Search in sources :

Example 31 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter in project hadoop by apache.

the class TestRecovery method writeBadOutput.

private void writeBadOutput(TaskAttempt attempt, Configuration conf) throws Exception {
    TaskAttemptContext tContext = new TaskAttemptContextImpl(conf, TypeConverter.fromYarn(attempt.getID()));
    TextOutputFormat<?, ?> theOutputFormat = new TextOutputFormat();
    RecordWriter theRecordWriter = theOutputFormat.getRecordWriter(tContext);
    NullWritable nullWritable = NullWritable.get();
    try {
        theRecordWriter.write(key2, val2);
        theRecordWriter.write(null, nullWritable);
        theRecordWriter.write(null, val2);
        theRecordWriter.write(nullWritable, val1);
        theRecordWriter.write(key1, nullWritable);
        theRecordWriter.write(key2, null);
        theRecordWriter.write(null, null);
        theRecordWriter.write(key1, val1);
    } finally {
        theRecordWriter.close(tContext);
    }
    OutputFormat outputFormat = ReflectionUtils.newInstance(tContext.getOutputFormatClass(), conf);
    OutputCommitter committer = outputFormat.getOutputCommitter(tContext);
    committer.commitTask(tContext);
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) RecordWriter(org.apache.hadoop.mapreduce.RecordWriter) TextOutputFormat(org.apache.hadoop.mapreduce.lib.output.TextOutputFormat) TaskAttemptContextImpl(org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl) OutputFormat(org.apache.hadoop.mapreduce.OutputFormat) TextOutputFormat(org.apache.hadoop.mapreduce.lib.output.TextOutputFormat) FileOutputFormat(org.apache.hadoop.mapreduce.lib.output.FileOutputFormat) TaskAttemptContext(org.apache.hadoop.mapreduce.TaskAttemptContext) NullWritable(org.apache.hadoop.io.NullWritable)

Example 32 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter in project hadoop by apache.

the class TestRecovery method testRecoveryTaskSuccessAllAttemptsSucceed.

@Test
public void testRecoveryTaskSuccessAllAttemptsSucceed() {
    LOG.info("--- START:  testRecoveryTaskSuccessAllAttemptsFail ---");
    long clusterTimestamp = System.currentTimeMillis();
    EventHandler mockEventHandler = mock(EventHandler.class);
    MapTaskImpl recoverMapTask = getMockMapTask(clusterTimestamp, mockEventHandler);
    TaskId taskId = recoverMapTask.getID();
    JobID jobID = new JobID(Long.toString(clusterTimestamp), 1);
    TaskID taskID = new TaskID(jobID, org.apache.hadoop.mapreduce.TaskType.MAP, taskId.getId());
    //Mock up the TaskAttempts
    Map<TaskAttemptID, TaskAttemptInfo> mockTaskAttempts = new HashMap<TaskAttemptID, TaskAttemptInfo>();
    TaskAttemptID taId1 = new TaskAttemptID(taskID, 2);
    TaskAttemptInfo mockTAinfo1 = getMockTaskAttemptInfo(taId1, TaskAttemptState.SUCCEEDED);
    mockTaskAttempts.put(taId1, mockTAinfo1);
    TaskAttemptID taId2 = new TaskAttemptID(taskID, 1);
    TaskAttemptInfo mockTAinfo2 = getMockTaskAttemptInfo(taId2, TaskAttemptState.SUCCEEDED);
    mockTaskAttempts.put(taId2, mockTAinfo2);
    OutputCommitter mockCommitter = mock(OutputCommitter.class);
    TaskInfo mockTaskInfo = mock(TaskInfo.class);
    when(mockTaskInfo.getTaskStatus()).thenReturn("SUCCEEDED");
    when(mockTaskInfo.getTaskId()).thenReturn(taskID);
    when(mockTaskInfo.getAllTaskAttempts()).thenReturn(mockTaskAttempts);
    recoverMapTask.handle(new TaskRecoverEvent(taskId, mockTaskInfo, mockCommitter, true));
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    verify(mockEventHandler, atLeast(1)).handle((org.apache.hadoop.yarn.event.Event) arg.capture());
    Map<TaskAttemptID, TaskAttemptState> finalAttemptStates = new HashMap<TaskAttemptID, TaskAttemptState>();
    finalAttemptStates.put(taId1, TaskAttemptState.SUCCEEDED);
    finalAttemptStates.put(taId2, TaskAttemptState.SUCCEEDED);
    List<EventType> jobHistoryEvents = new ArrayList<EventType>();
    jobHistoryEvents.add(EventType.TASK_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_FINISHED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_FINISHED);
    jobHistoryEvents.add(EventType.TASK_FINISHED);
    recoveryChecker(recoverMapTask, TaskState.SUCCEEDED, finalAttemptStates, arg, jobHistoryEvents, 2L, 0L);
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) TaskId(org.apache.hadoop.mapreduce.v2.api.records.TaskId) TaskID(org.apache.hadoop.mapreduce.TaskID) HashMap(java.util.HashMap) TaskAttemptEventType(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEventType) EventType(org.apache.hadoop.mapreduce.jobhistory.EventType) TaskEventType(org.apache.hadoop.mapreduce.v2.app.job.event.TaskEventType) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) ArrayList(java.util.ArrayList) EventHandler(org.apache.hadoop.yarn.event.EventHandler) JobHistoryEventHandler(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEventHandler) MapTaskImpl(org.apache.hadoop.mapreduce.v2.app.job.impl.MapTaskImpl) TaskInfo(org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskInfo) TaskAttemptState(org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState) TaskAttemptInfo(org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskAttemptInfo) TaskAttemptContainerLaunchedEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptContainerLaunchedEvent) Event(org.apache.hadoop.mapreduce.jobhistory.Event) TaskRecoverEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskRecoverEvent) JobTaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobTaskEvent) JobHistoryEvent(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEvent) JobCounterUpdateEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobCounterUpdateEvent) ContainerLauncherEvent(org.apache.hadoop.mapreduce.v2.app.launcher.ContainerLauncherEvent) TaskAttemptEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent) TaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskEvent) JobID(org.apache.hadoop.mapreduce.JobID) TaskRecoverEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskRecoverEvent) Test(org.junit.Test)

