Search in sources :

Example 1 with TaskEventTAFailed

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

the class TestVertexImpl method testFailuresMaxPercentSourceTaskAttemptCompletionEvents.

@Test(timeout = 5000)
public void testFailuresMaxPercentSourceTaskAttemptCompletionEvents() 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 less that the max percent failure
    dispatcher.getEventHandler().handle(new TaskEventTAFailed(ta1_t1_v4, TaskFailureType.NON_FATAL, null));
    dispatcher.getEventHandler().handle(new TaskEventTASucceeded(ta1_t2_v4));
    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.SUCCEEDED, v4.getState());
    Assert.assertEquals(VertexState.SUCCEEDED, v5.getState());
    Assert.assertEquals(VertexState.RUNNING, v6.getState());
    Assert.assertEquals(4, v6.numSuccessSourceAttemptCompletions);
}
Also used : TaskSpec(org.apache.tez.runtime.api.impl.TaskSpec) TaskEventTAFailed(org.apache.tez.dag.app.dag.event.TaskEventTAFailed) ProcessorDescriptor(org.apache.tez.dag.api.ProcessorDescriptor) TaskEventScheduleTask(org.apache.tez.dag.app.dag.event.TaskEventScheduleTask) GroupInputSpec(org.apache.tez.runtime.api.impl.GroupInputSpec) InputSpec(org.apache.tez.runtime.api.impl.InputSpec) TaskEventTASucceeded(org.apache.tez.dag.app.dag.event.TaskEventTASucceeded) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) PlanTaskLocationHint(org.apache.tez.dag.api.records.DAGProtos.PlanTaskLocationHint) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) OutputSpec(org.apache.tez.runtime.api.impl.OutputSpec) VertexManagerPluginForTest(org.apache.tez.test.VertexManagerPluginForTest) Test(org.junit.Test) GraceShuffleVertexManagerForTest(org.apache.tez.test.GraceShuffleVertexManagerForTest) StateChangeNotifierForTest(org.apache.tez.dag.app.dag.TestStateChangeNotifier.StateChangeNotifierForTest) EdgeManagerForTest(org.apache.tez.test.EdgeManagerForTest)

Example 2 with TaskEventTAFailed

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

the class TestTaskAttempt method testContainerTerminationWhileRunning.

@Test(timeout = 5000)
public // the TaskAttempt
void testContainerTerminationWhileRunning() 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));
    assertEquals("Task attempt is not in the STARTING state", taImpl.getState(), TaskAttemptState.STARTING);
    assertEquals("Task attempt internal state is not at SUBMITTED", taImpl.getInternalState(), TaskAttemptStateInternal.SUBMITTED);
    // At state STARTING.
    taImpl.handle(new TaskAttemptEventStartedRemotely(taskAttemptID));
    assertEquals("Task attempt is not in the RUNNING state", taImpl.getState(), TaskAttemptState.RUNNING);
    verify(mockHeartbeatHandler).register(taskAttemptID);
    int expectedEventsAtRunning = 5;
    verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
    taImpl.handle(new TaskAttemptEventContainerTerminating(taskAttemptID, "Terminating", TaskAttemptTerminationCause.APPLICATION_ERROR));
    assertFalse("InternalError occurred trying to handle TA_CONTAINER_TERMINATING", eventHandler.internalError);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals("Task attempt is not in the  FAILED state", taImpl.getState(), TaskAttemptState.FAILED);
    assertEquals(1, taImpl.getDiagnostics().size());
    assertEquals("Terminating", taImpl.getDiagnostics().get(0));
    assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
    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());
    assertEquals(2, taImpl.getDiagnostics().size());
    assertEquals("Terminated", taImpl.getDiagnostics().get(1));
    // check that original error cause is retained
    assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, taImpl.getTerminationCause());
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) Container(org.apache.hadoop.yarn.api.records.Container) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TezDAGID(org.apache.tez.dag.records.TezDAGID) TaskHeartbeatHandler(org.apache.tez.dag.app.TaskHeartbeatHandler) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TezVertexID(org.apache.tez.dag.records.TezVertexID) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) HashSet(java.util.HashSet) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskEventTAFailed(org.apache.tez.dag.app.dag.event.TaskEventTAFailed) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) TaskAttemptEventContainerTerminating(org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminating) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) NodeId(org.apache.hadoop.yarn.api.records.NodeId) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) TaskAttemptFinishedEvent(org.apache.tez.dag.history.events.TaskAttemptFinishedEvent) TaskEvent(org.apache.tez.dag.app.dag.event.TaskEvent) DAGEvent(org.apache.tez.dag.app.dag.event.DAGEvent) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) Event(org.apache.hadoop.yarn.event.Event) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) TaskAttemptEventContainerTerminated(org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminated) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Example 3 with TaskEventTAFailed

