Examples with MRAppWithHistory org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory used on opensource projects

Example 11 with MRAppWithHistory

use of org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory in project hadoop by apache.

the class TestRecovery method testOutputRecovery.

@Test
public void testOutputRecovery() throws Exception {
    int runCount = 0;
    MRApp app = new MRAppWithHistory(1, 2, false, this.getClass().getName(), true, ++runCount);
    Configuration conf = new Configuration();
    conf.setBoolean("mapred.mapper.new-api", true);
    conf.setBoolean("mapred.reducer.new-api", true);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task mapTask1 = it.next();
    Task reduceTask1 = it.next();
    // all maps must be running
    app.waitForState(mapTask1, TaskState.RUNNING);
    TaskAttempt task1Attempt1 = mapTask1.getAttempts().values().iterator().next();
    //before sending the TA_DONE, event make sure attempt has come to 
    //RUNNING state
    app.waitForState(task1Attempt1, TaskAttemptState.RUNNING);
    //send the done signal to the map
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt1.getID(), TaskAttemptEventType.TA_DONE));
    //wait for map task to complete
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    // Verify the shuffle-port
    Assert.assertEquals(5467, task1Attempt1.getShufflePort());
    app.waitForState(reduceTask1, TaskState.RUNNING);
    TaskAttempt reduce1Attempt1 = reduceTask1.getAttempts().values().iterator().next();
    // write output corresponding to reduce1
    writeOutput(reduce1Attempt1, conf);
    //send the done signal to the 1st reduce
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduce1Attempt1.getID(), TaskAttemptEventType.TA_DONE));
    //wait for first reduce task to complete
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    //stop the app before the job completes.
    app.stop();
    //rerun
    //in rerun the map will be recovered from previous run
    app = new MRAppWithHistory(1, 2, false, this.getClass().getName(), false, ++runCount);
    conf = new Configuration();
    conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, true);
    conf.setBoolean("mapred.mapper.new-api", true);
    conf.setBoolean("mapred.reducer.new-api", true);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    it = job.getTasks().values().iterator();
    mapTask1 = it.next();
    reduceTask1 = it.next();
    Task reduceTask2 = it.next();
    // map will be recovered, no need to send done
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    // Verify the shuffle-port after recovery
    task1Attempt1 = mapTask1.getAttempts().values().iterator().next();
    Assert.assertEquals(5467, task1Attempt1.getShufflePort());
    // first reduce will be recovered, no need to send done
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    app.waitForState(reduceTask2, TaskState.RUNNING);
    TaskAttempt reduce2Attempt = reduceTask2.getAttempts().values().iterator().next();
    //before sending the TA_DONE, event make sure attempt has come to 
    //RUNNING state
    app.waitForState(reduce2Attempt, TaskAttemptState.RUNNING);
    //send the done signal to the 2nd reduce task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduce2Attempt.getID(), TaskAttemptEventType.TA_DONE));
    //wait to get it completed
    app.waitForState(reduceTask2, TaskState.SUCCEEDED);
    app.waitForState(job, JobState.SUCCEEDED);
    app.verifyCompleted();
    validateOutput();
}

Example 12 with MRAppWithHistory

use of org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory in project hadoop by apache.

the class TestRecovery method testRecoveryFailsUsingCustomOutputCommitter.