Example 33 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter in project hadoop by apache.

the class TestRecovery method testRecoveryAllFailAttempts.

@Test
public void testRecoveryAllFailAttempts() {
    LOG.info("--- START: testRecoveryAllFailAttempts ---");
    long clusterTimestamp = System.currentTimeMillis();
    EventHandler mockEventHandler = mock(EventHandler.class);
    MapTaskImpl recoverMapTask = getMockMapTask(clusterTimestamp, mockEventHandler);
    TaskId taskId = recoverMapTask.getID();
    JobID jobID = new JobID(Long.toString(clusterTimestamp), 1);
    TaskID taskID = new TaskID(jobID, org.apache.hadoop.mapreduce.TaskType.MAP, taskId.getId());
    //Mock up the TaskAttempts
    Map<TaskAttemptID, TaskAttemptInfo> mockTaskAttempts = new HashMap<TaskAttemptID, TaskAttemptInfo>();
    TaskAttemptID taId1 = new TaskAttemptID(taskID, 2);
    TaskAttemptInfo mockTAinfo1 = getMockTaskAttemptInfo(taId1, TaskAttemptState.FAILED);
    mockTaskAttempts.put(taId1, mockTAinfo1);
    TaskAttemptID taId2 = new TaskAttemptID(taskID, 1);
    TaskAttemptInfo mockTAinfo2 = getMockTaskAttemptInfo(taId2, TaskAttemptState.FAILED);
    mockTaskAttempts.put(taId2, mockTAinfo2);
    OutputCommitter mockCommitter = mock(OutputCommitter.class);
    TaskInfo mockTaskInfo = mock(TaskInfo.class);
    when(mockTaskInfo.getTaskStatus()).thenReturn("FAILED");
    when(mockTaskInfo.getTaskId()).thenReturn(taskID);
    when(mockTaskInfo.getAllTaskAttempts()).thenReturn(mockTaskAttempts);
    recoverMapTask.handle(new TaskRecoverEvent(taskId, mockTaskInfo, mockCommitter, true));
    ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
    verify(mockEventHandler, atLeast(1)).handle((org.apache.hadoop.yarn.event.Event) arg.capture());
    Map<TaskAttemptID, TaskAttemptState> finalAttemptStates = new HashMap<TaskAttemptID, TaskAttemptState>();
    finalAttemptStates.put(taId1, TaskAttemptState.FAILED);
    finalAttemptStates.put(taId2, TaskAttemptState.FAILED);
    List<EventType> jobHistoryEvents = new ArrayList<EventType>();
    jobHistoryEvents.add(EventType.TASK_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_FAILED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
    jobHistoryEvents.add(EventType.MAP_ATTEMPT_FAILED);
    jobHistoryEvents.add(EventType.TASK_FAILED);
    recoveryChecker(recoverMapTask, TaskState.FAILED, finalAttemptStates, arg, jobHistoryEvents, 2L, 2L);
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) TaskId(org.apache.hadoop.mapreduce.v2.api.records.TaskId) TaskID(org.apache.hadoop.mapreduce.TaskID) HashMap(java.util.HashMap) TaskAttemptEventType(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEventType) EventType(org.apache.hadoop.mapreduce.jobhistory.EventType) TaskEventType(org.apache.hadoop.mapreduce.v2.app.job.event.TaskEventType) TaskAttemptID(org.apache.hadoop.mapreduce.TaskAttemptID) ArrayList(java.util.ArrayList) EventHandler(org.apache.hadoop.yarn.event.EventHandler) JobHistoryEventHandler(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEventHandler) MapTaskImpl(org.apache.hadoop.mapreduce.v2.app.job.impl.MapTaskImpl) TaskInfo(org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskInfo) TaskAttemptState(org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState) TaskAttemptInfo(org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskAttemptInfo) TaskAttemptContainerLaunchedEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptContainerLaunchedEvent) Event(org.apache.hadoop.mapreduce.jobhistory.Event) TaskRecoverEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskRecoverEvent) JobTaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobTaskEvent) JobHistoryEvent(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEvent) JobCounterUpdateEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobCounterUpdateEvent) ContainerLauncherEvent(org.apache.hadoop.mapreduce.v2.app.launcher.ContainerLauncherEvent) TaskAttemptEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent) TaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskEvent) JobID(org.apache.hadoop.mapreduce.JobID) TaskRecoverEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskRecoverEvent) Test(org.junit.Test)

