Examples with TaskAttemptEvent org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent used on opensource projects

Example 41 with TaskAttemptEvent

use of org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent in project hadoop by apache.

the class TaskAttemptImpl method handle.

@SuppressWarnings("unchecked")
@Override
public void handle(TaskAttemptEvent event) {
    if (LOG.isDebugEnabled()) {
        LOG.debug("Processing " + event.getTaskAttemptID() + " of type " + event.getType());
    }
    writeLock.lock();
    try {
        final TaskAttemptStateInternal oldState = getInternalState();
        try {
            stateMachine.doTransition(event.getType(), event);
        } catch (InvalidStateTransitionException e) {
            LOG.error("Can't handle this event at current state for " + this.attemptId, e);
            eventHandler.handle(new JobDiagnosticsUpdateEvent(this.attemptId.getTaskId().getJobId(), "Invalid event " + event.getType() + " on TaskAttempt " + this.attemptId));
            eventHandler.handle(new JobEvent(this.attemptId.getTaskId().getJobId(), JobEventType.INTERNAL_ERROR));
        }
        if (oldState != getInternalState()) {
            if (getInternalState() == TaskAttemptStateInternal.FAILED) {
                String nodeId = null == this.container ? "Not-assigned" : this.container.getNodeId().toString();
                LOG.info(attemptId + " transitioned from state " + oldState + " to " + getInternalState() + ", event type is " + event.getType() + " and nodeId=" + nodeId);
            } else {
                LOG.info(attemptId + " TaskAttempt Transitioned from " + oldState + " to " + getInternalState());
            }
        }
    } finally {
        writeLock.unlock();
    }
}

Example 42 with TaskAttemptEvent

use of org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent in project hadoop by apache.

the class TestLocalContainerLauncher method testKillJob.

@SuppressWarnings("rawtypes")
@Test(timeout = 10000)
public void testKillJob() throws Exception {
    JobConf conf = new JobConf();
    AppContext context = mock(AppContext.class);
    // a simple event handler solely to detect the container cleaned event
    final CountDownLatch isDone = new CountDownLatch(1);
    EventHandler<Event> handler = new EventHandler<Event>() {

        @Override
        public void handle(Event event) {
            LOG.info("handling event " + event.getClass() + " with type " + event.getType());
            if (event instanceof TaskAttemptEvent) {
                if (event.getType() == TaskAttemptEventType.TA_CONTAINER_CLEANED) {
                    isDone.countDown();
                }
            }
        }
    };
    when(context.getEventHandler()).thenReturn(handler);
    // create and start the launcher
    LocalContainerLauncher launcher = new LocalContainerLauncher(context, mock(TaskUmbilicalProtocol.class));
    launcher.init(conf);
    launcher.start();
    // create mocked job, task, and task attempt
    // a single-mapper job
    JobId jobId = MRBuilderUtils.newJobId(System.currentTimeMillis(), 1, 1);
    TaskId taskId = MRBuilderUtils.newTaskId(jobId, 1, TaskType.MAP);
    TaskAttemptId taId = MRBuilderUtils.newTaskAttemptId(taskId, 0);
    Job job = mock(Job.class);
    when(job.getTotalMaps()).thenReturn(1);
    when(job.getTotalReduces()).thenReturn(0);
    Map<JobId, Job> jobs = new HashMap<JobId, Job>();
    jobs.put(jobId, job);
    // app context returns the one and only job
    when(context.getAllJobs()).thenReturn(jobs);
    org.apache.hadoop.mapreduce.v2.app.job.Task ytask = mock(org.apache.hadoop.mapreduce.v2.app.job.Task.class);
    when(ytask.getType()).thenReturn(TaskType.MAP);
    when(job.getTask(taskId)).thenReturn(ytask);
    // create a sleeping mapper that runs beyond the test timeout
    MapTask mapTask = mock(MapTask.class);
    when(mapTask.isMapOrReduce()).thenReturn(true);
    when(mapTask.isMapTask()).thenReturn(true);
    TaskAttemptID taskID = TypeConverter.fromYarn(taId);
    when(mapTask.getTaskID()).thenReturn(taskID);
    when(mapTask.getJobID()).thenReturn(taskID.getJobID());
    doAnswer(new Answer<Void>() {

