Search in sources :

Example 21 with TaskAttemptEventStartedRemotely

use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely 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());
}
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) 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) TaskAttemptEventNodeFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventNodeFailed) 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) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Example 22 with TaskAttemptEventStartedRemotely

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

the class TestTaskImpl method testFailedAttemptStatus.

@SuppressWarnings("rawtypes")
@Test
public void testFailedAttemptStatus() 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.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 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);
}
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) TaskAttemptEventSubmitted(org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted) TezTaskID(org.apache.tez.dag.records.TezTaskID) TaskAttemptEventContainerTerminated(org.apache.tez.dag.app.dag.event.TaskAttemptEventContainerTerminated) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) TaskAttemptEventAttemptFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed) Test(org.junit.Test)

Example 23 with TaskAttemptEventStartedRemotely

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

the class TestTaskImpl method testSucceededAttemptStatusWithRetroActiveFailures.

@SuppressWarnings("rawtypes")
@Test
public void testSucceededAttemptStatusWithRetroActiveFailures() throws InterruptedException {
    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 TaskAttemptEvent(TezTaskAttemptID.fromString(firstMockTaskAttempt.toString()), TaskAttemptEventType.TA_DONE));
    mockTask.handle(new TaskEventTASucceeded(secondMockTaskAttempt.getID()));
    mockTask.handle(new TaskEventTASucceeded(firstMockTaskAttempt.getID()));
    assertTrue("Attempts should have succeeded!", firstMockTaskAttempt.getInternalState() == TaskAttemptStateInternal.SUCCEEDED && secondMockTaskAttempt.getInternalState() == TaskAttemptStateInternal.SUCCEEDED);
    assertEquals("Task should have no uncompleted attempts!", 0, mockTask.getUncompletedAttemptsCount());
    assertTrue("Task should have Succeeded!", mockTask.getState() == TaskState.SUCCEEDED);
    // Failing the attempt that finished after the task was marked succeeded, should not schedule another attempt
    failAttempt(firstMockTaskAttempt, 0, 0);
    assertTaskSucceededState();
    // Failing the attempt that allowed the task to succeed, should schedule another attempt
    failAttempt(secondMockTaskAttempt, 1, 1);
    assertTaskScheduledState();
}
Also used : TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) TaskAttemptEventSchedule(org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule) 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) TaskAttemptEventStartedRemotely(org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely) Test(org.junit.Test)

Aggregations

TaskAttemptEventSchedule (org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule)23 TaskAttemptEventStartedRemotely (org.apache.tez.dag.app.dag.event.TaskAttemptEventStartedRemotely)23 TaskAttemptEventSubmitted (org.apache.tez.dag.app.dag.event.TaskAttemptEventSubmitted)23 Test (org.junit.Test)23 TezTaskID (org.apache.tez.dag.records.TezTaskID)20 Container (org.apache.hadoop.yarn.api.records.Container)18 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)18 NodeId (org.apache.hadoop.yarn.api.records.NodeId)18 ContainerHeartbeatHandler (org.apache.tez.dag.app.ContainerHeartbeatHandler)18 TaskCommunicatorManagerInterface (org.apache.tez.dag.app.TaskCommunicatorManagerInterface)18 AMContainerMap (org.apache.tez.dag.app.rm.container.AMContainerMap)18 ContainerContextMatcher (org.apache.tez.dag.app.rm.container.ContainerContextMatcher)18 Configuration (org.apache.hadoop.conf.Configuration)17 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)17 TaskAttemptEvent (org.apache.tez.dag.app.dag.event.TaskAttemptEvent)16 TezTaskAttemptID (org.apache.tez.dag.records.TezTaskAttemptID)16 HashSet (java.util.HashSet)14 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)14 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)14 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)14