Example 34 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter in project hadoop by apache.

the class TestCommitterEventHandler method testCommitWindow.

@Test
public void testCommitWindow() throws Exception {
    Configuration conf = new Configuration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    AsyncDispatcher dispatcher = new AsyncDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    TestingJobEventHandler jeh = new TestingJobEventHandler();
    dispatcher.register(JobEventType.class, jeh);
    SystemClock clock = SystemClock.getInstance();
    AppContext appContext = mock(AppContext.class);
    ApplicationAttemptId attemptid = ApplicationAttemptId.fromString("appattempt_1234567890000_0001_0");
    when(appContext.getApplicationID()).thenReturn(attemptid.getApplicationId());
    when(appContext.getApplicationAttemptId()).thenReturn(attemptid);
    when(appContext.getEventHandler()).thenReturn(dispatcher.getEventHandler());
    when(appContext.getClock()).thenReturn(clock);
    OutputCommitter committer = mock(OutputCommitter.class);
    TestingRMHeartbeatHandler rmhh = new TestingRMHeartbeatHandler();
    CommitterEventHandler ceh = new CommitterEventHandler(appContext, committer, rmhh);
    ceh.init(conf);
    ceh.start();
    // verify trying to commit when RM heartbeats are stale does not commit
    ceh.handle(new CommitterJobCommitEvent(null, null));
    long timeToWaitMs = 5000;
    while (rmhh.getNumCallbacks() != 1 && timeToWaitMs > 0) {
        Thread.sleep(10);
        timeToWaitMs -= 10;
    }
    Assert.assertEquals("committer did not register a heartbeat callback", 1, rmhh.getNumCallbacks());
    verify(committer, never()).commitJob(any(JobContext.class));
    Assert.assertEquals("committer should not have committed", 0, jeh.numCommitCompletedEvents);
    // set a fresh heartbeat and verify commit completes
    rmhh.setLastHeartbeatTime(clock.getTime());
    timeToWaitMs = 5000;
    while (jeh.numCommitCompletedEvents != 1 && timeToWaitMs > 0) {
        Thread.sleep(10);
        timeToWaitMs -= 10;
    }
    Assert.assertEquals("committer did not complete commit after RM hearbeat", 1, jeh.numCommitCompletedEvents);
    verify(committer, times(1)).commitJob(any(JobContext.class));
    //Clean up so we can try to commit again (Don't do this at home)
    cleanup();
    // try to commit again and verify it goes through since the heartbeat
    // is still fresh
    ceh.handle(new CommitterJobCommitEvent(null, null));
    timeToWaitMs = 5000;
    while (jeh.numCommitCompletedEvents != 2 && timeToWaitMs > 0) {
        Thread.sleep(10);
        timeToWaitMs -= 10;
    }
    Assert.assertEquals("committer did not commit", 2, jeh.numCommitCompletedEvents);
    verify(committer, times(2)).commitJob(any(JobContext.class));
    ceh.stop();
    dispatcher.stop();
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) Configuration(org.apache.hadoop.conf.Configuration) AsyncDispatcher(org.apache.hadoop.yarn.event.AsyncDispatcher) SystemClock(org.apache.hadoop.yarn.util.SystemClock) AppContext(org.apache.hadoop.mapreduce.v2.app.AppContext) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) JobContext(org.apache.hadoop.mapreduce.JobContext) Test(org.junit.Test)

Example 35 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter in project hadoop by apache.

the class TestCommitterEventHandler method testFailure.