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

the class TestTaskAttempt method testMultipleOutputFailed.

@Test(timeout = 5000)
public // TaskAttempt.
void testMultipleOutputFailed() 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 mockEh = new MockEventHandler();
    MockEventHandler eventHandler = spy(mockEh);
    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();
    HistoryEventHandler mockHistHandler = mock(HistoryEventHandler.class);
    doReturn(mockHistHandler).when(appCtx).getHistoryHandler();
    TaskHeartbeatHandler mockHeartbeatHandler = mock(TaskHeartbeatHandler.class);
    MockTaskAttemptImpl 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));
    verify(mockHeartbeatHandler).register(taskAttemptID);
    taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    int expectedEventsTillSucceeded = 8;
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    ArgumentCaptor<DAGHistoryEvent> histArg = ArgumentCaptor.forClass(DAGHistoryEvent.class);
    verify(eventHandler, times(expectedEventsTillSucceeded)).handle(arg.capture());
    // start and finish
    verify(mockHistHandler, times(2)).handle(histArg.capture());
    DAGHistoryEvent histEvent = histArg.getValue();
    TaskAttemptFinishedEvent finishEvent = (TaskAttemptFinishedEvent) histEvent.getHistoryEvent();
    long finishTime = finishEvent.getFinishTime();
    verifyEventType(arg.getAllValues(), TaskEventTAUpdate.class, 2);
    InputReadErrorEvent mockReEvent = InputReadErrorEvent.create("", 0, 1);
    EventMetaData mockMeta = mock(EventMetaData.class);
    TezTaskAttemptID mockDestId1 = mock(TezTaskAttemptID.class);
    when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
    TezEvent tzEvent = new TezEvent(mockReEvent, mockMeta);
    taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
    // failure threshold not met. state is SUCCEEDED
    assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    // sending same error again doesnt change anything
    taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
    assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    // default value of error cause
    assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
    // different destination attempt reports error. now threshold crossed
    TezTaskAttemptID mockDestId2 = mock(TezTaskAttemptID.class);
    when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId2);
    taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 11));
    assertEquals("Task attempt is not in FAILED state", taImpl.getState(), TaskAttemptState.FAILED);
    assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl.getTerminationCause());
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    verify(mockHistHandler, times(3)).handle(histArg.capture());
    histEvent = histArg.getValue();
    finishEvent = (TaskAttemptFinishedEvent) histEvent.getHistoryEvent();
    assertEquals(TaskFailureType.NON_FATAL, finishEvent.getTaskFailureType());
    long newFinishTime = finishEvent.getFinishTime();
    Assert.assertEquals(finishTime, newFinishTime);
    assertEquals(true, taImpl.inputFailedReported);
    int expectedEventsAfterFetchFailure = expectedEventsTillSucceeded + 2;
    arg.getAllValues().clear();
    verify(eventHandler, times(expectedEventsAfterFetchFailure)).handle(arg.capture());
    Event e = verifyEventType(arg.getAllValues().subList(expectedEventsTillSucceeded, expectedEventsAfterFetchFailure), TaskEventTAFailed.class, 1);
    TaskEventTAFailed failedEvent = (TaskEventTAFailed) e;
    assertEquals(TaskFailureType.NON_FATAL, failedEvent.getTaskFailureType());
    taImpl.handle(new TaskAttemptEventOutputFailed(taskAttemptID, tzEvent, 1));
    assertEquals("Task attempt is not in FAILED state, still", taImpl.getState(), TaskAttemptState.FAILED);
    assertFalse("InternalError occurred trying to handle TA_TOO_MANY_FETCH_FAILURES", eventHandler.internalError);
    // No new events.
    verify(eventHandler, times(expectedEventsAfterFetchFailure)).handle(arg.capture());
    taskConf.setInt(TezConfiguration.TEZ_TASK_MAX_ALLOWED_OUTPUT_FAILURES, 1);
    TezTaskID taskID2 = TezTaskID.getInstance(vertexID, 2);
    MockTaskAttemptImpl taImpl2 = new MockTaskAttemptImpl(taskID2, 1, eventHandler, taListener, taskConf, new SystemClock(), mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
    TezTaskAttemptID taskAttemptID2 = taImpl2.getID();
    taImpl2.handle(new TaskAttemptEventSchedule(taskAttemptID2, 0, 0));
    taImpl2.handle(new TaskAttemptEventSubmitted(taskAttemptID2, contId));
    taImpl2.handle(new TaskAttemptEventStartedRemotely(taskAttemptID2));
    verify(mockHeartbeatHandler).register(taskAttemptID2);
