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

Example 96 with TaskAttempt

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

the class TestTaskAttempt method verifyMillisCounters.

public void verifyMillisCounters(Resource containerResource, int minContainerSize) throws Exception {
    Clock actualClock = SystemClock.getInstance();
    ControlledClock clock = new ControlledClock(actualClock);
    clock.setTime(10);
    MRApp app = new MRApp(1, 1, false, "testSlotMillisCounterUpdate", true, clock);
    app.setAllocatedContainerResource(containerResource);
    Configuration conf = new Configuration();
    conf.setInt(YarnConfiguration.RM_SCHEDULER_MINIMUM_ALLOCATION_MB, minContainerSize);
    app.setClusterInfo(new ClusterInfo(Resource.newInstance(10240, 1)));
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Map<TaskId, Task> tasks = job.getTasks();
    Assert.assertEquals("Num tasks is not correct", 2, tasks.size());
    Iterator<Task> taskIter = tasks.values().iterator();
    Task mTask = taskIter.next();
    app.waitForState(mTask, TaskState.RUNNING);
    Task rTask = taskIter.next();
    app.waitForState(rTask, TaskState.RUNNING);
    Map<TaskAttemptId, TaskAttempt> mAttempts = mTask.getAttempts();
    Assert.assertEquals("Num attempts is not correct", 1, mAttempts.size());
    Map<TaskAttemptId, TaskAttempt> rAttempts = rTask.getAttempts();
    Assert.assertEquals("Num attempts is not correct", 1, rAttempts.size());
    TaskAttempt mta = mAttempts.values().iterator().next();
    TaskAttempt rta = rAttempts.values().iterator().next();
    app.waitForState(mta, TaskAttemptState.RUNNING);
    app.waitForState(rta, TaskAttemptState.RUNNING);
    clock.setTime(11);
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mta.getID(), TaskAttemptEventType.TA_DONE));
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(rta.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(job, JobState.SUCCEEDED);
    Assert.assertEquals(mta.getFinishTime(), 11);
    Assert.assertEquals(mta.getLaunchTime(), 10);
    Assert.assertEquals(rta.getFinishTime(), 11);
    Assert.assertEquals(rta.getLaunchTime(), 10);
    Counters counters = job.getAllCounters();
    int memoryMb = (int) containerResource.getMemorySize();
    int vcores = containerResource.getVirtualCores();
    Assert.assertEquals((int) Math.ceil((float) memoryMb / minContainerSize), counters.findCounter(JobCounter.SLOTS_MILLIS_MAPS).getValue());
    Assert.assertEquals((int) Math.ceil((float) memoryMb / minContainerSize), counters.findCounter(JobCounter.SLOTS_MILLIS_REDUCES).getValue());
    Assert.assertEquals(1, counters.findCounter(JobCounter.MILLIS_MAPS).getValue());
    Assert.assertEquals(1, counters.findCounter(JobCounter.MILLIS_REDUCES).getValue());
    Assert.assertEquals(memoryMb, counters.findCounter(JobCounter.MB_MILLIS_MAPS).getValue());
    Assert.assertEquals(memoryMb, counters.findCounter(JobCounter.MB_MILLIS_REDUCES).getValue());
    Assert.assertEquals(vcores, counters.findCounter(JobCounter.VCORES_MILLIS_MAPS).getValue());
    Assert.assertEquals(vcores, counters.findCounter(JobCounter.VCORES_MILLIS_REDUCES).getValue());
}

Example 97 with TaskAttempt

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

the class TestMRApp method testUpdatedNodes.

