Search in sources :

Example 36 with OutputCommitter

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

the class TestJobImpl method testKilledDuringSetup.

@Test(timeout = 20000)
public void testKilledDuringSetup() throws Exception {
    Configuration conf = new Configuration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    AsyncDispatcher dispatcher = new AsyncDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    OutputCommitter committer = new StubbedOutputCommitter() {

        @Override
        public synchronized void setupJob(JobContext jobContext) throws IOException {
            while (!Thread.interrupted()) {
                try {
                    wait();
                } catch (InterruptedException e) {
                }
            }
        }
    };
    CommitterEventHandler commitHandler = createCommitterEventHandler(dispatcher, committer);
    commitHandler.init(conf);
    commitHandler.start();
    JobImpl job = createStubbedJob(conf, dispatcher, 2, null);
    JobId jobId = job.getID();
    job.handle(new JobEvent(jobId, JobEventType.JOB_INIT));
    assertJobState(job, JobStateInternal.INITED);
    job.handle(new JobStartEvent(jobId));
    assertJobState(job, JobStateInternal.SETUP);
    job.handle(new JobEvent(job.getID(), JobEventType.JOB_KILL));
    assertJobState(job, JobStateInternal.KILLED);
    dispatcher.stop();
    commitHandler.stop();
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) Configuration(org.apache.hadoop.conf.Configuration) AsyncDispatcher(org.apache.hadoop.yarn.event.AsyncDispatcher) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) JobStartEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobStartEvent) CommitterEventHandler(org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler) JobContext(org.apache.hadoop.mapreduce.JobContext) JobId(org.apache.hadoop.mapreduce.v2.api.records.JobId) Test(org.junit.Test)

Example 37 with OutputCommitter

use of org.apache.hadoop.mapreduce.OutputCommitter 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();
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) Task(org.apache.hadoop.mapreduce.v2.app.job.Task) Configuration(org.apache.hadoop.conf.Configuration) JobTaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobTaskEvent) TaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskEvent) CommitterEventHandler(org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler) TaskTAttemptEvent(org.apache.hadoop.mapreduce.v2.app.job.event.TaskTAttemptEvent) TaskAttempt(org.apache.hadoop.mapreduce.v2.app.job.TaskAttempt) Test(org.junit.Test)

Example 38 with OutputCommitter

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

the class TestJobImpl method testRebootedDuringCommit.

@Test(timeout = 20000)
public void testRebootedDuringCommit() throws Exception {
    Configuration conf = new Configuration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    conf.setInt(MRJobConfig.MR_AM_MAX_ATTEMPTS, 2);
    AsyncDispatcher dispatcher = new AsyncDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    CyclicBarrier syncBarrier = new CyclicBarrier(2);
    OutputCommitter committer = new WaitingOutputCommitter(syncBarrier, true);
    CommitterEventHandler commitHandler = createCommitterEventHandler(dispatcher, committer);
    commitHandler.init(conf);
    commitHandler.start();
    AppContext mockContext = mock(AppContext.class);
    when(mockContext.isLastAMRetry()).thenReturn(true);
    when(mockContext.hasSuccessfullyUnregistered()).thenReturn(false);
    JobImpl job = createRunningStubbedJob(conf, dispatcher, 2, mockContext);
    completeJobTasks(job);
    assertJobState(job, JobStateInternal.COMMITTING);
    syncBarrier.await();
    job.handle(new JobEvent(job.getID(), JobEventType.JOB_AM_REBOOT));
    assertJobState(job, JobStateInternal.REBOOT);
    // return the external state as ERROR since this is last retry.
    Assert.assertEquals(JobState.RUNNING, job.getState());
    when(mockContext.hasSuccessfullyUnregistered()).thenReturn(true);
    Assert.assertEquals(JobState.ERROR, job.getState());
    dispatcher.stop();
    commitHandler.stop();
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) Configuration(org.apache.hadoop.conf.Configuration) AsyncDispatcher(org.apache.hadoop.yarn.event.AsyncDispatcher) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) AppContext(org.apache.hadoop.mapreduce.v2.app.AppContext) CommitterEventHandler(org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler) CyclicBarrier(java.util.concurrent.CyclicBarrier) Test(org.junit.Test)

Example 39 with OutputCommitter

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

the class TestJobImpl method testAbortJobCalledAfterKillingTasks.

