Examples with AMContainerMap org.apache.tez.dag.app.rm.container.AMContainerMap used on opensource projects

Example 21 with AMContainerMap

use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.

the class DAGAppMaster method serviceInit.

@Override
protected void serviceInit(final Configuration conf) throws Exception {
    this.amConf = conf;
    initResourceCalculatorPlugins();
    this.hadoopShim = new HadoopShimsLoader(this.amConf).getHadoopShim();
    long sleepTimeBeforeSecs = this.amConf.getLong(TezConfiguration.TEZ_AM_SLEEP_TIME_BEFORE_EXIT_MILLIS, TezConstants.TEZ_DAG_SLEEP_TIME_BEFORE_EXIT);
    if (sleepTimeBeforeSecs >= 0) {
        this.shutdownHandler.setSleepTimeBeforeExit(sleepTimeBeforeSecs);
    }
    this.isLocal = conf.getBoolean(TezConfiguration.TEZ_LOCAL_MODE, TezConfiguration.TEZ_LOCAL_MODE_DEFAULT);
    UserPayload defaultPayload = TezUtils.createUserPayloadFromConf(amConf);
    List<NamedEntityDescriptor> taskSchedulerDescriptors = Lists.newLinkedList();
    List<NamedEntityDescriptor> containerLauncherDescriptors = Lists.newLinkedList();
    List<NamedEntityDescriptor> taskCommunicatorDescriptors = Lists.newLinkedList();
    parseAllPlugins(taskSchedulerDescriptors, taskSchedulers, containerLauncherDescriptors, containerLaunchers, taskCommunicatorDescriptors, taskCommunicators, amPluginDescriptorProto, isLocal, defaultPayload);
    LOG.info(buildPluginComponentLog(taskSchedulerDescriptors, taskSchedulers, "TaskSchedulers"));
    LOG.info(buildPluginComponentLog(containerLauncherDescriptors, containerLaunchers, "ContainerLaunchers"));
    LOG.info(buildPluginComponentLog(taskCommunicatorDescriptors, taskCommunicators, "TaskCommunicators"));
    boolean disableVersionCheck = conf.getBoolean(TezConfiguration.TEZ_AM_DISABLE_CLIENT_VERSION_CHECK, TezConfiguration.TEZ_AM_DISABLE_CLIENT_VERSION_CHECK_DEFAULT);
    // Check client - AM version compatibility
    LOG.info("Comparing client version with AM version" + ", clientVersion=" + clientVersion + ", AMVersion=" + dagVersionInfo.getVersion());
    Simple2LevelVersionComparator versionComparator = new Simple2LevelVersionComparator();
    if (versionComparator.compare(clientVersion, dagVersionInfo.getVersion()) != 0) {
        versionMismatchDiagnostics = "Incompatible versions found" + ", clientVersion=" + clientVersion + ", AMVersion=" + dagVersionInfo.getVersion();
        addDiagnostic(versionMismatchDiagnostics);
        if (disableVersionCheck) {
            LOG.warn("Ignoring client-AM version mismatch as check disabled. " + versionMismatchDiagnostics);
        } else {
            LOG.error(versionMismatchDiagnostics);
            versionMismatch = true;
        }
    }
    dispatcher = createDispatcher();
    if (isLocal) {
        conf.setBoolean(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_ENABLED, false);
        conf.set(TezConfiguration.TEZ_HISTORY_LOGGING_SERVICE_CLASS, TezConfiguration.TEZ_HISTORY_LOGGING_SERVICE_CLASS_DEFAULT);
    } else {
        dispatcher.enableExitOnDispatchException();
    }
    String strAppId = this.appAttemptID.getApplicationId().toString();
    this.tezSystemStagingDir = TezCommonUtils.getTezSystemStagingPath(conf, strAppId);
    context = new RunningAppContext(conf);
    this.aclManager = new ACLManager(appMasterUgi.getShortUserName(), this.amConf);
    clientHandler = new DAGClientHandler(this);
    addIfService(dispatcher, false);
    recoveryDataDir = TezCommonUtils.getRecoveryPath(tezSystemStagingDir, conf);
    recoveryFS = recoveryDataDir.getFileSystem(conf);
    currentRecoveryDataDir = TezCommonUtils.getAttemptRecoveryPath(recoveryDataDir, appAttemptID.getAttemptId());
    if (LOG.isDebugEnabled()) {
        LOG.debug("Stage directory information for AppAttemptId :" + this.appAttemptID + " tezSystemStagingDir :" + tezSystemStagingDir + " recoveryDataDir :" + recoveryDataDir + " recoveryAttemptDir :" + currentRecoveryDataDir);
    }
    recoveryEnabled = conf.getBoolean(TezConfiguration.DAG_RECOVERY_ENABLED, TezConfiguration.DAG_RECOVERY_ENABLED_DEFAULT);
    initClientRpcServer();
    taskHeartbeatHandler = createTaskHeartbeatHandler(context, conf);
    addIfService(taskHeartbeatHandler, true);
    containerHeartbeatHandler = createContainerHeartbeatHandler(context, conf);
    addIfService(containerHeartbeatHandler, true);
    sessionToken = TokenCache.getSessionToken(amCredentials);
    if (sessionToken == null) {
        throw new RuntimeException("Could not find session token in AM Credentials");
    }
    // Prepare the TaskAttemptListener server for authentication of Containers
    // TaskAttemptListener gets the information via jobTokenSecretManager.
    jobTokenSecretManager.addTokenForJob(appAttemptID.getApplicationId().toString(), sessionToken);
    // service to handle requests to TaskUmbilicalProtocol