        @Override
        public Void answer(InvocationOnMock invocation) throws Throwable {
            // sleep for a long time
            LOG.info("sleeping for 5 minutes...");
            Thread.sleep(5 * 60 * 1000);
            return null;
        }
    }).when(mapTask).run(isA(JobConf.class), isA(TaskUmbilicalProtocol.class));
    // pump in a task attempt launch event
    ContainerLauncherEvent launchEvent = new ContainerRemoteLaunchEvent(taId, null, createMockContainer(), mapTask);
    launcher.handle(launchEvent);
    Thread.sleep(200);
    // now pump in a container clean-up event
    ContainerLauncherEvent cleanupEvent = new ContainerLauncherEvent(taId, null, null, null, ContainerLauncher.EventType.CONTAINER_REMOTE_CLEANUP);
    launcher.handle(cleanupEvent);
    // wait for the event to fire: this should be received promptly
    isDone.await();
    launcher.close();
}

Example 43 with TaskAttemptEvent

use of org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent in project hadoop by apache.

the class TestFetchFailure method testFetchFailureMultipleReduces.

@Test
public void testFetchFailureMultipleReduces() throws Exception {
    MRApp app = new MRApp(1, 3, false, this.getClass().getName(), true);
    Configuration conf = new Configuration();
    // map -> reduce -> fetch-failure -> map retry is incompatible with
    // sequential, single-task-attempt approach in uber-AM, so disable:
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    //all maps would be running
    Assert.assertEquals("Num tasks not correct", 4, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task mapTask = it.next();
    Task reduceTask = it.next();
    Task reduceTask2 = it.next();
    Task reduceTask3 = it.next();
    //wait for Task state move to RUNNING
    app.waitForState(mapTask, TaskState.RUNNING);
    TaskAttempt mapAttempt1 = mapTask.getAttempts().values().iterator().next();
    app.waitForState(mapAttempt1, TaskAttemptState.RUNNING);
    //send the done signal to the map attempt
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapAttempt1.getID(), TaskAttemptEventType.TA_DONE));
    // wait for map success
    app.waitForState(mapTask, TaskState.SUCCEEDED);
    TaskAttemptCompletionEvent[] events = job.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Num completion events not correct", 1, events.length);
    Assert.assertEquals("Event status not correct", TaskAttemptCompletionEventStatus.SUCCEEDED, events[0].getStatus());
    // wait for reduce to start running
    app.waitForState(reduceTask, TaskState.RUNNING);
    app.waitForState(reduceTask2, TaskState.RUNNING);
    app.waitForState(reduceTask3, TaskState.RUNNING);
    TaskAttempt reduceAttempt = reduceTask.getAttempts().values().iterator().next();
    app.waitForState(reduceAttempt, TaskAttemptState.RUNNING);
    updateStatus(app, reduceAttempt, Phase.SHUFFLE);
    TaskAttempt reduceAttempt2 = reduceTask2.getAttempts().values().iterator().next();
    app.waitForState(reduceAttempt2, TaskAttemptState.RUNNING);
    updateStatus(app, reduceAttempt2, Phase.SHUFFLE);
    TaskAttempt reduceAttempt3 = reduceTask3.getAttempts().values().iterator().next();
    app.waitForState(reduceAttempt3, TaskAttemptState.RUNNING);
    updateStatus(app, reduceAttempt3, Phase.SHUFFLE);
    //send 2 fetch failures from reduce to prepare for map re execution
    sendFetchFailure(app, reduceAttempt, mapAttempt1, "host1");
    sendFetchFailure(app, reduceAttempt2, mapAttempt1, "host2");
    //We should not re-launch the map task yet
    assertEquals(TaskState.SUCCEEDED, mapTask.getState());
    updateStatus(app, reduceAttempt2, Phase.REDUCE);
