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

Example 1 with TaskAttemptEventNodeFailed

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

the class TestAMContainer method testNodeFailedAtRunningMultipleAttempts.

@SuppressWarnings("rawtypes")
@Test(timeout = 5000)
public void testNodeFailedAtRunningMultipleAttempts() {
    WrappedContainer wc = new WrappedContainer();
    List<Event> outgoingEvents;
    wc.launchContainer();
    wc.containerLaunched();
    wc.assignTaskAttempt(wc.taskAttemptID);
    wc.taskAttemptSucceeded(wc.taskAttemptID);
    TezTaskAttemptID taID2 = TezTaskAttemptID.getInstance(wc.taskID, 2);
    wc.assignTaskAttempt(taID2);
    wc.verifyState(AMContainerState.RUNNING);
    wc.nodeFailed();
    // Expecting a complete event from the RM
    wc.verifyState(AMContainerState.STOPPING);
    outgoingEvents = wc.verifyCountAndGetOutgoingEvents(5);
    verifyUnOrderedOutgoingEventTypes(outgoingEvents, TaskAttemptEventType.TA_NODE_FAILED, TaskAttemptEventType.TA_NODE_FAILED, TaskAttemptEventType.TA_CONTAINER_TERMINATING, AMSchedulerEventType.S_CONTAINER_DEALLOCATE, AMNodeEventType.N_CONTAINER_COMPLETED);
    for (Event event : outgoingEvents) {
        if (event.getType() == TaskAttemptEventType.TA_NODE_FAILED) {
            TaskAttemptEventNodeFailed nfEvent = (TaskAttemptEventNodeFailed) event;
            assertTrue(nfEvent.getDiagnosticInfo().contains("nodeFailed"));
        }
    }
    wc.containerCompleted();
    wc.verifyHistoryStopEvent();
    outgoingEvents = wc.verifyCountAndGetOutgoingEvents(1);
    verifyUnOrderedOutgoingEventTypes(outgoingEvents, TaskAttemptEventType.TA_CONTAINER_TERMINATED);
    assertFalse(wc.amContainer.isInErrorState());
    assertNull(wc.amContainer.getCurrentTaskAttempt());
    assertEquals(2, wc.amContainer.getAllTaskAttempts().size());
}

Example 2 with TaskAttemptEventNodeFailed

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

the class TestAMContainer method testNodeFailedAtRunning.

@SuppressWarnings("rawtypes")
@Test(timeout = 5000)
public void testNodeFailedAtRunning() {
    WrappedContainer wc = new WrappedContainer();
    List<Event> outgoingEvents;
    wc.launchContainer();
    wc.containerLaunched();
    wc.assignTaskAttempt(wc.taskAttemptID);
    wc.verifyState(AMContainerState.RUNNING);
    wc.nodeFailed();
    // Expecting a complete event from the RM
    wc.verifyState(AMContainerState.STOPPING);
    outgoingEvents = wc.verifyCountAndGetOutgoingEvents(4);
    verifyUnOrderedOutgoingEventTypes(outgoingEvents, TaskAttemptEventType.TA_NODE_FAILED, TaskAttemptEventType.TA_CONTAINER_TERMINATING, AMSchedulerEventType.S_CONTAINER_DEALLOCATE, AMNodeEventType.N_CONTAINER_COMPLETED);
    for (Event event : outgoingEvents) {
        if (event.getType() == TaskAttemptEventType.TA_NODE_FAILED) {
            TaskAttemptEventNodeFailed nfEvent = (TaskAttemptEventNodeFailed) event;
            assertTrue(nfEvent.getDiagnosticInfo().contains("nodeFailed"));
        }
    }
    wc.containerCompleted();
    wc.verifyHistoryStopEvent();
    outgoingEvents = wc.verifyCountAndGetOutgoingEvents(1);
    verifyUnOrderedOutgoingEventTypes(outgoingEvents, TaskAttemptEventType.TA_CONTAINER_TERMINATED);
    assertFalse(wc.amContainer.isInErrorState());
}

Example 3 with TaskAttemptEventNodeFailed

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

the class TestTaskAttempt method testNodeFailedNonLeafVertex.