/**
   * The test verifies that the AM re-runs maps that have run on bad nodes. It
   * also verifies that the AM records all success/killed events so that reduces
   * are notified about map output status changes. It also verifies that the
   * re-run information is preserved across AM restart
   */
@Test
public void testUpdatedNodes() throws Exception {
    int runCount = 0;
    Dispatcher disp = Mockito.spy(new AsyncDispatcher());
    MRApp app = new MRAppWithHistory(2, 2, false, this.getClass().getName(), true, ++runCount, disp);
    Configuration conf = new Configuration();
    // after half of the map completion, reduce will start
    conf.setFloat(MRJobConfig.COMPLETED_MAPS_FOR_REDUCE_SLOWSTART, 0.5f);
    // uberization forces full slowstart (1.0), so disable that
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    ContainerAllocEventHandler handler = new ContainerAllocEventHandler();
    disp.register(ContainerAllocator.EventType.class, handler);
    final Job job1 = app.submit(conf);
    app.waitForState(job1, JobState.RUNNING);
    Assert.assertEquals("Num tasks not correct", 4, job1.getTasks().size());
    Iterator<Task> it = job1.getTasks().values().iterator();
    Task mapTask1 = it.next();
    Task mapTask2 = it.next();
    // all maps must be running
    app.waitForState(mapTask1, TaskState.RUNNING);
    app.waitForState(mapTask2, TaskState.RUNNING);
    TaskAttempt task1Attempt = mapTask1.getAttempts().values().iterator().next();
    TaskAttempt task2Attempt = mapTask2.getAttempts().values().iterator().next();
    NodeId node1 = task1Attempt.getNodeId();
    NodeId node2 = task2Attempt.getNodeId();
    Assert.assertEquals(node1, node2);
    // send the done signal to the task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt.getID(), TaskAttemptEventType.TA_DONE));
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task2Attempt.getID(), TaskAttemptEventType.TA_DONE));
    // all maps must be succeeded
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    app.waitForState(mapTask2, TaskState.SUCCEEDED);
    final int checkIntervalMillis = 100;
    final int waitForMillis = 800;
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job1.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 2;
        }
    }, checkIntervalMillis, waitForMillis);
    TaskAttemptCompletionEvent[] events = job1.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 2 completion events for success", 2, events.length);
    // send updated nodes info
    ArrayList<NodeReport> updatedNodes = new ArrayList<NodeReport>();
    NodeReport nr = RecordFactoryProvider.getRecordFactory(null).newRecordInstance(NodeReport.class);
    nr.setNodeId(node1);
    nr.setNodeState(NodeState.UNHEALTHY);
    updatedNodes.add(nr);
    app.getContext().getEventHandler().handle(new JobUpdatedNodesEvent(job1.getID(), updatedNodes));
    app.waitForState(task1Attempt, TaskAttemptState.KILLED);
    app.waitForState(task2Attempt, TaskAttemptState.KILLED);
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job1.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 4;
        }
    }, checkIntervalMillis, waitForMillis);
    events = job1.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 2 more completion events for killed", 4, events.length);
    // 2 map task attempts which were killed above should be requested from
    // container allocator with the previous map task marked as failed. If
    // this happens allocator will request the container for this mapper from
    // RM at a higher priority of 5(i.e. with a priority equivalent to that of
    // a fail fast map).
    handler.waitForFailedMapContainerReqEvents(2);
    // all maps must be back to running
    app.waitForState(mapTask1, TaskState.RUNNING);
    app.waitForState(mapTask2, TaskState.RUNNING);
    Iterator<TaskAttempt> itr = mapTask1.getAttempts().values().iterator();
    itr.next();
    task1Attempt = itr.next();
    // send the done signal to the task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt.getID(), TaskAttemptEventType.TA_DONE));
    // map1 must be succeeded. map2 must be running
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    app.waitForState(mapTask2, TaskState.RUNNING);
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job1.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 5;
        }
    }, checkIntervalMillis, waitForMillis);
    events = job1.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 1 more completion events for success", 5, events.length);
    // Crash the app again.
    app.stop();
    // rerun
    // in rerun the 1st map will be recovered from previous run
    app = new MRAppWithHistory(2, 2, false, this.getClass().getName(), false, ++runCount, (Dispatcher) new AsyncDispatcher());
    conf = new Configuration();
    conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, true);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    final Job job2 = app.submit(conf);
    app.waitForState(job2, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 4, job2.getTasks().size());
    it = job2.getTasks().values().iterator();
    mapTask1 = it.next();
    mapTask2 = it.next();
    Task reduceTask1 = it.next();
    Task reduceTask2 = it.next();
    // map 1 will be recovered, no need to send done
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    app.waitForState(mapTask2, TaskState.RUNNING);
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job2.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 2;
        }
    }, checkIntervalMillis, waitForMillis);
    events = job2.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 2 completion events for killed & success of map1", 2, events.length);
    task2Attempt = mapTask2.getAttempts().values().iterator().next();
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task2Attempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(mapTask2, TaskState.SUCCEEDED);
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job2.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 3;
        }
    }, checkIntervalMillis, waitForMillis);
    events = job2.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 1 more completion events for success", 3, events.length);
    app.waitForState(reduceTask1, TaskState.RUNNING);
    app.waitForState(reduceTask2, TaskState.RUNNING);
    TaskAttempt task3Attempt = reduceTask1.getAttempts().values().iterator().next();
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task3Attempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task3Attempt.getID(), TaskAttemptEventType.TA_KILL));
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    TaskAttempt task4Attempt = reduceTask2.getAttempts().values().iterator().next();
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task4Attempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(reduceTask2, TaskState.SUCCEEDED);
    waitFor(new Supplier<Boolean>() {

        @Override
        public Boolean get() {
            TaskAttemptCompletionEvent[] events = job2.getTaskAttemptCompletionEvents(0, 100);
            return events.length == 5;
        }
    }, checkIntervalMillis, waitForMillis);
    events = job2.getTaskAttemptCompletionEvents(0, 100);
    Assert.assertEquals("Expecting 2 more completion events for reduce success", 5, events.length);
    // job succeeds
    app.waitForState(job2, JobState.SUCCEEDED);
}

Example 98 with TaskAttempt

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

the class TestMRApp method testContainerPassThrough.

