Search in sources :

Example 6 with MRAppMaster

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

the class TestStagingCleanup method testDeletionofStaging.

@Test
public void testDeletionofStaging() throws IOException {
    conf.set(MRJobConfig.MAPREDUCE_JOB_DIR, stagingJobDir);
    fs = mock(FileSystem.class);
    when(fs.delete(any(Path.class), anyBoolean())).thenReturn(true);
    //Staging Dir exists
    String user = UserGroupInformation.getCurrentUser().getShortUserName();
    Path stagingDir = MRApps.getStagingAreaDir(conf, user);
    when(fs.exists(stagingDir)).thenReturn(true);
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 0);
    ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
    JobId jobid = recordFactory.newRecordInstance(JobId.class);
    jobid.setAppId(appId);
    ContainerAllocator mockAlloc = mock(ContainerAllocator.class);
    Assert.assertTrue(MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS > 1);
    MRAppMaster appMaster = new TestMRApp(attemptId, mockAlloc, JobStateInternal.RUNNING, MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS);
    appMaster.init(conf);
    appMaster.start();
    appMaster.shutDownJob();
    //test whether notifyIsLastAMRetry called
    Assert.assertEquals(true, ((TestMRApp) appMaster).getTestIsLastAMRetry());
    verify(fs).delete(stagingJobPath, true);
}
Also used : Path(org.apache.hadoop.fs.Path) FileSystem(org.apache.hadoop.fs.FileSystem) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) JobId(org.apache.hadoop.mapreduce.v2.api.records.JobId) ContainerAllocator(org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator) Test(org.junit.Test)

Example 7 with MRAppMaster

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

the class TestStagingCleanup method testPreservePatternMatchedStaging.

@Test
public void testPreservePatternMatchedStaging() throws IOException {
    conf.set(MRJobConfig.MAPREDUCE_JOB_DIR, stagingJobDir);
    // The staging files that are matched to the pattern
    // should not be deleted
    conf.set(MRJobConfig.PRESERVE_FILES_PATTERN, "JobDir");
    fs = mock(FileSystem.class);
    when(fs.delete(any(Path.class), anyBoolean())).thenReturn(true);
    //Staging Dir exists
    String user = UserGroupInformation.getCurrentUser().getShortUserName();
    Path stagingDir = MRApps.getStagingAreaDir(conf, user);
    when(fs.exists(stagingDir)).thenReturn(true);
    ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 0);
    ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
    JobId jobid = recordFactory.newRecordInstance(JobId.class);
    jobid.setAppId(appId);
    ContainerAllocator mockAlloc = mock(ContainerAllocator.class);
    Assert.assertTrue(MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS > 1);
    MRAppMaster appMaster = new TestMRApp(attemptId, mockAlloc, JobStateInternal.RUNNING, MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS);
    appMaster.init(conf);
    appMaster.start();
    appMaster.shutDownJob();
    //test whether notifyIsLastAMRetry called
    Assert.assertEquals(true, ((TestMRApp) appMaster).getTestIsLastAMRetry());
    verify(fs, times(0)).delete(stagingJobPath, true);
}
Also used : Path(org.apache.hadoop.fs.Path) FileSystem(org.apache.hadoop.fs.FileSystem) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId) JobId(org.apache.hadoop.mapreduce.v2.api.records.JobId) ContainerAllocator(org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator) Test(org.junit.Test)

Example 8 with MRAppMaster

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

the class MRAppMaster method serviceInit.

