Examples with TaskStatusUpdateEvent org.apache.tez.runtime.api.events.TaskStatusUpdateEvent used on opensource projects

Example 11 with TaskStatusUpdateEvent

use of org.apache.tez.runtime.api.events.TaskStatusUpdateEvent 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 12 with TaskStatusUpdateEvent

use of org.apache.tez.runtime.api.events.TaskStatusUpdateEvent in project tez by apache.

the class TestTaskAttempt method testNoProgressFail.

@Test(timeout = 5000)
public void testNoProgressFail() 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.setLong(TezConfiguration.TEZ_TASK_PROGRESS_STUCK_INTERVAL_MS, 75);
    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);
    Clock mockClock = mock(Clock.class);
    TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, mockClock, 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);
    when(mockClock.getTime()).thenReturn(100l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, true)));
    // invocations and time updated
    assertEquals(100l, taImpl.lastNotifyProgressTimestamp);
    when(mockClock.getTime()).thenReturn(150l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, true)));
    // invocations and time updated
    assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
    when(mockClock.getTime()).thenReturn(200l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    // invocations and time not updated
    assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
    when(mockClock.getTime()).thenReturn(250l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    // invocations and time not updated
    assertEquals(150l, taImpl.lastNotifyProgressTimestamp);
    // failed event sent to self
    verify(eventHandler, atLeast(1)).handle(arg.capture());
    TaskAttemptEventAttemptFailed fEvent = (TaskAttemptEventAttemptFailed) arg.getValue();
    assertEquals(taImpl.getID(), fEvent.getTaskAttemptID());
    assertEquals(TaskAttemptTerminationCause.NO_PROGRESS, fEvent.getTerminationCause());
    assertEquals(TaskFailureType.NON_FATAL, fEvent.getTaskFailureType());
    taImpl.handle(fEvent);
    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(TaskAttemptTerminationCause.NO_PROGRESS, taImpl.getTerminationCause());
}

Example 13 with TaskStatusUpdateEvent

use of org.apache.tez.runtime.api.events.TaskStatusUpdateEvent in project tez by apache.

the class TestTaskAttempt method testLastDataEventRecording.

@Test(timeout = 5000)
public void testLastDataEventRecording() 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();
    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);
    long ts1 = 1024;
    long ts2 = 2048;
    TezTaskAttemptID mockId1 = mock(TezTaskAttemptID.class);
    TezTaskAttemptID mockId2 = mock(TezTaskAttemptID.class);
    TezEvent mockTezEvent1 = mock(TezEvent.class, RETURNS_DEEP_STUBS);
    when(mockTezEvent1.getEventReceivedTime()).thenReturn(ts1);
    when(mockTezEvent1.getSourceInfo().getTaskAttemptID()).thenReturn(mockId1);
    TezEvent mockTezEvent2 = mock(TezEvent.class, RETURNS_DEEP_STUBS);
    when(mockTezEvent2.getEventReceivedTime()).thenReturn(ts2);
    when(mockTezEvent2.getSourceInfo().getTaskAttemptID()).thenReturn(mockId2);
    TaskAttemptEventStatusUpdate statusEvent = new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false));
    assertEquals(0, taImpl.lastDataEvents.size());
    taImpl.setLastEventSent(mockTezEvent1);
    assertEquals(1, taImpl.lastDataEvents.size());
    assertEquals(ts1, taImpl.lastDataEvents.get(0).getTimestamp());
    assertEquals(mockId1, taImpl.lastDataEvents.get(0).getTaskAttemptId());
    taImpl.handle(statusEvent);
    taImpl.setLastEventSent(mockTezEvent2);
    assertEquals(1, taImpl.lastDataEvents.size());
    assertEquals(ts2, taImpl.lastDataEvents.get(0).getTimestamp());
    // over-write earlier value
    assertEquals(mockId2, taImpl.lastDataEvents.get(0).getTaskAttemptId());
    statusEvent.setReadErrorReported(true);
    taImpl.handle(statusEvent);
    taImpl.setLastEventSent(mockTezEvent1);
    assertEquals(2, taImpl.lastDataEvents.size());
    assertEquals(ts1, taImpl.lastDataEvents.get(1).getTimestamp());
    // add new event
    assertEquals(mockId1, taImpl.lastDataEvents.get(1).getTaskAttemptId());
    taImpl.setLastEventSent(mockTezEvent2);
    assertEquals(2, taImpl.lastDataEvents.size());
    assertEquals(ts2, taImpl.lastDataEvents.get(1).getTimestamp());
    // over-write earlier value
    assertEquals(mockId2, taImpl.lastDataEvents.get(1).getTaskAttemptId());
}