@Test
public void testFailure() throws Exception {
    AppContext mockContext = mock(AppContext.class);
    OutputCommitter mockCommitter = mock(OutputCommitter.class);
    Clock mockClock = mock(Clock.class);
    CommitterEventHandler handler = new CommitterEventHandler(mockContext, mockCommitter, new TestingRMHeartbeatHandler());
    YarnConfiguration conf = new YarnConfiguration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    JobContext mockJobContext = mock(JobContext.class);
    ApplicationAttemptId attemptid = ApplicationAttemptId.fromString("appattempt_1234567890000_0001_0");
    JobId jobId = TypeConverter.toYarn(TypeConverter.fromYarn(attemptid.getApplicationId()));
    WaitForItHandler waitForItHandler = new WaitForItHandler();
    when(mockContext.getApplicationID()).thenReturn(attemptid.getApplicationId());
    when(mockContext.getApplicationAttemptId()).thenReturn(attemptid);
    when(mockContext.getEventHandler()).thenReturn(waitForItHandler);
    when(mockContext.getClock()).thenReturn(mockClock);
    doThrow(new YarnRuntimeException("Intentional Failure")).when(mockCommitter).commitJob(any(JobContext.class));
    handler.init(conf);
    handler.start();
    try {
        handler.handle(new CommitterJobCommitEvent(jobId, mockJobContext));
        String user = UserGroupInformation.getCurrentUser().getShortUserName();
        Path startCommitFile = MRApps.getStartJobCommitFile(conf, user, jobId);
        Path endCommitSuccessFile = MRApps.getEndJobCommitSuccessFile(conf, user, jobId);
        Path endCommitFailureFile = MRApps.getEndJobCommitFailureFile(conf, user, jobId);
        Event e = waitForItHandler.getAndClearEvent();
        assertNotNull(e);
        assertTrue(e instanceof JobCommitFailedEvent);
        FileSystem fs = FileSystem.get(conf);
        assertTrue(fs.exists(startCommitFile));
        assertFalse(fs.exists(endCommitSuccessFile));
        assertTrue(fs.exists(endCommitFailureFile));
        verify(mockCommitter).commitJob(any(JobContext.class));
    } finally {
        handler.stop();
    }
}
Also used : Path(org.apache.hadoop.fs.Path) OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) AppContext(org.apache.hadoop.mapreduce.v2.app.AppContext) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) Clock(org.apache.hadoop.yarn.util.Clock) SystemClock(org.apache.hadoop.yarn.util.SystemClock) YarnRuntimeException(org.apache.hadoop.yarn.exceptions.YarnRuntimeException) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) FileSystem(org.apache.hadoop.fs.FileSystem) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) Event(org.apache.hadoop.yarn.event.Event) JobCommitFailedEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobCommitFailedEvent) JobCommitCompletedEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobCommitCompletedEvent) JobContext(org.apache.hadoop.mapreduce.JobContext) JobId(org.apache.hadoop.mapreduce.v2.api.records.JobId) JobCommitFailedEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobCommitFailedEvent) Test(org.junit.Test)

Aggregations

OutputCommitter (org.apache.hadoop.mapreduce.OutputCommitter)47 Test (org.junit.Test)29 Configuration (org.apache.hadoop.conf.Configuration)23 TaskAttemptContext (org.apache.hadoop.mapreduce.TaskAttemptContext)18 JobContext (org.apache.hadoop.mapreduce.JobContext)13 CommitterEventHandler (org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler)13 JobEvent (org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent)11 AsyncDispatcher (org.apache.hadoop.yarn.event.AsyncDispatcher)11 TaskAttemptID (org.apache.hadoop.mapreduce.TaskAttemptID)10 IOException (java.io.IOException)8 JobTaskEvent (org.apache.hadoop.mapreduce.v2.app.job.event.JobTaskEvent)8 HashMap (java.util.HashMap)7 JobId (org.apache.hadoop.mapreduce.v2.api.records.JobId)7 ArrayList (java.util.ArrayList)6 Map (java.util.Map)6 NullWritable (org.apache.hadoop.io.NullWritable)6 TaskId (org.apache.hadoop.mapreduce.v2.api.records.TaskId)6 AppContext (org.apache.hadoop.mapreduce.v2.app.AppContext)6 JobStartEvent (org.apache.hadoop.mapreduce.v2.app.job.event.JobStartEvent)6 TaskAttemptEvent (org.apache.hadoop.mapreduce.v2.app.job.event.TaskAttemptEvent)6