@Test(timeout = 5000)
public // Ensure node failure on Successful Non-Leaf tasks cause them to be marked as KILLED
void testNodeFailedNonLeafVertex() 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);
    MockTaskAttemptImpl 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", TaskAttemptState.RUNNING, taImpl.getState());
    verify(mockHeartbeatHandler).register(taskAttemptID);
    int expectedEventsAtRunning = 5;
    verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
    taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in the  SUCCEEDED state", TaskAttemptState.SUCCEEDED, taImpl.getState());
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals(0, taImpl.getDiagnostics().size());
    int expectedEvenstAfterTerminating = expectedEventsAtRunning + 3;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTASucceeded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
    // Send out a Node Failure.
    taImpl.handle(new TaskAttemptEventNodeFailed(taskAttemptID, "NodeDecomissioned", TaskAttemptTerminationCause.NODE_FAILED));
    // Verify in KILLED state
    assertEquals("Task attempt is not in the  KILLED state", TaskAttemptState.KILLED, taImpl.getState());
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    assertEquals(true, taImpl.inputFailedReported);
    // Verify one event to the Task informing it about FAILURE. No events to scheduler. Counter event.
    int expectedEventsNodeFailure = expectedEvenstAfterTerminating + 2;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEventsNodeFailure)).handle(arg.capture());
    verifyEventType(arg.getAllValues().subList(expectedEvenstAfterTerminating, expectedEventsNodeFailure), TaskEventTAKilled.class, 1);
    // Verify still in KILLED state
    assertEquals("Task attempt is not in the  KILLED state", TaskAttemptState.KILLED, taImpl.getState());
    assertEquals(TaskAttemptTerminationCause.NODE_FAILED, taImpl.getTerminationCause());
    assertEquals(TaskAttemptStateInternal.KILLED, taImpl.getInternalState());
    taImpl.handle(new TaskAttemptEventTezEventUpdate(taImpl.getID(), Collections.EMPTY_LIST));
    assertFalse("InternalError occurred trying to handle TA_TEZ_EVENT_UPDATE in KILLED state", eventHandler.internalError);
}

Example 4 with TaskAttemptEventNodeFailed

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

the class TestTaskAttempt method testNodeFailedLeafVertex.

@Test(timeout = 5000)
public // Ensure node failure on Successful Leaf tasks do not cause them to be marked as KILLED
void testNodeFailedLeafVertex() 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(), true);
    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", TaskAttemptState.RUNNING, taImpl.getState());
    verify(mockHeartbeatHandler).register(taskAttemptID);
    int expectedEventsAtRunning = 5;
    verify(eventHandler, times(expectedEventsAtRunning)).handle(arg.capture());
    taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_DONE));
    assertEquals("Task attempt is not in the  SUCCEEDED state", TaskAttemptState.SUCCEEDED, taImpl.getState());
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals(0, taImpl.getDiagnostics().size());
    int expectedEvenstAfterTerminating = expectedEventsAtRunning + 3;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTASucceeded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), DAGEventCounterUpdate.class, 1);
    // Send out a Node Failure.
    taImpl.handle(new TaskAttemptEventNodeFailed(taskAttemptID, "NodeDecomissioned", TaskAttemptTerminationCause.NODE_FAILED));
    // Verify no additional events
    int expectedEventsNodeFailure = expectedEvenstAfterTerminating + 0;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEventsNodeFailure)).handle(arg.capture());
    // Verify still in SUCCEEDED state
    assertEquals("Task attempt is not in the  SUCCEEDED state", TaskAttemptState.SUCCEEDED, taImpl.getState());
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    // error cause remains as default value
    assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
}

Example 5 with TaskAttemptEventNodeFailed

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

the class TestAMContainer method testNodeFailedAtIdleMultipleAttempts.

@SuppressWarnings("rawtypes")
@Test(timeout = 5000)
public void testNodeFailedAtIdleMultipleAttempts() {
    WrappedContainer wc = new WrappedContainer();
    List<Event> outgoingEvents;
    wc.launchContainer();
    wc.containerLaunched();
    wc.assignTaskAttempt(wc.taskAttemptID);
    wc.taskAttemptSucceeded(wc.taskAttemptID);
    wc.verifyState(AMContainerState.IDLE);
    TezTaskAttemptID taID2 = TezTaskAttemptID.getInstance(wc.taskID, 2);
    wc.assignTaskAttempt(taID2);
    wc.taskAttemptSucceeded(taID2);
    wc.verifyState(AMContainerState.IDLE);
    wc.nodeFailed();
    // Expecting a complete event from the RM
    wc.verifyState(AMContainerState.STOPPING);
    outgoingEvents = wc.verifyCountAndGetOutgoingEvents(4);
    verifyUnOrderedOutgoingEventTypes(outgoingEvents, TaskAttemptEventType.TA_NODE_FAILED, TaskAttemptEventType.TA_NODE_FAILED, AMSchedulerEventType.S_CONTAINER_DEALLOCATE, AMNodeEventType.N_CONTAINER_COMPLETED);
    for (Event event : outgoingEvents) {
        if (event.getType() == TaskAttemptEventType.TA_NODE_FAILED) {
            TaskAttemptEventNodeFailed nfEvent = (TaskAttemptEventNodeFailed) event;
            assertTrue(nfEvent.getDiagnosticInfo().contains("nodeFailed"));
        }
    }
    assertFalse(wc.amContainer.isInErrorState());
    wc.containerCompleted();
    wc.verifyNoOutgoingEvents();
    wc.verifyHistoryStopEvent();
    assertNull(wc.amContainer.getCurrentTaskAttempt());
    assertEquals(2, wc.amContainer.getAllTaskAttempts().size());
}