@Test
public void testContainerPassThrough() throws Exception {
    MRApp app = new MRApp(0, 1, true, this.getClass().getName(), true) {

        @Override
        protected ContainerLauncher createContainerLauncher(AppContext context) {
            return new MockContainerLauncher() {

                @Override
                public void handle(ContainerLauncherEvent event) {
                    if (event instanceof ContainerRemoteLaunchEvent) {
                        containerObtainedByContainerLauncher = ((ContainerRemoteLaunchEvent) event).getAllocatedContainer();
                    }
                    super.handle(event);
                }
            };
        }

        ;
    };
    Job job = app.submit(new Configuration());
    app.waitForState(job, JobState.SUCCEEDED);
    app.verifyCompleted();
    Collection<Task> tasks = job.getTasks().values();
    Collection<TaskAttempt> taskAttempts = tasks.iterator().next().getAttempts().values();
    TaskAttemptImpl taskAttempt = (TaskAttemptImpl) taskAttempts.iterator().next();
    // Container from RM should pass through to the launcher. Container object
    // should be the same.
    Assert.assertTrue(taskAttempt.container == containerObtainedByContainerLauncher);
}

Example 99 with TaskAttempt

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

the class TestJobImpl method testFailAbortDoesntHang.

@Test(timeout = 10000)
public void testFailAbortDoesntHang() throws IOException {
    Configuration conf = new Configuration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    conf.set(MRJobConfig.MR_AM_COMMITTER_CANCEL_TIMEOUT_MS, "1000");
    DrainDispatcher dispatcher = new DrainDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    OutputCommitter committer = Mockito.mock(OutputCommitter.class);
    CommitterEventHandler commitHandler = createCommitterEventHandler(dispatcher, committer);
    commitHandler.init(conf);
    commitHandler.start();
    //Job has only 1 mapper task. No reducers
    conf.setInt(MRJobConfig.NUM_REDUCES, 0);
    conf.setInt(MRJobConfig.MAP_MAX_ATTEMPTS, 1);
    JobImpl job = createRunningStubbedJob(conf, dispatcher, 1, null);
    //FAIL_ABORT state
    for (Task t : job.tasks.values()) {
        TaskImpl task = (TaskImpl) t;
        task.handle(new TaskEvent(task.getID(), TaskEventType.T_SCHEDULE));
        for (TaskAttempt ta : task.getAttempts().values()) {
            task.handle(new TaskTAttemptEvent(ta.getID(), TaskEventType.T_ATTEMPT_FAILED));
        }
    }
    dispatcher.await();
    //Verify abortJob is called once and the job failed
    Mockito.verify(committer, Mockito.timeout(2000).times(1)).abortJob((JobContext) Mockito.any(), (State) Mockito.any());
    assertJobState(job, JobStateInternal.FAILED);
    dispatcher.stop();
}

Example 100 with TaskAttempt

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

the class TestKill method testKillTask.

@Test
public void testKillTask() throws Exception {
    final CountDownLatch latch = new CountDownLatch(1);
    MRApp app = new BlockingMRApp(2, 0, latch);
    //this will start the job but job won't complete as Task is blocked
    Job job = app.submit(new Configuration());
    //wait and vailidate for Job to become RUNNING
    app.waitForInternalState((JobImpl) job, JobStateInternal.RUNNING);
    Map<TaskId, Task> tasks = job.getTasks();
    Assert.assertEquals("No of tasks is not correct", 2, tasks.size());
    Iterator<Task> it = tasks.values().iterator();
    Task task1 = it.next();
    Task task2 = it.next();
    //send the kill signal to the first Task
    app.getContext().getEventHandler().handle(new TaskEvent(task1.getID(), TaskEventType.T_KILL));
    //unblock Task
    latch.countDown();
    //wait and validate for Job to become SUCCEEDED
    app.waitForState(job, JobState.SUCCEEDED);
    //first Task is killed and second is Succeeded
    //Job is succeeded
    Assert.assertEquals("Task state not correct", TaskState.KILLED, task1.getReport().getTaskState());
    Assert.assertEquals("Task state not correct", TaskState.SUCCEEDED, task2.getReport().getTaskState());
    Map<TaskAttemptId, TaskAttempt> attempts = task1.getAttempts();
    Assert.assertEquals("No of attempts is not correct", 1, attempts.size());
    Iterator<TaskAttempt> iter = attempts.values().iterator();
    Assert.assertEquals("Attempt state not correct", TaskAttemptState.KILLED, iter.next().getReport().getTaskAttemptState());
    attempts = task2.getAttempts();
    Assert.assertEquals("No of attempts is not correct", 1, attempts.size());
    iter = attempts.values().iterator();
    Assert.assertEquals("Attempt state not correct", TaskAttemptState.SUCCEEDED, iter.next().getReport().getTaskAttemptState());
}