Example 14 with TaskStatusUpdateEvent

use of org.apache.tez.runtime.api.events.TaskStatusUpdateEvent in project tez by apache.

the class TestTaskReporter method testStatusUpdateAfterInitializationAndCounterFlag.

@Test(timeout = 5000)
public void testStatusUpdateAfterInitializationAndCounterFlag() {
    TezTaskAttemptID mockTaskAttemptId = mock(TezTaskAttemptID.class);
    LogicalIOProcessorRuntimeTask mockTask = mock(LogicalIOProcessorRuntimeTask.class);
    doReturn("vertexName").when(mockTask).getVertexName();
    doReturn(mockTaskAttemptId).when(mockTask).getTaskAttemptID();
    boolean progressNotified = false;
    doReturn(progressNotified).when(mockTask).getAndClearProgressNotification();
    TezTaskUmbilicalProtocol mockUmbilical = mock(TezTaskUmbilicalProtocol.class);
    float progress = 0.5f;
    TaskStatistics stats = new TaskStatistics();
    TezCounters counters = new TezCounters();
    doReturn(progress).when(mockTask).getProgress();
    doReturn(stats).when(mockTask).getTaskStatistics();
    doReturn(counters).when(mockTask).getCounters();
    // Setup the sleep time to be way higher than the test timeout
    TaskReporter.HeartbeatCallable heartbeatCallable = new TaskReporter.HeartbeatCallable(mockTask, mockUmbilical, 100000, 100000, 5, new AtomicLong(0), "containerIdStr");
    // task not initialized - nothing obtained from task
    doReturn(false).when(mockTask).hasInitialized();
    TaskStatusUpdateEvent event = heartbeatCallable.getStatusUpdateEvent(true);
    verify(mockTask, times(1)).hasInitialized();
    verify(mockTask, times(0)).getProgress();
    verify(mockTask, times(0)).getAndClearProgressNotification();
    verify(mockTask, times(0)).getTaskStatistics();
    verify(mockTask, times(0)).getCounters();
    Assert.assertEquals(0, event.getProgress(), 0);
    Assert.assertEquals(false, event.getProgressNotified());
    Assert.assertNull(event.getCounters());
    Assert.assertNull(event.getStatistics());
    // task is initialized - progress obtained but not counters since flag is false
    doReturn(true).when(mockTask).hasInitialized();
    event = heartbeatCallable.getStatusUpdateEvent(false);
    verify(mockTask, times(2)).hasInitialized();
    verify(mockTask, times(1)).getProgress();
    verify(mockTask, times(1)).getAndClearProgressNotification();
    verify(mockTask, times(0)).getTaskStatistics();
    verify(mockTask, times(0)).getCounters();
    Assert.assertEquals(progress, event.getProgress(), 0);
    Assert.assertEquals(progressNotified, event.getProgressNotified());
    Assert.assertNull(event.getCounters());
    Assert.assertNull(event.getStatistics());
    // task is initialized - progress obtained and also counters since flag is true
    progressNotified = true;
    doReturn(progressNotified).when(mockTask).getAndClearProgressNotification();
    doReturn(true).when(mockTask).hasInitialized();
    event = heartbeatCallable.getStatusUpdateEvent(true);
    verify(mockTask, times(3)).hasInitialized();
    verify(mockTask, times(2)).getProgress();
    verify(mockTask, times(2)).getAndClearProgressNotification();
    verify(mockTask, times(1)).getTaskStatistics();
    verify(mockTask, times(1)).getCounters();
    Assert.assertEquals(progress, event.getProgress(), 0);
    Assert.assertEquals(progressNotified, event.getProgressNotified());
    Assert.assertEquals(counters, event.getCounters());
    Assert.assertEquals(stats, event.getStatistics());
}