@Test
public void testAbortJobCalledAfterKillingTasks() throws IOException {
    Configuration conf = new Configuration();
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    conf.set(MRJobConfig.MR_AM_COMMITTER_CANCEL_TIMEOUT_MS, "1000");
    InlineDispatcher dispatcher = new InlineDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    OutputCommitter committer = Mockito.mock(OutputCommitter.class);
    CommitterEventHandler commitHandler = createCommitterEventHandler(dispatcher, committer);
    commitHandler.init(conf);
    commitHandler.start();
    JobImpl job = createRunningStubbedJob(conf, dispatcher, 2, null);
    //Fail one task. This should land the JobImpl in the FAIL_WAIT state
    job.handle(new JobTaskEvent(MRBuilderUtils.newTaskId(job.getID(), 1, TaskType.MAP), TaskState.FAILED));
    //Verify abort job hasn't been called
    Mockito.verify(committer, Mockito.never()).abortJob((JobContext) Mockito.any(), (State) Mockito.any());
    assertJobState(job, JobStateInternal.FAIL_WAIT);
    //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();
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) Configuration(org.apache.hadoop.conf.Configuration) JobTaskEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobTaskEvent) InlineDispatcher(org.apache.hadoop.yarn.event.InlineDispatcher) CommitterEventHandler(org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler) Test(org.junit.Test)

Example 40 with OutputCommitter

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

the class TestJobImpl method testJobNoTasks.

@Test
public void testJobNoTasks() {
    Configuration conf = new Configuration();
    conf.setInt(MRJobConfig.NUM_REDUCES, 0);
    conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
    conf.set(MRJobConfig.WORKFLOW_ID, "testId");
    conf.set(MRJobConfig.WORKFLOW_NAME, "testName");
    conf.set(MRJobConfig.WORKFLOW_NODE_NAME, "testNodeName");
    conf.set(MRJobConfig.WORKFLOW_ADJACENCY_PREFIX_STRING + "key1", "value1");
    conf.set(MRJobConfig.WORKFLOW_ADJACENCY_PREFIX_STRING + "key2", "value2");
    conf.set(MRJobConfig.WORKFLOW_TAGS, "tag1,tag2");
    AsyncDispatcher dispatcher = new AsyncDispatcher();
    dispatcher.init(conf);
    dispatcher.start();
    OutputCommitter committer = mock(OutputCommitter.class);
    CommitterEventHandler commitHandler = createCommitterEventHandler(dispatcher, committer);
    commitHandler.init(conf);
    commitHandler.start();
    JobSubmittedEventHandler jseHandler = new JobSubmittedEventHandler("testId", "testName", "testNodeName", "\"key2\"=\"value2\" \"key1\"=\"value1\" ", "tag1,tag2");
    dispatcher.register(EventType.class, jseHandler);
    JobImpl job = createStubbedJob(conf, dispatcher, 0, null);
    job.handle(new JobEvent(job.getID(), JobEventType.JOB_INIT));
    assertJobState(job, JobStateInternal.INITED);
    job.handle(new JobStartEvent(job.getID()));
    assertJobState(job, JobStateInternal.SUCCEEDED);
    dispatcher.stop();
    commitHandler.stop();
    try {
        Assert.assertTrue(jseHandler.getAssertValue());
    } catch (InterruptedException e) {
        Assert.fail("Workflow related attributes are not tested properly");
    }
}
Also used : OutputCommitter(org.apache.hadoop.mapreduce.OutputCommitter) Configuration(org.apache.hadoop.conf.Configuration) AsyncDispatcher(org.apache.hadoop.yarn.event.AsyncDispatcher) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) JobStartEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobStartEvent) CommitterEventHandler(org.apache.hadoop.mapreduce.v2.app.commit.CommitterEventHandler) 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