    taImpl2.handle(new TaskAttemptEvent(taskAttemptID2, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in succeeded state", taImpl2.getState(), TaskAttemptState.SUCCEEDED);
    verify(mockHeartbeatHandler).unregister(taskAttemptID2);
    mockReEvent = InputReadErrorEvent.create("", 1, 1);
    mockMeta = mock(EventMetaData.class);
    mockDestId1 = mock(TezTaskAttemptID.class);
    when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
    tzEvent = new TezEvent(mockReEvent, mockMeta);
    // This should fail even when MAX_ALLOWED_OUTPUT_FAILURES_FRACTION is within limits, as
    // MAX_ALLOWED_OUTPUT_FAILURES has crossed the limit.
    taImpl2.handle(new TaskAttemptEventOutputFailed(taskAttemptID2, tzEvent, 8));
    assertEquals("Task attempt is not in failed state", taImpl2.getState(), TaskAttemptState.FAILED);
    assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl2.getTerminationCause());
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID2);
    Clock mockClock = mock(Clock.class);
    int readErrorTimespanSec = 1;
    taskConf.setInt(TezConfiguration.TEZ_TASK_MAX_ALLOWED_OUTPUT_FAILURES, 10);
    taskConf.setInt(TezConfiguration.TEZ_AM_MAX_ALLOWED_TIME_FOR_TASK_READ_ERROR_SEC, readErrorTimespanSec);
    TezTaskID taskID3 = TezTaskID.getInstance(vertexID, 3);
    MockTaskAttemptImpl taImpl3 = new MockTaskAttemptImpl(taskID3, 1, eventHandler, taListener, taskConf, mockClock, mockHeartbeatHandler, appCtx, false, resource, createFakeContainerContext(), false);
    TezTaskAttemptID taskAttemptID3 = taImpl3.getID();
    taImpl3.handle(new TaskAttemptEventSchedule(taskAttemptID3, 0, 0));
    taImpl3.handle(new TaskAttemptEventSubmitted(taskAttemptID3, contId));
    taImpl3.handle(new TaskAttemptEventStartedRemotely(taskAttemptID3));
    verify(mockHeartbeatHandler).register(taskAttemptID3);
    taImpl3.handle(new TaskAttemptEvent(taskAttemptID3, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
    verify(mockHeartbeatHandler).unregister(taskAttemptID3);
    mockReEvent = InputReadErrorEvent.create("", 1, 1);
    mockMeta = mock(EventMetaData.class);
    mockDestId1 = mock(TezTaskAttemptID.class);
    when(mockMeta.getTaskAttemptID()).thenReturn(mockDestId1);
    tzEvent = new TezEvent(mockReEvent, mockMeta);
    when(mockClock.getTime()).thenReturn(1000L);
    // time deadline not exceeded for a couple of read error events
    taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
    assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
    when(mockClock.getTime()).thenReturn(1500L);
    taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
    assertEquals("Task attempt is not in succeeded state", taImpl3.getState(), TaskAttemptState.SUCCEEDED);
    // exceed the time threshold
    when(mockClock.getTime()).thenReturn(2001L);
    taImpl3.handle(new TaskAttemptEventOutputFailed(taskAttemptID3, tzEvent, 1000));
    assertEquals("Task attempt is not in FAILED state", taImpl3.getState(), TaskAttemptState.FAILED);
    assertEquals(TaskAttemptTerminationCause.OUTPUT_LOST, taImpl3.getTerminationCause());
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID3);
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) Clock(org.apache.hadoop.yarn.util.Clock) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Container(org.apache.hadoop.yarn.api.records.Container) HistoryEventHandler(org.apache.tez.dag.history.HistoryEventHandler) TaskAttemptEventOutputFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventOutputFailed) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TezDAGID(org.apache.tez.dag.records.TezDAGID) TaskHeartbeatHandler(org.apache.tez.dag.app.TaskHeartbeatHandler) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TaskAttemptFinishedEvent(org.apache.tez.dag.history.events.TaskAttemptFinishedEvent) TezVertexID(org.apache.tez.dag.records.TezVertexID) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) EventMetaData(org.apache.tez.runtime.api.impl.EventMetaData) HashSet(java.util.HashSet) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskEventTAFailed(org.apache.tez.dag.app.dag.event.TaskEventTAFailed) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) NodeId(org.apache.hadoop.yarn.api.records.NodeId) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) TaskAttemptFinishedEvent(org.apache.tez.dag.history.events.TaskAttemptFinishedEvent) TaskEvent(org.apache.tez.dag.app.dag.event.TaskEvent) DAGEvent(org.apache.tez.dag.app.dag.event.DAGEvent) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) Event(org.apache.hadoop.yarn.event.Event) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Example 4 with TaskEventTAFailed