@Override
protected void serviceInit(final Configuration conf) throws Exception {
    // create the job classloader if enabled
    createJobClassLoader(conf);
    conf.setBoolean(Dispatcher.DISPATCHER_EXIT_ON_ERROR_KEY, true);
    initJobCredentialsAndUGI(conf);
    dispatcher = createDispatcher();
    addIfService(dispatcher);
    taskAttemptFinishingMonitor = createTaskAttemptFinishingMonitor(dispatcher.getEventHandler());
    addIfService(taskAttemptFinishingMonitor);
    context = new RunningAppContext(conf, taskAttemptFinishingMonitor);
    // Job name is the same as the app name util we support DAG of jobs
    // for an app later
    appName = conf.get(MRJobConfig.JOB_NAME, "<missing app name>");
    conf.setInt(MRJobConfig.APPLICATION_ATTEMPT_ID, appAttemptID.getAttemptId());
    newApiCommitter = false;
    jobId = MRBuilderUtils.newJobId(appAttemptID.getApplicationId(), appAttemptID.getApplicationId().getId());
    int numReduceTasks = conf.getInt(MRJobConfig.NUM_REDUCES, 0);
    if ((numReduceTasks > 0 && conf.getBoolean("mapred.reducer.new-api", false)) || (numReduceTasks == 0 && conf.getBoolean("mapred.mapper.new-api", false))) {
        newApiCommitter = true;
        LOG.info("Using mapred newApiCommitter.");
    }
    boolean copyHistory = false;
    committer = createOutputCommitter(conf);
    try {
        String user = UserGroupInformation.getCurrentUser().getShortUserName();
        Path stagingDir = MRApps.getStagingAreaDir(conf, user);
        FileSystem fs = getFileSystem(conf);
        boolean stagingExists = fs.exists(stagingDir);
        Path startCommitFile = MRApps.getStartJobCommitFile(conf, user, jobId);
        boolean commitStarted = fs.exists(startCommitFile);
        Path endCommitSuccessFile = MRApps.getEndJobCommitSuccessFile(conf, user, jobId);
        boolean commitSuccess = fs.exists(endCommitSuccessFile);
        Path endCommitFailureFile = MRApps.getEndJobCommitFailureFile(conf, user, jobId);
        boolean commitFailure = fs.exists(endCommitFailureFile);
        if (!stagingExists) {
            isLastAMRetry = true;
            LOG.info("Attempt num: " + appAttemptID.getAttemptId() + " is last retry: " + isLastAMRetry + " because the staging dir doesn't exist.");
            errorHappenedShutDown = true;
            forcedState = JobStateInternal.ERROR;
            shutDownMessage = "Staging dir does not exist " + stagingDir;
            LOG.fatal(shutDownMessage);
        } else if (commitStarted) {
            //A commit was started so this is the last time, we just need to know
            // what result we will use to notify, and how we will unregister
            errorHappenedShutDown = true;
            isLastAMRetry = true;
            LOG.info("Attempt num: " + appAttemptID.getAttemptId() + " is last retry: " + isLastAMRetry + " because a commit was started.");
            copyHistory = true;
            if (commitSuccess) {
                shutDownMessage = "Job commit succeeded in a prior MRAppMaster attempt " + "before it crashed. Recovering.";
                forcedState = JobStateInternal.SUCCEEDED;
            } else if (commitFailure) {
                shutDownMessage = "Job commit failed in a prior MRAppMaster attempt " + "before it crashed. Not retrying.";
                forcedState = JobStateInternal.FAILED;
            } else {
                if (isCommitJobRepeatable()) {
                    // cleanup previous half done commits if committer supports
                    // repeatable job commit.
                    errorHappenedShutDown = false;
                    cleanupInterruptedCommit(conf, fs, startCommitFile);
                } else {
                    //The commit is still pending, commit error
                    shutDownMessage = "Job commit from a prior MRAppMaster attempt is " + "potentially in progress. Preventing multiple commit executions";
                    forcedState = JobStateInternal.ERROR;
                }
            }
        }
    } catch (IOException e) {
        throw new YarnRuntimeException("Error while initializing", e);
    }
    if (errorHappenedShutDown) {
        NoopEventHandler eater = new NoopEventHandler();
        //We do not have a JobEventDispatcher in this path
        dispatcher.register(JobEventType.class, eater);
        EventHandler<JobHistoryEvent> historyService = null;
        if (copyHistory) {
            historyService = createJobHistoryHandler(context);
            dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, historyService);
        } else {
            dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, eater);
        }
        if (copyHistory) {
            // Now that there's a FINISHING state for application on RM to give AMs
            // plenty of time to clean up after unregister it's safe to clean staging
            // directory after unregistering with RM. So, we start the staging-dir
            // cleaner BEFORE the ContainerAllocator so that on shut-down,
            // ContainerAllocator unregisters first and then the staging-dir cleaner
            // deletes staging directory.
            addService(createStagingDirCleaningService());
        }
        // service to allocate containers from RM (if non-uber) or to fake it (uber)
        containerAllocator = createContainerAllocator(null, context);
        addIfService(containerAllocator);
        dispatcher.register(ContainerAllocator.EventType.class, containerAllocator);
        if (copyHistory) {
            // Add the JobHistoryEventHandler last so that it is properly stopped first.
            // This will guarantee that all history-events are flushed before AM goes
            // ahead with shutdown.
            // Note: Even though JobHistoryEventHandler is started last, if any
            // component creates a JobHistoryEvent in the meanwhile, it will be just be
            // queued inside the JobHistoryEventHandler 
            addIfService(historyService);
            JobHistoryCopyService cpHist = new JobHistoryCopyService(appAttemptID, dispatcher.getEventHandler());
            addIfService(cpHist);
        }
    } else {
        //service to handle requests from JobClient
        clientService = createClientService(context);
        // Init ClientService separately so that we stop it separately, since this
        // service needs to wait some time before it stops so clients can know the
        // final states
        clientService.init(conf);
        containerAllocator = createContainerAllocator(clientService, context);
        //service to handle the output committer
        committerEventHandler = createCommitterEventHandler(context, committer);
        addIfService(committerEventHandler);
        //policy handling preemption requests from RM
        callWithJobClassLoader(conf, new Action<Void>() {

