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

Example 11 with ClusterInfo

use of org.apache.tez.dag.app.ClusterInfo in project tez by apache.

the class TestDAGImpl method testDAGInitFailedDuetoInvalidResource.

@Test(timeout = 5000)
public void testDAGInitFailedDuetoInvalidResource() {
    // cluster maxContainerCapability is less than the vertex resource request
    ClusterInfo clusterInfo = new ClusterInfo(Resource.newInstance(512, 10));
    doReturn(clusterInfo).when(appContext).getClusterInfo();
    dag.handle(new DAGEvent(dag.getID(), DAGEventType.DAG_INIT));
    dispatcher.await();
    Assert.assertEquals(DAGState.FAILED, dag.getState());
    Assert.assertEquals(DAGTerminationCause.INIT_FAILURE, dag.getTerminationCause());
    Assert.assertTrue(StringUtils.join(dag.getDiagnostics(), ",").contains("Vertex's TaskResource is beyond the cluster container capability"));
}

Example 12 with ClusterInfo

use of org.apache.tez.dag.app.ClusterInfo in project tez by apache.

the class TestCommit method setupDAG.

public void setupDAG(DAGPlan dagPlan) {
    conf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, false);
    appAttemptId = ApplicationAttemptId.newInstance(ApplicationId.newInstance(100, 1), 1);
    dagId = TezDAGID.getInstance(appAttemptId.getApplicationId(), 1);
    Assert.assertNotNull(dagId);
    dispatcher = new DrainDispatcher();
    fsTokens = new Credentials();
    appContext = mock(AppContext.class);
    when(appContext.getHadoopShim()).thenReturn(new DefaultHadoopShim());
    rawExecutor = Executors.newCachedThreadPool(new ThreadFactoryBuilder().setDaemon(true).setNameFormat("App Shared Pool - " + "#%d").build());
    execService = MoreExecutors.listeningDecorator(rawExecutor);
    doReturn(execService).when(appContext).getExecService();
    historyEventHandler = new MockHistoryEventHandler(appContext);
    aclManager = new ACLManager("amUser");
    doReturn(conf).when(appContext).getAMConf();
    doReturn(appAttemptId).when(appContext).getApplicationAttemptId();
    doReturn(appAttemptId.getApplicationId()).when(appContext).getApplicationID();
    doReturn(dagId).when(appContext).getCurrentDAGID();
    doReturn(historyEventHandler).when(appContext).getHistoryHandler();
    doReturn(aclManager).when(appContext).getAMACLManager();
    dag = new DAGImpl(dagId, conf, dagPlan, dispatcher.getEventHandler(), taskCommunicatorManagerInterface, fsTokens, clock, "user", thh, appContext);
    doReturn(dag).when(appContext).getCurrentDAG();
    doReturn(dispatcher.getEventHandler()).when(appContext).getEventHandler();
    ClusterInfo clusterInfo = new ClusterInfo(Resource.newInstance(8192, 10));
    doReturn(clusterInfo).when(appContext).getClusterInfo();
    dispatcher.register(CallableEventType.class, new CallableEventDispatcher());
    taskEventDispatcher = new TaskEventDispatcher();
    dispatcher.register(TaskEventType.class, taskEventDispatcher);
    taskAttemptEventDispatcher = new TaskAttemptEventDispatcher();
    dispatcher.register(TaskAttemptEventType.class, taskAttemptEventDispatcher);
    vertexEventDispatcher = new VertexEventDispatcher();
    dispatcher.register(VertexEventType.class, vertexEventDispatcher);
    dagEventDispatcher = new DagEventDispatcher();
    dispatcher.register(DAGEventType.class, dagEventDispatcher);
    dagFinishEventHandler = new DAGFinishEventHandler();
    dispatcher.register(DAGAppMasterEventType.class, dagFinishEventHandler);
    dispatcher.init(conf);
    dispatcher.start();
}

Example 13 with ClusterInfo

use of org.apache.tez.dag.app.ClusterInfo 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 14 with ClusterInfo

use of org.apache.tez.dag.app.ClusterInfo 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 15 with ClusterInfo

use of org.apache.tez.dag.app.ClusterInfo 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());
}