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

the class TestTaskAttempt method testFailureFatalError.

@Test(timeout = 5000)
public void testFailureFatalError() 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.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.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);
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) TaskAttemptEventStatusUpdate(org.apache.tez.dag.app.dag.event.TaskAttemptEventStatusUpdate) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) AMContainerMap(org.apache.tez.dag.app.rm.container.AMContainerMap) Container(org.apache.hadoop.yarn.api.records.Container) TaskAttemptEventTezEventUpdate(org.apache.tez.dag.app.dag.event.TaskAttemptEventTezEventUpdate) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TezDAGID(org.apache.tez.dag.records.TezDAGID) TaskHeartbeatHandler(org.apache.tez.dag.app.TaskHeartbeatHandler) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TezVertexID(org.apache.tez.dag.records.TezVertexID) ContainerHeartbeatHandler(org.apache.tez.dag.app.ContainerHeartbeatHandler) HashSet(java.util.HashSet) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) SystemClock(org.apache.hadoop.yarn.util.SystemClock) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) TaskEventTAFailed(org.apache.tez.dag.app.dag.event.TaskEventTAFailed) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) TaskCommunicatorManagerInterface(org.apache.tez.dag.app.TaskCommunicatorManagerInterface) ContainerContextMatcher(org.apache.tez.dag.app.rm.container.ContainerContextMatcher) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) NodeId(org.apache.hadoop.yarn.api.records.NodeId) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) TaskAttemptFinishedEvent(org.apache.tez.dag.history.events.TaskAttemptFinishedEvent) TaskEvent(org.apache.tez.dag.app.dag.event.TaskEvent) DAGEvent(org.apache.tez.dag.app.dag.event.DAGEvent) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) Event(org.apache.hadoop.yarn.event.Event) DAGHistoryEvent(org.apache.tez.dag.history.DAGHistoryEvent) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) TaskAttemptEventContainerTerminated(org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminated) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) TaskAttemptEventAttemptFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed) Test(org.junit.Test)

Example 5 with TaskEventTAFailed

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

the class TestTaskImpl method testSucceededLeafTaskWithRetroFailures.