Example 15 with TaskStatusUpdateEvent

use of org.apache.tez.runtime.api.events.TaskStatusUpdateEvent in project tez by apache.

the class TezEvent method serializeEvent.

private void serializeEvent(DataOutput out) throws IOException {
    if (event == null) {
        out.writeBoolean(false);
        return;
    }
    out.writeBoolean(true);
    out.writeInt(eventType.ordinal());
    out.writeLong(eventReceivedTime);
    if (eventType.equals(EventType.TASK_STATUS_UPDATE_EVENT)) {
        // TODO NEWTEZ convert to PB
        TaskStatusUpdateEvent sEvt = (TaskStatusUpdateEvent) event;
        sEvt.write(out);
    } else {
        AbstractMessage message;
        switch(eventType) {
            case CUSTOM_PROCESSOR_EVENT:
                message = ProtoConverters.convertCustomProcessorEventToProto((CustomProcessorEvent) event);
                break;
            case DATA_MOVEMENT_EVENT:
                message = ProtoConverters.convertDataMovementEventToProto((DataMovementEvent) event);
                break;
            case COMPOSITE_ROUTED_DATA_MOVEMENT_EVENT:
                message = ProtoConverters.convertCompositeRoutedDataMovementEventToProto((CompositeRoutedDataMovementEvent) event);
                break;
            case COMPOSITE_DATA_MOVEMENT_EVENT:
                message = ProtoConverters.convertCompositeDataMovementEventToProto((CompositeDataMovementEvent) event);
                break;
            case VERTEX_MANAGER_EVENT:
                message = ProtoConverters.convertVertexManagerEventToProto((VertexManagerEvent) event);
                break;
            case INPUT_READ_ERROR_EVENT:
                InputReadErrorEvent ideEvt = (InputReadErrorEvent) event;
                message = InputReadErrorEventProto.newBuilder().setIndex(ideEvt.getIndex()).setDiagnostics(ideEvt.getDiagnostics()).setVersion(ideEvt.getVersion()).build();
                break;
            case TASK_ATTEMPT_FAILED_EVENT:
                TaskAttemptFailedEvent tfEvt = (TaskAttemptFailedEvent) event;
                message = TaskAttemptFailedEventProto.newBuilder().setDiagnostics(tfEvt.getDiagnostics()).setTaskFailureType(TezConverterUtils.failureTypeToProto(tfEvt.getTaskFailureType())).build();
                break;
            case TASK_ATTEMPT_KILLED_EVENT:
                TaskAttemptKilledEvent tkEvent = (TaskAttemptKilledEvent) event;
                message = TaskAttemptKilledEventProto.newBuilder().setDiagnostics(tkEvent.getDiagnostics()).build();
                break;
            case TASK_ATTEMPT_COMPLETED_EVENT:
                message = TaskAttemptCompletedEventProto.newBuilder().build();
                break;
            case INPUT_FAILED_EVENT:
                InputFailedEvent ifEvt = (InputFailedEvent) event;
                message = InputFailedEventProto.newBuilder().setTargetIndex(ifEvt.getTargetIndex()).setVersion(ifEvt.getVersion()).build();
                break;
            case ROOT_INPUT_DATA_INFORMATION_EVENT:
                message = ProtoConverters.convertRootInputDataInformationEventToProto((InputDataInformationEvent) event);
                break;
            case ROOT_INPUT_INITIALIZER_EVENT:
                message = ProtoConverters.convertRootInputInitializerEventToProto((InputInitializerEvent) event);
                break;
            default:
                throw new TezUncheckedException("Unknown TezEvent" + ", type=" + eventType);
        }
        if (out instanceof OutputStream) {
            // DataOutputBuffer extends DataOutputStream
            int serializedSize = message.getSerializedSize();
            out.writeInt(serializedSize);
            int buffersize = serializedSize < CodedOutputStream.DEFAULT_BUFFER_SIZE ? serializedSize : CodedOutputStream.DEFAULT_BUFFER_SIZE;
            CodedOutputStream codedOut = CodedOutputStream.newInstance((OutputStream) out, buffersize);
            message.writeTo(codedOut);
            codedOut.flush();
        } else {
            byte[] eventBytes = message.toByteArray();
            out.writeInt(eventBytes.length);
            out.write(eventBytes);
        }
    }
}