttemptID.getInstance(t1_v4, 0);
    TezTaskAttemptID ta1_t2_v4 = TezTaskAttemptID.getInstance(t2_v4, 0);
    TezTaskAttemptID ta1_t1_v5 = TezTaskAttemptID.getInstance(t1_v5, 0);
    TezTaskAttemptID ta1_t2_v5 = TezTaskAttemptID.getInstance(t2_v5, 0);
    TaskSpec taskSpec = new TaskSpec("dag", "vertex", 2, new ProcessorDescriptor(), new ArrayList<InputSpec>(), new ArrayList<OutputSpec>(), null, conf);
    TaskLocationHint locationHint = TaskLocationHint.createTaskLocationHint(null, null);
    // Tasks can only succeed from a scheduled or running state
    dispatcher.getEventHandler().handle(new TaskEventScheduleTask(t1_v4, taskSpec, locationHint, false));
    dispatcher.getEventHandler().handle(new TaskEventScheduleTask(t2_v4, taskSpec, locationHint, false));
    // Completed tasks are more that the max percent failure
    dispatcher.getEventHandler().handle(new TaskEventTAFailed(ta1_t1_v4, TaskFailureType.NON_FATAL, null));
    dispatcher.getEventHandler().handle(new TaskEventTAFailed(ta1_t2_v4, TaskFailureType.NON_FATAL, null));
    dispatcher.getEventHandler().handle(new TaskEventTASucceeded(ta1_t1_v5));
    dispatcher.getEventHandler().handle(new TaskEventTAFailed(ta1_t2_v5, TaskFailureType.NON_FATAL, null));
    dispatcher.await();
    Assert.assertEquals(VertexState.FAILED, v4.getState());
    Assert.assertEquals(VertexState.SUCCEEDED, v5.getState());
    Assert.assertEquals(VertexState.RUNNING, v6.getState());
    Assert.assertEquals(2, v6.numSuccessSourceAttemptCompletions);
}

Example 9 with TaskEventTAFailed

use of org.apache.tez.dag.app.dag.event.TaskEventTAFailed in project tez by apache.

the class TestTaskAttempt method testFailure.

@Test(timeout = 5000)
public void testFailure() throws Exception {
    ApplicationId appId = ApplicationId.newInstance(1, 2);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 0);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    TezVertexID vertexID = TezVertexID.getInstance(dagID, 1);
    TezTaskID taskID = TezTaskID.getInstance(vertexID, 1);
    MockEventHandler eventHandler = spy(new MockEventHandler());
    TaskCommunicatorManagerInterface taListener = createMockTaskAttemptListener();
    Configuration taskConf = new Configuration();
    taskConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class);
    taskConf.setBoolean("fs.file.impl.disable.cache", true);
    taskConf.setBoolean(TezConfiguration.TEZ_AM_SPECULATION_ENABLED, true);
    locationHint = TaskLocationHint.createTaskLocationHint(new HashSet<String>(Arrays.asList(new String[] { "127.0.0.1" })), null);
    Resource resource = Resource.newInstance(1024, 1);
    NodeId nid = NodeId.newInstance("127.0.0.1", 0);
    @SuppressWarnings("deprecation") ContainerId contId = ContainerId.newInstance(appAttemptId, 3);
    Container container = mock(Container.class);
    when(container.getId()).thenReturn(contId);
    when(container.getNodeId()).thenReturn(nid);
    when(container.getNodeHttpAddress()).thenReturn("localhost:0");
    AMContainerMap containers = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskCommunicatorManagerInterface.class), new ContainerContextMatcher(), appCtx);
    containers.addContainerIfNew(container, 0, 0, 0);
    doReturn(new ClusterInfo()).when(appCtx).getClusterInfo();
    doReturn(containers).when(appCtx).getAllContainers();
    TaskHeartbeatHandler mockHeartbeatHandler = mock(TaskHeartbeatHandler.class);
    TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
    TezTaskAttemptID taskAttemptID = taImpl.getID();
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0));
    taImpl.handle(new TaskAttemptEventSubmitted(taskAttemptID, contId));
    taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
    assertEquals("Task attempt is not in the RUNNING state", taImpl.getState(), TaskAttemptState.RUNNING);
    verify(mockHeartbeatHandler).register(taskAttemptID);
    int expectedEventsAtRunning = 6;
    verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
    verifyEventType(arg.getAllValues().subList(0, expectedEventsAtRunning), SpeculatorEventTaskAttemptStatusUpdate.class, 1);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    taImpl.handle(new TaskAttemptEventAttemptFailed(taskAttemptID, TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "0", TaskAttemptTerminationCause.APPLICATION_ERROR));
    assertEquals("Task attempt is not in the  FAIL_IN_PROGRESS state", taImpl.getInternalState(), TaskAttemptStateInternal.FAIL_IN_PROGRESS);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals(1, taImpl.getDiagnostics().size());
    assertEquals("0", taImpl.getDiagnostics().get(0));
    assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
    assertEquals(TaskAttemptStateInternal.FAIL_IN_PROGRESS, taImpl.getInternalState());
    taImpl.handle(new TaskAttemptEventTezEventUpdate(taImpl.getID(), Collections.EMPTY_LIST));
    assertFalse("InternalError occurred trying to handle TA_TEZ_EVENT_UPDATE in FAIL_IN_PROGRESS state", eventHandler.internalError);
    taImpl.handle(new TaskAttemptEventContainerTerminated(contId, taskAttemptID, "1", TaskAttemptTerminationCause.CONTAINER_EXITED));
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    assertEquals(2, taImpl.getDiagnostics().size());
    assertEquals("1", taImpl.getDiagnostics().get(1));
    // err cause does not change
    assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
    int expectedEvenstAfterTerminating = expectedEventsAtRunning + 5;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    Event e = verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTAFailed.class, 1);
    TaskEventTAFailed failedEvent = (TaskEventTAFailed) e;
    assertEquals(TaskFailureType.NON_FATAL, failedEvent.getTaskFailureType());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), SpeculatorEventTaskAttemptStatusUpdate.class, 2);
}