    updateStatus(app, reduceAttempt3, Phase.REDUCE);
    //send 3rd fetch failures from reduce to trigger map re execution
    sendFetchFailure(app, reduceAttempt3, mapAttempt1, "host3");
    //wait for map Task state move back to RUNNING
    app.waitForState(mapTask, TaskState.RUNNING);
    //map attempt must have become FAILED
    Assert.assertEquals("Map TaskAttempt state not correct", TaskAttemptState.FAILED, mapAttempt1.getState());
    Assert.assertEquals(mapAttempt1.getDiagnostics().get(0), "Too many fetch failures. Failing the attempt. " + "Last failure reported by " + reduceAttempt3.getID().toString() + " from host host3");
    Assert.assertEquals("Num attempts in Map Task not correct", 2, mapTask.getAttempts().size());
    Iterator<TaskAttempt> atIt = mapTask.getAttempts().values().iterator();
    atIt.next();
    TaskAttempt mapAttempt2 = atIt.next();
    app.waitForState(mapAttempt2, TaskAttemptState.RUNNING);
    //send the done signal to the second map attempt
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapAttempt2.getID(), TaskAttemptEventType.TA_DONE));
    // wait for map success
    app.waitForState(mapTask, TaskState.SUCCEEDED);
    //send done to reduce
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduceAttempt.getID(), TaskAttemptEventType.TA_DONE));
    //send done to reduce
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduceAttempt2.getID(), TaskAttemptEventType.TA_DONE));
    //send done to reduce
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduceAttempt3.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(job, JobState.SUCCEEDED);
    //previous completion event now becomes obsolete
    Assert.assertEquals("Event status not correct", TaskAttemptCompletionEventStatus.OBSOLETE, events[0].getStatus());
    events = job.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Num completion events not correct", 6, events.length);
    Assert.assertEquals("Event map attempt id not correct", mapAttempt1.getID(), events[0].getAttemptId());
    Assert.assertEquals("Event map attempt id not correct", mapAttempt1.getID(), events[1].getAttemptId());
    Assert.assertEquals("Event map attempt id not correct", mapAttempt2.getID(), events[2].getAttemptId());
    Assert.assertEquals("Event reduce attempt id not correct", reduceAttempt.getID(), events[3].getAttemptId());
    Assert.assertEquals("Event status not correct for map attempt1", TaskAttemptCompletionEventStatus.OBSOLETE, events[0].getStatus());
    Assert.assertEquals("Event status not correct for map attempt1", TaskAttemptCompletionEventStatus.FAILED, events[1].getStatus());
    Assert.assertEquals("Event status not correct for map attempt2", TaskAttemptCompletionEventStatus.SUCCEEDED, events[2].getStatus());
    Assert.assertEquals("Event status not correct for reduce attempt1", TaskAttemptCompletionEventStatus.SUCCEEDED, events[3].getStatus());
    TaskCompletionEvent[] mapEvents = job.getMapAttemptCompletionEvents(0, 2);
    TaskCompletionEvent[] convertedEvents = TypeConverter.fromYarn(events);
    Assert.assertEquals("Incorrect number of map events", 2, mapEvents.length);
    Assert.assertArrayEquals("Unexpected map events", Arrays.copyOfRange(convertedEvents, 0, 2), mapEvents);
    mapEvents = job.getMapAttemptCompletionEvents(2, 200);
    Assert.assertEquals("Incorrect number of map events", 1, mapEvents.length);
    Assert.assertEquals("Unexpected map event", convertedEvents[2], mapEvents[0]);
}

Example 44 with TaskAttemptEvent

use of org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent in project hadoop by apache.

the class TestMRClientService method test.