@SuppressWarnings("rawtypes")
@Test(timeout = 10000L)
public void testSucceededLeafTaskWithRetroFailures() 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, true, taskResource, containerContext, vertex);
    TezTaskID taskId = getNewTaskID();
    scheduleTaskAttempt(taskId);
    MockTaskAttemptImpl firstMockTaskAttempt = mockTask.getAttemptList().get(0);
    launchTaskAttempt(firstMockTaskAttempt.getID());
    mockTask.handle(createTaskTAAddSpecAttempt(mockTask.getLastAttempt().getID()));
    MockTaskAttemptImpl secondMockTaskAttempt = mockTask.getAttemptList().get(1);
    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 TaskAttemptEvent(TezTaskAttemptID.fromString(secondMockTaskAttempt.toString()), TaskAttemptEventType.TA_DONE));
    firstMockTaskAttempt.handle(new TaskAttemptEventAttemptFailed(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "test", TaskAttemptTerminationCause.CONTAINER_EXITED));
    mockTask.handle(new TaskEventTASucceeded(secondMockTaskAttempt.getID()));
    firstMockTaskAttempt.handle(new TaskAttemptEventContainerTerminated(mockContainerId, firstMockTaskAttempt.getID(), "test", TaskAttemptTerminationCause.NO_PROGRESS));
    InputReadErrorEvent mockReEvent = InputReadErrorEvent.create("", 0, 0);
    TezTaskAttemptID mockDestId = firstMockTaskAttempt.getID();
    EventMetaData meta = new EventMetaData(EventProducerConsumerType.INPUT, "Vertex", "Edge", mockDestId);
    TezEvent tzEvent = new TezEvent(mockReEvent, meta);
    TaskAttemptEventOutputFailed outputFailedEvent = new TaskAttemptEventOutputFailed(mockDestId, tzEvent, 1);
    firstMockTaskAttempt.handle(outputFailedEvent);
    mockTask.handle(new TaskEventTAFailed(firstMockTaskAttempt.getID(), TaskFailureType.NON_FATAL, mock(TaskAttemptEvent.class)));
    Assert.assertEquals(mockTask.getInternalState(), TaskStateInternal.SUCCEEDED);
}
Also used : Vertex(org.apache.tez.dag.app.dag.Vertex) Configuration(org.apache.hadoop.conf.Configuration) TezConfiguration(org.apache.tez.dag.api.TezConfiguration) TaskEventTAFailed(org.apache.tez.dag.app.dag.event.TaskEventTAFailed) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) InputReadErrorEvent(org.apache.tez.runtime.api.events.InputReadErrorEvent) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TaskEventTASucceeded(org.apache.tez.dag.app.dag.event.TaskEventTASucceeded) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskAttemptEventOutputFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventOutputFailed) TaskAttemptEventContainerTerminated(org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminated) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) EventMetaData(org.apache.tez.runtime.api.impl.EventMetaData) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) TaskAttemptEventAttemptFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Aggregations

TaskEventTAFailed (org.apache.tez.dag.app.dag.event.TaskEventTAFailed)11 TezTaskID (org.apache.tez.dag.records.TezTaskID)10 Test (org.junit.Test)10 TezTaskAttemptID (org.apache.tez.dag.records.TezTaskAttemptID)9 Configuration (org.apache.hadoop.conf.Configuration)8 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)8 TaskAttemptEventSchedule (org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule)8 TaskAttemptEventStartedRemotely (org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely)8 TaskAttemptEventSubmitted (org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted)8 TaskLocationHint (org.apache.tez.dag.api.TaskLocationHint)7 TaskAttemptEventContainerTerminated (org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminated)7 InputReadErrorEvent (org.apache.tez.runtime.api.events.InputReadErrorEvent)7 TezEvent (org.apache.tez.runtime.api.impl.TezEvent)7 TaskAttemptEvent (org.apache.tez.dag.app.dag.event.TaskAttemptEvent)6 TaskEvent (org.apache.tez.dag.app.dag.event.TaskEvent)6 HashSet (java.util.HashSet)5 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)5 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)5 Container (org.apache.hadoop.yarn.api.records.Container)5 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)5