/**
   * This test case primarily verifies if the recovery is controlled through config
   * property. In this case, recover is turned OFF. AM with 3 maps and 0 reduce.
   * AM crashes after the first two tasks finishes and recovery fails and have
   * to rerun fully in the second generation and succeeds.
   * 
   * @throws Exception
   */
@Test
public void testRecoveryFailsUsingCustomOutputCommitter() throws Exception {
    int runCount = 0;
    MRApp app = new MRAppWithHistory(3, 0, false, this.getClass().getName(), true, ++runCount);
    Configuration conf = new Configuration();
    conf.setClass("mapred.output.committer.class", TestFileOutputCommitter.class, org.apache.hadoop.mapred.OutputCommitter.class);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    conf.setBoolean("want.am.recovery", false);
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    // all maps would be running
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task mapTask1 = it.next();
    Task mapTask2 = it.next();
    Task mapTask3 = it.next();
    // all maps must be running
    app.waitForState(mapTask1, TaskState.RUNNING);
    app.waitForState(mapTask2, TaskState.RUNNING);
    app.waitForState(mapTask3, TaskState.RUNNING);
    TaskAttempt task1Attempt = mapTask1.getAttempts().values().iterator().next();
    TaskAttempt task2Attempt = mapTask2.getAttempts().values().iterator().next();
    TaskAttempt task3Attempt = mapTask3.getAttempts().values().iterator().next();
    // before sending the TA_DONE, event make sure attempt has come to
    // RUNNING state
    app.waitForState(task1Attempt, TaskAttemptState.RUNNING);
    app.waitForState(task2Attempt, TaskAttemptState.RUNNING);
    app.waitForState(task3Attempt, TaskAttemptState.RUNNING);
    // send the done signal to the 1st two maps
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt.getID(), TaskAttemptEventType.TA_DONE));
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task2Attempt.getID(), TaskAttemptEventType.TA_DONE));
    // wait for first two map task to complete
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    app.waitForState(mapTask2, TaskState.SUCCEEDED);
    // stop the app
    app.stop();
    // rerun
    // in rerun the 1st two map will be recovered from previous run
    app = new MRAppWithHistory(2, 1, false, this.getClass().getName(), false, ++runCount);
    conf = new Configuration();
    conf.setClass("mapred.output.committer.class", TestFileOutputCommitter.class, org.apache.hadoop.mapred.OutputCommitter.class);
    conf.setBoolean("want.am.recovery", false);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    // Set num-reduces explicitly in conf as recovery logic depends on it.
    conf.setInt(MRJobConfig.NUM_REDUCES, 0);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    it = job.getTasks().values().iterator();
    mapTask1 = it.next();
    mapTask2 = it.next();
    mapTask3 = it.next();
    // first two maps will NOT  be recovered, need to send done from them
    app.waitForState(mapTask1, TaskState.RUNNING);
    app.waitForState(mapTask2, TaskState.RUNNING);
    app.waitForState(mapTask3, TaskState.RUNNING);
    task3Attempt = mapTask3.getAttempts().values().iterator().next();
    // before sending the TA_DONE, event make sure attempt has come to
    // RUNNING state
    app.waitForState(task3Attempt, TaskAttemptState.RUNNING);
    // send the done signal to all 3 tasks map task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapTask1.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapTask2.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapTask3.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
    // wait to get it completed
    app.waitForState(mapTask3, TaskState.SUCCEEDED);
    app.waitForState(job, JobState.SUCCEEDED);
    app.verifyCompleted();
}

Example 13 with MRAppWithHistory

use of org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory in project hadoop by apache.

the class TestAMInfos method testAMInfosWithoutRecoveryEnabled.

@Test
public void testAMInfosWithoutRecoveryEnabled() throws Exception {
    int runCount = 0;
    MRApp app = new MRAppWithHistory(1, 0, false, this.getClass().getName(), true, ++runCount);
    Configuration conf = new Configuration();
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    long am1StartTime = app.getAllAMInfos().get(0).getStartTime();
    Assert.assertEquals("No of tasks not correct", 1, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task mapTask = it.next();
    app.waitForState(mapTask, TaskState.RUNNING);
    TaskAttempt taskAttempt = mapTask.getAttempts().values().iterator().next();
    app.waitForState(taskAttempt, TaskAttemptState.RUNNING);
    // stop the app
    app.stop();
    // rerun
    app = new MRAppWithHistory(1, 0, false, this.getClass().getName(), false, ++runCount);
    conf = new Configuration();
    // in rerun the AMInfo will be recovered from previous run even if recovery
    // is not enabled.
    conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, false);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 1, job.getTasks().size());
    it = job.getTasks().values().iterator();
    mapTask = it.next();
    // There should be two AMInfos
    List<AMInfo> amInfos = app.getAllAMInfos();
    Assert.assertEquals(2, amInfos.size());
    AMInfo amInfoOne = amInfos.get(0);
    Assert.assertEquals(am1StartTime, amInfoOne.getStartTime());
    app.stop();
}

Example 14 with MRAppWithHistory

use of org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory in project hadoop by apache.

the class TestRecovery method testRecoveryWithOldCommiter.