@Test
public void test() throws Exception {
    MRAppWithClientService app = new MRAppWithClientService(1, 0, false);
    Configuration conf = new Configuration();
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("Num tasks not correct", 1, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task task = it.next();
    app.waitForState(task, TaskState.RUNNING);
    TaskAttempt attempt = task.getAttempts().values().iterator().next();
    app.waitForState(attempt, TaskAttemptState.RUNNING);
    // send the diagnostic
    String diagnostic1 = "Diagnostic1";
    String diagnostic2 = "Diagnostic2";
    app.getContext().getEventHandler().handle(new TaskAttemptDiagnosticsUpdateEvent(attempt.getID(), diagnostic1));
    // send the status update
    TaskAttemptStatus taskAttemptStatus = new TaskAttemptStatus();
    taskAttemptStatus.id = attempt.getID();
    taskAttemptStatus.progress = 0.5f;
    taskAttemptStatus.stateString = "RUNNING";
    taskAttemptStatus.taskState = TaskAttemptState.RUNNING;
    taskAttemptStatus.phase = Phase.MAP;
    // send the status update
    app.getContext().getEventHandler().handle(new TaskAttemptStatusUpdateEvent(attempt.getID(), taskAttemptStatus));
    //verify that all object are fully populated by invoking RPCs.
    YarnRPC rpc = YarnRPC.create(conf);
    MRClientProtocol proxy = (MRClientProtocol) rpc.getProxy(MRClientProtocol.class, app.clientService.getBindAddress(), conf);
    GetCountersRequest gcRequest = recordFactory.newRecordInstance(GetCountersRequest.class);
    gcRequest.setJobId(job.getID());
    Assert.assertNotNull("Counters is null", proxy.getCounters(gcRequest).getCounters());
    GetJobReportRequest gjrRequest = recordFactory.newRecordInstance(GetJobReportRequest.class);
    gjrRequest.setJobId(job.getID());
    JobReport jr = proxy.getJobReport(gjrRequest).getJobReport();
    verifyJobReport(jr);
    GetTaskAttemptCompletionEventsRequest gtaceRequest = recordFactory.newRecordInstance(GetTaskAttemptCompletionEventsRequest.class);
    gtaceRequest.setJobId(job.getID());
    gtaceRequest.setFromEventId(0);
    gtaceRequest.setMaxEvents(10);
    Assert.assertNotNull("TaskCompletionEvents is null", proxy.getTaskAttemptCompletionEvents(gtaceRequest).getCompletionEventList());
    GetDiagnosticsRequest gdRequest = recordFactory.newRecordInstance(GetDiagnosticsRequest.class);
    gdRequest.setTaskAttemptId(attempt.getID());
    Assert.assertNotNull("Diagnostics is null", proxy.getDiagnostics(gdRequest).getDiagnosticsList());
    GetTaskAttemptReportRequest gtarRequest = recordFactory.newRecordInstance(GetTaskAttemptReportRequest.class);
    gtarRequest.setTaskAttemptId(attempt.getID());
    TaskAttemptReport tar = proxy.getTaskAttemptReport(gtarRequest).getTaskAttemptReport();
    verifyTaskAttemptReport(tar);
    GetTaskReportRequest gtrRequest = recordFactory.newRecordInstance(GetTaskReportRequest.class);
    gtrRequest.setTaskId(task.getID());
    Assert.assertNotNull("TaskReport is null", proxy.getTaskReport(gtrRequest).getTaskReport());
    GetTaskReportsRequest gtreportsRequest = recordFactory.newRecordInstance(GetTaskReportsRequest.class);
    gtreportsRequest.setJobId(job.getID());
    gtreportsRequest.setTaskType(TaskType.MAP);
    Assert.assertNotNull("TaskReports for map is null", proxy.getTaskReports(gtreportsRequest).getTaskReportList());
    gtreportsRequest = recordFactory.newRecordInstance(GetTaskReportsRequest.class);
    gtreportsRequest.setJobId(job.getID());
    gtreportsRequest.setTaskType(TaskType.REDUCE);
    Assert.assertNotNull("TaskReports for reduce is null", proxy.getTaskReports(gtreportsRequest).getTaskReportList());
    List<String> diag = proxy.getDiagnostics(gdRequest).getDiagnosticsList();
    Assert.assertEquals("Num diagnostics not correct", 1, diag.size());
    Assert.assertEquals("Diag 1 not correct", diagnostic1, diag.get(0).toString());
    TaskReport taskReport = proxy.getTaskReport(gtrRequest).getTaskReport();
    Assert.assertEquals("Num diagnostics not correct", 1, taskReport.getDiagnosticsCount());
    //send the done signal to the task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(job, JobState.SUCCEEDED);
    // For invalid jobid, throw IOException
    gtreportsRequest = recordFactory.newRecordInstance(GetTaskReportsRequest.class);
    gtreportsRequest.setJobId(TypeConverter.toYarn(JobID.forName("job_1415730144495_0001")));
    gtreportsRequest.setTaskType(TaskType.REDUCE);
    try {
        proxy.getTaskReports(gtreportsRequest);
        fail("IOException not thrown for invalid job id");
    } catch (IOException e) {
    // Expected
    }
}

