Search in sources :

Example 1 with TaskAttemptEventStatusUpdate

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

the class TaskCommunicatorManager method heartbeat.

public TaskHeartbeatResponse heartbeat(TaskHeartbeatRequest request) throws IOException, TezException {
    ContainerId containerId = ConverterUtils.toContainerId(request.getContainerIdentifier());
    if (LOG.isDebugEnabled()) {
        LOG.debug("Received heartbeat from container" + ", request=" + request);
    }
    if (!registeredContainers.containsKey(containerId)) {
        LOG.warn("Received task heartbeat from unknown container with id: " + containerId + ", asking it to die");
        return RESPONSE_SHOULD_DIE;
    }
    // A heartbeat can come in anytime. The AM may have made a decision to kill a running task/container
    // meanwhile. If the decision is processed through the pipeline before the heartbeat is processed,
    // the heartbeat will be dropped. Otherwise the heartbeat will be processed - and the system
    // know how to handle this - via FailedInputEvents for example (relevant only if the heartbeat has events).
    // So - avoiding synchronization.
    pingContainerHeartbeatHandler(containerId);
    TaskAttemptEventInfo eventInfo = new TaskAttemptEventInfo(0, null, 0);
    TezTaskAttemptID taskAttemptID = request.getTaskAttemptId();
    if (taskAttemptID != null) {
        ContainerId containerIdFromMap = registeredAttempts.get(taskAttemptID);
        if (containerIdFromMap == null || !containerIdFromMap.equals(containerId)) {
            // This can happen when a task heartbeats. Meanwhile the container is unregistered.
            // The information will eventually make it through to the plugin via a corresponding unregister.
            // There's a race in that case between the unregister making it through, and this method returning.
            // TODO TEZ-2003 (post) TEZ-2666. An exception back is likely a better approach than sending a shouldDie = true,
            // so that the plugin can handle the scenario. Alternately augment the response with error codes.
            // Error codes would be better than exceptions.
            LOG.info("Attempt: " + taskAttemptID + " is not recognized for heartbeats");
            return RESPONSE_SHOULD_DIE;
        }
        List<TezEvent> inEvents = request.getEvents();
        if (LOG.isDebugEnabled()) {
            LOG.debug("Ping from " + taskAttemptID.toString() + " events: " + (inEvents != null ? inEvents.size() : -1));
        }
        long currTime = context.getClock().getTime();
        // taFinishedEvents - means the TaskAttemptFinishedEvent
        // taGeneratedEvents - for recovery, means the events generated by this task attempt and is needed by its downstream vertices
        // eventsForVertex - including all the taGeneratedEvents and other events such as INPUT_READ_ERROR_EVENT/INPUT_FAILED_EVENT
        // taGeneratedEvents is routed both to TaskAttempt & Vertex. Route to Vertex is for performance consideration
        // taFinishedEvents must be routed before taGeneratedEvents
        List<TezEvent> taFinishedEvents = new ArrayList<TezEvent>();
        List<TezEvent> taGeneratedEvents = new ArrayList<TezEvent>();
        List<TezEvent> eventsForVertex = new ArrayList<TezEvent>();
        TaskAttemptEventStatusUpdate taskAttemptEvent = null;
        boolean readErrorReported = false;
        for (TezEvent tezEvent : ListUtils.emptyIfNull(inEvents)) {
            // for now, set the event time on the AM when it is received.
            // this avoids any time disparity between machines.
            tezEvent.setEventReceivedTime(currTime);
            final EventType eventType = tezEvent.getEventType();
            if (eventType == EventType.TASK_STATUS_UPDATE_EVENT) {
                // send TA_STATUS_UPDATE before TA_DONE/TA_FAILED/TA_KILLED otherwise Status may be missed
                taskAttemptEvent = new TaskAttemptEventStatusUpdate(taskAttemptID, (TaskStatusUpdateEvent) tezEvent.getEvent());
            } else if (eventType == EventType.TASK_ATTEMPT_COMPLETED_EVENT || eventType == EventType.TASK_ATTEMPT_FAILED_EVENT || eventType == EventType.TASK_ATTEMPT_KILLED_EVENT) {
                taFinishedEvents.add(tezEvent);
            } else {
                if (eventType == EventType.INPUT_READ_ERROR_EVENT) {
                    readErrorReported = true;
                }
                if (eventType == EventType.DATA_MOVEMENT_EVENT || eventType == EventType.COMPOSITE_DATA_MOVEMENT_EVENT || eventType == EventType.ROOT_INPUT_INITIALIZER_EVENT || eventType == EventType.VERTEX_MANAGER_EVENT) {
                    taGeneratedEvents.add(tezEvent);
                }
                eventsForVertex.add(tezEvent);
            }
        }
        if (taskAttemptEvent != null) {
            taskAttemptEvent.setReadErrorReported(readErrorReported);
            sendEvent(taskAttemptEvent);
        }