@Test
public void testRecoveryWithOldCommiter() throws Exception {
    int runCount = 0;
    MRApp app = new MRAppWithHistory(1, 2, false, this.getClass().getName(), true, ++runCount);
    Configuration conf = new Configuration();
    conf.setBoolean("mapred.mapper.new-api", false);
    conf.setBoolean("mapred.reducer.new-api", false);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    Job job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    Iterator<Task> it = job.getTasks().values().iterator();
    Task mapTask1 = it.next();
    Task reduceTask1 = it.next();
    // all maps must be running
    app.waitForState(mapTask1, TaskState.RUNNING);
    TaskAttempt task1Attempt1 = mapTask1.getAttempts().values().iterator().next();
    //before sending the TA_DONE, event make sure attempt has come to 
    //RUNNING state
    app.waitForState(task1Attempt1, TaskAttemptState.RUNNING);
    //send the done signal to the map
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt1.getID(), TaskAttemptEventType.TA_DONE));
    //wait for map task to complete
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    // Verify the shuffle-port
    Assert.assertEquals(5467, task1Attempt1.getShufflePort());
    app.waitForState(reduceTask1, TaskState.RUNNING);
    TaskAttempt reduce1Attempt1 = reduceTask1.getAttempts().values().iterator().next();
    // write output corresponding to reduce1
    writeOutput(reduce1Attempt1, conf);
    //send the done signal to the 1st reduce
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduce1Attempt1.getID(), TaskAttemptEventType.TA_DONE));
    //wait for first reduce task to complete
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    //stop the app before the job completes.
    app.stop();
    //rerun
    //in rerun the map will be recovered from previous run
    app = new MRAppWithHistory(1, 2, false, this.getClass().getName(), false, ++runCount);
    conf = new Configuration();
    conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, true);
    conf.setBoolean("mapred.mapper.new-api", false);
    conf.setBoolean("mapred.reducer.new-api", false);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    job = app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
    it = job.getTasks().values().iterator();
    mapTask1 = it.next();
    reduceTask1 = it.next();
    Task reduceTask2 = it.next();
    // map will be recovered, no need to send done
    app.waitForState(mapTask1, TaskState.SUCCEEDED);
    // Verify the shuffle-port after recovery
    task1Attempt1 = mapTask1.getAttempts().values().iterator().next();
    Assert.assertEquals(5467, task1Attempt1.getShufflePort());
    // first reduce will be recovered, no need to send done
    app.waitForState(reduceTask1, TaskState.SUCCEEDED);
    app.waitForState(reduceTask2, TaskState.RUNNING);
    TaskAttempt reduce2Attempt = reduceTask2.getAttempts().values().iterator().next();
    //before sending the TA_DONE, event make sure attempt has come to 
    //RUNNING state
    app.waitForState(reduce2Attempt, TaskAttemptState.RUNNING);
    //send the done signal to the 2nd reduce task
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduce2Attempt.getID(), TaskAttemptEventType.TA_DONE));
    //wait to get it completed
    app.waitForState(reduceTask2, TaskState.SUCCEEDED);
    app.waitForState(job, JobState.SUCCEEDED);
    app.verifyCompleted();
    validateOutput();
}

Example 15 with MRAppWithHistory

use of org.apache.hadoop.mapreduce.v2.app.TestRecovery.MRAppWithHistory in project hadoop by apache.

the class TestRecovery method testRecoveryWithSpillEncryption.

@Test
public void testRecoveryWithSpillEncryption() throws Exception {
    int runCount = 0;
    MRApp app = new MRAppWithHistory(1, 1, false, this.getClass().getName(), true, ++runCount) {
    };
    Configuration conf = new Configuration();
    conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, true);
    conf.setBoolean("mapred.mapper.new-api", true);
    conf.setBoolean("mapred.reducer.new-api", true);
    conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
    conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
    conf.setBoolean(MRJobConfig.MR_ENCRYPTED_INTERMEDIATE_DATA, true);
    // run the MR job at the first attempt
    Job jobAttempt1 = app.submit(conf);
    app.waitForState(jobAttempt1, JobState.RUNNING);
    Iterator<Task> tasks = jobAttempt1.getTasks().values().iterator();
    // finish the map task but the reduce task
    Task mapper = tasks.next();
    app.waitForState(mapper, TaskState.RUNNING);
    TaskAttempt mapAttempt = mapper.getAttempts().values().iterator().next();
    app.waitForState(mapAttempt, TaskAttemptState.RUNNING);
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapAttempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(mapper, TaskState.SUCCEEDED);
    // crash the first attempt of the MR job
    app.stop();
    // run the MR job again at the second attempt
    app = new MRAppWithHistory(1, 1, false, this.getClass().getName(), false, ++runCount);
    Job jobAttempt2 = app.submit(conf);
    Assert.assertTrue("Recovery from previous job attempt is processed even " + "though intermediate data encryption is enabled.", !app.recovered());
    // The map task succeeded from previous job attempt will not be recovered
    // because the data spill encryption is enabled.
    // Let's finish the job at the second attempt and verify its completion.
    app.waitForState(jobAttempt2, JobState.RUNNING);
    tasks = jobAttempt2.getTasks().values().iterator();
    mapper = tasks.next();
    Task reducer = tasks.next();
    // finish the map task first
    app.waitForState(mapper, TaskState.RUNNING);
    mapAttempt = mapper.getAttempts().values().iterator().next();
    app.waitForState(mapAttempt, TaskAttemptState.RUNNING);
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapAttempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(mapper, TaskState.SUCCEEDED);
    // then finish the reduce task
    TaskAttempt redAttempt = reducer.getAttempts().values().iterator().next();
    app.waitForState(redAttempt, TaskAttemptState.RUNNING);
    app.getContext().getEventHandler().handle(new TaskAttemptEvent(redAttempt.getID(), TaskAttemptEventType.TA_DONE));
    app.waitForState(reducer, TaskState.SUCCEEDED);
    // verify that the job succeeds at the 2rd attempt
    app.waitForState(jobAttempt2, JobState.SUCCEEDED);
}