    taskCommunicatorManager = createTaskCommunicatorManager(context, taskHeartbeatHandler, containerHeartbeatHandler, taskCommunicatorDescriptors);
    addIfService(taskCommunicatorManager, true);
    containerSignatureMatcher = createContainerSignatureMatcher();
    containers = new AMContainerMap(containerHeartbeatHandler, taskCommunicatorManager, containerSignatureMatcher, context);
    addIfService(containers, true);
    dispatcher.register(AMContainerEventType.class, containers);
    nodes = new AMNodeTracker(dispatcher.getEventHandler(), context);
    addIfService(nodes, true);
    dispatcher.register(AMNodeEventType.class, nodes);
    this.dagEventDispatcher = new DagEventDispatcher();
    this.vertexEventDispatcher = new VertexEventDispatcher();
    // register the event dispatchers
    dispatcher.register(DAGAppMasterEventType.class, new DAGAppMasterEventHandler());
    dispatcher.register(DAGEventType.class, dagEventDispatcher);
    dispatcher.register(VertexEventType.class, vertexEventDispatcher);
    boolean useConcurrentDispatcher = conf.getBoolean(TezConfiguration.TEZ_AM_USE_CONCURRENT_DISPATCHER, TezConfiguration.TEZ_AM_USE_CONCURRENT_DISPATCHER_DEFAULT);
    LOG.info("Using concurrent dispatcher: " + useConcurrentDispatcher);
    if (!useConcurrentDispatcher) {
        dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
        dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
    } else {
        int concurrency = conf.getInt(TezConfiguration.TEZ_AM_CONCURRENT_DISPATCHER_CONCURRENCY, TezConfiguration.TEZ_AM_CONCURRENT_DISPATCHER_CONCURRENCY_DEFAULT);
        AsyncDispatcherConcurrent sharedDispatcher = dispatcher.registerAndCreateDispatcher(TaskEventType.class, new TaskEventDispatcher(), "TaskAndAttemptEventThread", concurrency);
        dispatcher.registerWithExistingDispatcher(TaskAttemptEventType.class, new TaskAttemptEventDispatcher(), sharedDispatcher);
    }
    // register other delegating dispatchers
    dispatcher.registerAndCreateDispatcher(SpeculatorEventType.class, new SpeculatorEventHandler(), "Speculator");
    if (enableWebUIService()) {
        this.webUIService = new WebUIService(context);
        addIfService(webUIService, false);
    } else {
        LOG.debug("Web UI Service is not enabled.");
    }
    this.taskSchedulerManager = createTaskSchedulerManager(taskSchedulerDescriptors);
    addIfService(taskSchedulerManager, true);
    if (enableWebUIService()) {
        addIfServiceDependency(taskSchedulerManager, webUIService);
    }
    dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
    addIfServiceDependency(taskSchedulerManager, clientRpcServer);
    this.containerLauncherManager = createContainerLauncherManager(containerLauncherDescriptors, isLocal);
    addIfService(containerLauncherManager, true);
    dispatcher.register(ContainerLauncherEventType.class, containerLauncherManager);
    historyEventHandler = createHistoryEventHandler(context);
    addIfService(historyEventHandler, true);
    this.sessionTimeoutInterval = TezCommonUtils.getDAGSessionTimeout(amConf);
    this.clientAMHeartbeatTimeoutIntervalMillis = TezCommonUtils.getAMClientHeartBeatTimeoutMillis(amConf);
    if (!versionMismatch) {
        if (isSession) {
            try (BufferedInputStream sessionResourcesStream = new BufferedInputStream(new FileInputStream(new File(workingDirectory, TezConstants.TEZ_AM_LOCAL_RESOURCES_PB_FILE_NAME)))) {
                PlanLocalResourcesProto amLocalResourceProto = PlanLocalResourcesProto.parseDelimitedFrom(sessionResourcesStream);
                amResources.putAll(DagTypeConverters.convertFromPlanLocalResources(amLocalResourceProto));
            }
        }
    }
    int threadCount = conf.getInt(TezConfiguration.TEZ_AM_DAG_APPCONTEXT_THREAD_COUNT_LIMIT, TezConfiguration.TEZ_AM_DAG_APPCONTEXT_THREAD_COUNT_LIMIT_DEFAULT);
    // NOTE: LinkedBlockingQueue does not have a capacity Limit and can thus
    // occupy large memory chunks when numerous Runables are pending for execution
    ExecutorService rawExecutor = Executors.newFixedThreadPool(threadCount, new ThreadFactoryBuilder().setDaemon(true).setNameFormat("App Shared Pool - #%d").build());
    execService = MoreExecutors.listeningDecorator(rawExecutor);
    initServices(conf);
    super.serviceInit(conf);
    if (!versionMismatch) {
        if (this.appAttemptID.getAttemptId() == 1) {
            AppLaunchedEvent appLaunchedEvent = new AppLaunchedEvent(appAttemptID.getApplicationId(), startTime, appSubmitTime, appMasterUgi.getShortUserName(), this.amConf, dagVersionInfo);
            historyEventHandler.handle(new DAGHistoryEvent(appLaunchedEvent));
        }
        AMLaunchedEvent launchedEvent = new AMLaunchedEvent(appAttemptID, startTime, appSubmitTime, appMasterUgi.getShortUserName());
        historyEventHandler.handle(new DAGHistoryEvent(launchedEvent));
        this.state = DAGAppMasterState.INITED;
    } else {
        this.state = DAGAppMasterState.ERROR;
    }
}