        // route taGeneratedEvents to TaskAttempt
        if (!taGeneratedEvents.isEmpty()) {
            sendEvent(new TaskAttemptEventTezEventUpdate(taskAttemptID, taGeneratedEvents));
        }
        // route events to TaskAttempt
        Preconditions.checkArgument(taFinishedEvents.size() <= 1, "Multiple TaskAttemptFinishedEvent");
        for (TezEvent e : taFinishedEvents) {
            EventMetaData sourceMeta = e.getSourceInfo();
            switch(e.getEventType()) {
                case TASK_ATTEMPT_FAILED_EVENT:
                case TASK_ATTEMPT_KILLED_EVENT:
                    TaskAttemptTerminationCause errCause = null;
                    switch(sourceMeta.getEventGenerator()) {
                        case INPUT:
                            errCause = TaskAttemptTerminationCause.INPUT_READ_ERROR;
                            break;
                        case PROCESSOR:
                            errCause = TaskAttemptTerminationCause.APPLICATION_ERROR;
                            break;
                        case OUTPUT:
                            errCause = TaskAttemptTerminationCause.OUTPUT_WRITE_ERROR;
                            break;
                        case SYSTEM:
                            errCause = TaskAttemptTerminationCause.FRAMEWORK_ERROR;
                            break;
                        default:
                            throw new TezUncheckedException("Unknown EventProducerConsumerType: " + sourceMeta.getEventGenerator());
                    }
                    if (e.getEventType() == EventType.TASK_ATTEMPT_FAILED_EVENT) {
                        TaskAttemptFailedEvent taskFailedEvent = (TaskAttemptFailedEvent) e.getEvent();
                        sendEvent(new TaskAttemptEventAttemptFailed(sourceMeta.getTaskAttemptID(), TaskAttemptEventType.TA_FAILED, taskFailedEvent.getTaskFailureType(), "Error: " + taskFailedEvent.getDiagnostics(), errCause));
                    } else {
                        // Killed
                        TaskAttemptKilledEvent taskKilledEvent = (TaskAttemptKilledEvent) e.getEvent();
                        sendEvent(new TaskAttemptEventAttemptKilled(sourceMeta.getTaskAttemptID(), "Error: " + taskKilledEvent.getDiagnostics(), errCause));
                    }
                    break;
                case TASK_ATTEMPT_COMPLETED_EVENT:
                    sendEvent(new TaskAttemptEvent(sourceMeta.getTaskAttemptID(), TaskAttemptEventType.TA_DONE));
                    break;
                default:
                    throw new TezUncheckedException("Unhandled tez event type: " + e.getEventType());
            }
        }
        if (!eventsForVertex.isEmpty()) {
            TezVertexID vertexId = taskAttemptID.getTaskID().getVertexID();
            sendEvent(new VertexEventRouteEvent(vertexId, Collections.unmodifiableList(eventsForVertex)));
        }
        taskHeartbeatHandler.pinged(taskAttemptID);
        eventInfo = context.getCurrentDAG().getVertex(taskAttemptID.getTaskID().getVertexID()).getTaskAttemptTezEvents(taskAttemptID, request.getStartIndex(), request.getPreRoutedStartIndex(), request.getMaxEvents());
    }
    return new TaskHeartbeatResponse(false, eventInfo.getEvents(), eventInfo.getNextFromEventId(), eventInfo.getNextPreRoutedFromEventId());
}
Also used : TezUncheckedException(org.apache.tez.dag.api.TezUncheckedException) TaskAttemptEventStatusUpdate(org.apache.tez.dag.app.dag.event.TaskAttemptEventStatusUpdate) DAGAppMasterEventType(org.apache.tez.dag.app.dag.event.DAGAppMasterEventType) EventType(org.apache.tez.runtime.api.impl.EventType) TaskAttemptEventType(org.apache.tez.dag.app.dag.event.TaskAttemptEventType) ArrayList(java.util.ArrayList) TaskAttemptEvent(org.apache.tez.dag.app.dag.event.TaskAttemptEvent) TaskStatusUpdateEvent(org.apache.tez.runtime.api.events.TaskStatusUpdateEvent) VertexEventRouteEvent(org.apache.tez.dag.app.dag.event.VertexEventRouteEvent) TaskAttemptFailedEvent(org.apache.tez.runtime.api.events.TaskAttemptFailedEvent) TaskAttemptEventTezEventUpdate(org.apache.tez.dag.app.dag.event.TaskAttemptEventTezEventUpdate) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) TaskAttemptEventAttemptKilled(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptKilled) TaskHeartbeatResponse(org.apache.tez.serviceplugins.api.TaskHeartbeatResponse) TezEvent(org.apache.tez.runtime.api.impl.TezEvent) TaskAttemptTerminationCause(org.apache.tez.dag.records.TaskAttemptTerminationCause) TaskAttemptKilledEvent(org.apache.tez.runtime.api.events.TaskAttemptKilledEvent) EventMetaData(org.apache.tez.runtime.api.impl.EventMetaData) TezVertexID(org.apache.tez.dag.records.TezVertexID) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) TaskAttemptEventAttemptFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed)

Example 2 with TaskAttemptEventStatusUpdate

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

the class TestTaskAttempt method testSuccess.