Example 10 with TaskEventTAFailed

use of org.apache.tez.dag.app.dag.event.TaskEventTAFailed in project tez by apache.

the class TestTaskAttempt method testContainerTerminatedWhileRunning.

@Test(timeout = 5000)
public // Ensure ContainerTerminated is handled correctly by the TaskAttempt
void testContainerTerminatedWhileRunning() throws Exception {
    ApplicationId appId = ApplicationId.newInstance(1, 2);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 0);
    TezDAGID dagID = TezDAGID.getInstance(appId, 1);
    TezVertexID vertexID = TezVertexID.getInstance(dagID, 1);
    TezTaskID taskID = TezTaskID.getInstance(vertexID, 1);
    MockEventHandler eventHandler = spy(new MockEventHandler());
    TaskCommunicatorManagerInterface taListener = createMockTaskAttemptListener();
    Configuration taskConf = new Configuration();
    taskConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class);
    taskConf.setBoolean("fs.file.impl.disable.cache", true);
    locationHint = TaskLocationHint.createTaskLocationHint(new HashSet<String>(Arrays.asList(new String[] { "127.0.0.1" })), null);
    Resource resource = Resource.newInstance(1024, 1);
    NodeId nid = NodeId.newInstance("127.0.0.1", 0);
    @SuppressWarnings("deprecation") ContainerId contId = ContainerId.newInstance(appAttemptId, 3);
    Container container = mock(Container.class);
    when(container.getId()).thenReturn(contId);
    when(container.getNodeId()).thenReturn(nid);
    when(container.getNodeHttpAddress()).thenReturn("localhost:0");
    AMContainerMap containers = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskCommunicatorManagerInterface.class), new ContainerContextMatcher(), appCtx);
    containers.addContainerIfNew(container, 0, 0, 0);
    doReturn(new ClusterInfo()).when(appCtx).getClusterInfo();
    doReturn(containers).when(appCtx).getAllContainers();
    TaskHeartbeatHandler mockHeartbeatHandler = mock(TaskHeartbeatHandler.class);
    TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
    TezTaskAttemptID taskAttemptID = taImpl.getID();
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    taImpl.handle(new TaskAttemptEventSchedule(taskAttemptID, 0, 0));
    taImpl.handle(new TaskAttemptEventSubmitted(taskAttemptID, contId));
    taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
    int expectedEventsAtRunning = 5;
    verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
    assertEquals("Task attempt is not in running state", taImpl.getState(), TaskAttemptState.RUNNING);
    verify(mockHeartbeatHandler).register(taskAttemptID);
    taImpl.handle(new TaskAttemptEventContainerTerminated(contId, taskAttemptID, "Terminated", TaskAttemptTerminationCause.CONTAINER_EXITED));
    assertFalse("InternalError occurred trying to handle TA_CONTAINER_TERMINATED", eventHandler.internalError);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals("Terminated", taImpl.getDiagnostics().get(0));
    assertEquals(TaskAttemptTerminationCause.CONTAINER_EXITED, taImpl.getTerminationCause());
    // TODO Ensure TA_TERMINATING after this is ingored.
    int expectedEvenstAfterTerminating = expectedEventsAtRunning + 3;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    Event event = verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTAFailed.class, 1);
    TaskEventTAFailed failedEvent = (TaskEventTAFailed) event;
    assertEquals(TaskFailureType.NON_FATAL, failedEvent.getTaskFailureType());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
    taImpl.handle(new TaskAttemptEventContainerTerminated(contId, taskAttemptID, "Terminated", TaskAttemptTerminationCause.CONTAINER_EXITED));
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    int expectedEventAfterTerminated = expectedEvenstAfterTerminating + 0;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEventAfterTerminated)).handle(arg.capture());
}