Example 22 with AMContainerMap

use of org.apache.tez.dag.app.rm.container.AMContainerMap 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.getTaskAttemptID();
    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());
}

Example 23 with AMContainerMap

use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.

the class TestTaskAttempt method testStatusUpdateWithNullCounters.

@Test
public void testStatusUpdateWithNullCounters() 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);
    ContainerId contId = ContainerId.newContainerId(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.getTaskAttemptID();
    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);
    TezCounters counters = new TezCounters();
    counters.findCounter("group", "counter").increment(1);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(counters, 0.1f, null, false)));
    assertEquals(1, taImpl.getCounters().findCounter("group", "counter").getValue());
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    assertEquals(1, taImpl.getCounters().findCounter("group", "counter").getValue());
    counters.findCounter("group", "counter").increment(1);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(counters, 0.1f, null, false)));
    assertEquals(2, taImpl.getCounters().findCounter("group", "counter").getValue());
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    assertEquals(2, taImpl.getCounters().findCounter("group", "counter").getValue());
}

Example 24 with AMContainerMap

use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.

the class TestTaskAttempt method testContainerPreemptedAfterSuccess.

@Test(timeout = 5000)
public // the TaskAttempt
void testContainerPreemptedAfterSuccess() 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.getTaskAttemptID();
    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 = 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", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    verify(mockHeartbeatHandler).unregister(taskAttemptID);
    assertEquals(0, taImpl.getDiagnostics().size());
    int expectedEventsAfterTerminating = expectedEventsAtRunning + 3;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEventsAfterTerminating)).handle(arg.capture());
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), TaskEventTASucceeded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), AMSchedulerEventTAEnded.class, 1);
    verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEventsAfterTerminating), DAGEventCounterUpdate.class, 1);
    taImpl.handle(new TaskAttemptEvent(taskAttemptID, TaskAttemptEventType.TA_CONTAINER_TERMINATED_BY_SYSTEM));
    // verify unregister is not invoked again
    verify(mockHeartbeatHandler, times(1)).unregister(taskAttemptID);
    int expectedEventAfterTerminated = expectedEventsAfterTerminating + 0;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEventAfterTerminated)).handle(arg.capture());
    // Verify that the diagnostic message included in the Terminated event is not
    // captured - TA already succeeded. Error cause should be the default value
    assertEquals(0, taImpl.getDiagnostics().size());
    assertEquals(TaskAttemptTerminationCause.UNKNOWN_ERROR, taImpl.getTerminationCause());
}

Example 25 with AMContainerMap

use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.

the class TestVertexImpl method testVertexTaskAttemptProcessorFailure.

@SuppressWarnings("deprecation")
@Test(timeout = 5000)
public void testVertexTaskAttemptProcessorFailure() throws Exception {
    initAllVertices(VertexState.INITED);
    VertexImpl v = vertices.get("vertex1");
    startVertex(v);
    dispatcher.await();
    TaskAttemptImpl ta = (TaskAttemptImpl) v.getTask(0).getAttempts().values().iterator().next();
    ta.handle(new TaskAttemptEventSchedule(ta.getTaskAttemptID(), 2, 2));
    NodeId nid = NodeId.newInstance("127.0.0.1", 0);
    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(), appContext);
    containers.addContainerIfNew(container, 0, 0, 0);
    doReturn(containers).when(appContext).getAllContainers();
    ta.handle(new TaskAttemptEventSubmitted(ta.getTaskAttemptID(), contId));
    ta.handle(new TaskAttemptEventStartedRemotely(ta.getTaskAttemptID()));
    Assert.assertEquals(TaskAttemptStateInternal.RUNNING, ta.getInternalState());
    ta.handle(new TaskAttemptEventAttemptFailed(ta.getTaskAttemptID(), TaskAttemptEventType.TA_FAILED, TaskFailureType.NON_FATAL, "diag", TaskAttemptTerminationCause.APPLICATION_ERROR));
    dispatcher.await();
    Assert.assertEquals(VertexState.RUNNING, v.getState());
    Assert.assertEquals(TaskAttemptTerminationCause.APPLICATION_ERROR, ta.getTerminationCause());
}