@Test(timeout = 5000)
public void testSuccess() 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 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 expectedEvenstAfterTerminating = expectedEventsAtRunning + 5;
    arg = ArgumentCaptor.forClass(Event.class);
    verify(eventHandler, times(expectedEvenstAfterTerminating)).handle(arg.capture());
    Event e = verifyEventType(arg.getAllValues().subList(expectedEventsAtRunning, expectedEvenstAfterTerminating), TaskEventTASucceeded.class, 1);
    assertEquals(TaskEventType.T_ATTEMPT_SUCCEEDED, e.getType());
    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) 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) 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 3 with TaskAttemptEventStatusUpdate

use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStatusUpdate 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.getID();
    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());
}
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) 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) 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) TezCounters(org.apache.tez.common.counters.TezCounters) ClusterInfo(org.apache.tez.dag.app.ClusterInfo) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) TezTaskAttemptID(org.apache.tez.dag.records.TezTaskAttemptID) Test(org.junit.Test)

Example 4 with TaskAttemptEventStatusUpdate

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

the class TestTaskAttempt method testProgressTimeStampUpdate.

@Test
public void testProgressTimeStampUpdate() 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.setLong(TezConfiguration.TEZ_TASK_PROGRESS_STUCK_INTERVAL_MS, 75);
    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);
    Clock mockClock = mock(Clock.class);
    when(mockClock.getTime()).thenReturn(50l);
    TaskAttemptImpl taImpl = new MockTaskAttemptImpl(taskID, 1, eventHandler, taListener, taskConf, mockClock, 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);
    when(mockClock.getTime()).thenReturn(100l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    verify(eventHandler, atLeast(1)).handle(arg.capture());
    if (arg.getValue() instanceof TaskAttemptEventAttemptFailed) {
        TaskAttemptEventAttemptFailed fEvent = (TaskAttemptEventAttemptFailed) arg.getValue();
        assertEquals(taImpl.getID(), fEvent.getTaskAttemptID());
        assertEquals(TaskAttemptTerminationCause.NO_PROGRESS, fEvent.getTerminationCause());
        taImpl.handle(fEvent);
        fail("Should not fail since the timestamps do not differ by progress interval config");
    } else {
        Assert.assertEquals("Task Attempt's internal state should be RUNNING!", taImpl.getInternalState(), TaskAttemptStateInternal.RUNNING);
    }
    when(mockClock.getTime()).thenReturn(200l);
    taImpl.handle(new TaskAttemptEventStatusUpdate(taskAttemptID, new TaskStatusUpdateEvent(null, 0.1f, null, false)));
    verify(eventHandler, atLeast(1)).handle(arg.capture());
    Assert.assertTrue("This should have been an attempt failed event!", arg.getValue() instanceof TaskAttemptEventAttemptFailed);
}
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) Clock(org.apache.hadoop.yarn.util.Clock) SystemClock(org.apache.hadoop.yarn.util.SystemClock) 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) Resource(org.apache.hadoop.yarn.api.records.Resource) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) 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) 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) TaskAttemptEventAttemptFailed(org.apache.tez.dag.app.dag.event.TaskAttemptEventAttemptFailed) Test(org.junit.Test)

Example 5 with TaskAttemptEventStatusUpdate

use of org.apache.tez.dag.app.dag.event.TaskAttemptEventStatusUpdate 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)

Aggregations

ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)9 TaskAttemptEventStatusUpdate (org.apache.tez.dag.app.dag.event.TaskAttemptEventStatusUpdate)9 TezTaskAttemptID (org.apache.tez.dag.records.TezTaskAttemptID)9 TezVertexID (org.apache.tez.dag.records.TezVertexID)9 TaskStatusUpdateEvent (org.apache.tez.runtime.api.events.TaskStatusUpdateEvent)9 HashSet (java.util.HashSet)8 Configuration (org.apache.hadoop.conf.Configuration)8 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)8 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)8 Container (org.apache.hadoop.yarn.api.records.Container)8 LocalResource (org.apache.hadoop.yarn.api.records.LocalResource)8 NodeId (org.apache.hadoop.yarn.api.records.NodeId)8 Resource (org.apache.hadoop.yarn.api.records.Resource)8 SystemClock (org.apache.hadoop.yarn.util.SystemClock)8 TezConfiguration (org.apache.tez.dag.api.TezConfiguration)8 ClusterInfo (org.apache.tez.dag.app.ClusterInfo)8 ContainerHeartbeatHandler (org.apache.tez.dag.app.ContainerHeartbeatHandler)8 TaskCommunicatorManagerInterface (org.apache.tez.dag.app.TaskCommunicatorManagerInterface)8 TaskHeartbeatHandler (org.apache.tez.dag.app.TaskHeartbeatHandler)8 TaskAttemptEventSchedule (org.apache.tez.dag.app.dag.event.TaskAttemptEventSchedule)8