Example 45 with TaskAttemptEvent

use of org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent in project hadoop by apache.

the class TestTaskAttempt method testTooManyFetchFailureAfterKill.

@Test
public void testTooManyFetchFailureAfterKill() throws Exception {
    ApplicationId appId = ApplicationId.newInstance(1, 2);
    ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(appId, 0);
    JobId jobId = MRBuilderUtils.newJobId(appId, 1);
    TaskId taskId = MRBuilderUtils.newTaskId(jobId, 1, TaskType.MAP);
    TaskAttemptId attemptId = MRBuilderUtils.newTaskAttemptId(taskId, 0);
    Path jobFile = mock(Path.class);
    MockEventHandler eventHandler = new MockEventHandler();
    TaskAttemptListener taListener = mock(TaskAttemptListener.class);
    when(taListener.getAddress()).thenReturn(new InetSocketAddress("localhost", 0));
    JobConf jobConf = new JobConf();
    jobConf.setClass("fs.file.impl", StubbedFS.class, FileSystem.class);
    jobConf.setBoolean("fs.file.impl.disable.cache", true);
    jobConf.set(JobConf.MAPRED_MAP_TASK_ENV, "");
    jobConf.set(MRJobConfig.APPLICATION_ATTEMPT_ID, "10");
    TaskSplitMetaInfo splits = mock(TaskSplitMetaInfo.class);
    when(splits.getLocations()).thenReturn(new String[] { "127.0.0.1" });
    AppContext appCtx = mock(AppContext.class);
    ClusterInfo clusterInfo = mock(ClusterInfo.class);
    Resource resource = mock(Resource.class);
    when(appCtx.getClusterInfo()).thenReturn(clusterInfo);
    when(resource.getMemorySize()).thenReturn(1024L);
    setupTaskAttemptFinishingMonitor(eventHandler, jobConf, appCtx);
    TaskAttemptImpl taImpl = new MapTaskAttemptImpl(taskId, 1, eventHandler, jobFile, 1, splits, jobConf, taListener, mock(Token.class), new Credentials(), SystemClock.getInstance(), appCtx);
    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");
    taImpl.handle(new TaskAttemptEvent(attemptId, TaskAttemptEventType.TA_SCHEDULE));
    taImpl.handle(new TaskAttemptContainerAssignedEvent(attemptId, container, mock(Map.class)));
    taImpl.handle(new TaskAttemptContainerLaunchedEvent(attemptId, 0));
    taImpl.handle(new TaskAttemptEvent(attemptId, TaskAttemptEventType.TA_DONE));
    taImpl.handle(new TaskAttemptEvent(attemptId, TaskAttemptEventType.TA_CONTAINER_COMPLETED));
    assertEquals("Task attempt is not in succeeded state", taImpl.getState(), TaskAttemptState.SUCCEEDED);
    taImpl.handle(new TaskAttemptEvent(attemptId, TaskAttemptEventType.TA_KILL));
    assertEquals("Task attempt is not in KILLED state", taImpl.getState(), TaskAttemptState.KILLED);
    taImpl.handle(new TaskAttemptEvent(attemptId, TaskAttemptEventType.TA_TOO_MANY_FETCH_FAILURE));
    assertEquals("Task attempt is not in KILLED state, still", taImpl.getState(), TaskAttemptState.KILLED);
    assertFalse("InternalError occurred trying to handle TA_CONTAINER_CLEANED", eventHandler.internalError);
}