            public Void call(Configuration conf) {
                preemptionPolicy = createPreemptionPolicy(conf);
                preemptionPolicy.init(context);
                return null;
            }
        });
        //service to handle requests to TaskUmbilicalProtocol
        taskAttemptListener = createTaskAttemptListener(context, preemptionPolicy);
        addIfService(taskAttemptListener);
        //service to log job history events
        EventHandler<JobHistoryEvent> historyService = createJobHistoryHandler(context);
        dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, historyService);
        this.jobEventDispatcher = new JobEventDispatcher();
        //register the event dispatchers
        dispatcher.register(JobEventType.class, jobEventDispatcher);
        dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
        dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
        dispatcher.register(CommitterEventType.class, committerEventHandler);
        if (conf.getBoolean(MRJobConfig.MAP_SPECULATIVE, false) || conf.getBoolean(MRJobConfig.REDUCE_SPECULATIVE, false)) {
            //optional service to speculate on task attempts' progress
            speculator = createSpeculator(conf, context);
            addIfService(speculator);
        }
        speculatorEventDispatcher = new SpeculatorEventDispatcher(conf);
        dispatcher.register(Speculator.EventType.class, speculatorEventDispatcher);
        // Now that there's a FINISHING state for application on RM to give AMs
        // plenty of time to clean up after unregister it's safe to clean staging
        // directory after unregistering with RM. So, we start the staging-dir
        // cleaner BEFORE the ContainerAllocator so that on shut-down,
        // ContainerAllocator unregisters first and then the staging-dir cleaner
        // deletes staging directory.
        addService(createStagingDirCleaningService());
        // service to allocate containers from RM (if non-uber) or to fake it (uber)
        addIfService(containerAllocator);
        dispatcher.register(ContainerAllocator.EventType.class, containerAllocator);
        // corresponding service to launch allocated containers via NodeManager
        containerLauncher = createContainerLauncher(context);
        addIfService(containerLauncher);
        dispatcher.register(ContainerLauncher.EventType.class, containerLauncher);
        // Add the JobHistoryEventHandler last so that it is properly stopped first.
        // This will guarantee that all history-events are flushed before AM goes
        // ahead with shutdown.
        // Note: Even though JobHistoryEventHandler is started last, if any
        // component creates a JobHistoryEvent in the meanwhile, it will be just be
        // queued inside the JobHistoryEventHandler 
        addIfService(historyService);
    }
    super.serviceInit(conf);
}
Also used : JobHistoryCopyService(org.apache.hadoop.mapreduce.jobhistory.JobHistoryCopyService) Configuration(org.apache.hadoop.conf.Configuration) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) JobHistoryEvent(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEvent) ContainerLauncher(org.apache.hadoop.mapreduce.v2.app.launcher.ContainerLauncher) LocalContainerLauncher(org.apache.hadoop.mapred.LocalContainerLauncher) FileSystem(org.apache.hadoop.fs.FileSystem) Path(org.apache.hadoop.fs.Path) IOException(java.io.IOException) ContainerAllocator(org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator) RMContainerAllocator(org.apache.hadoop.mapreduce.v2.app.rm.RMContainerAllocator) LocalContainerAllocator(org.apache.hadoop.mapreduce.v2.app.local.LocalContainerAllocator) DefaultSpeculator(org.apache.hadoop.mapreduce.v2.app.speculate.DefaultSpeculator) Speculator(org.apache.hadoop.mapreduce.v2.app.speculate.Speculator) YarnRuntimeException(org.apache.hadoop.yarn.exceptions.YarnRuntimeException)

Example 9 with MRAppMaster

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

the class MRAppMaster method serviceStart.

@SuppressWarnings("unchecked")
@Override
protected void serviceStart() throws Exception {
    amInfos = new LinkedList<AMInfo>();
    completedTasksFromPreviousRun = new HashMap<TaskId, TaskInfo>();
    processRecovery();
    // Current an AMInfo for the current AM generation.
    AMInfo amInfo = MRBuilderUtils.newAMInfo(appAttemptID, startTime, containerID, nmHost, nmPort, nmHttpPort);
    // /////////////////// Create the job itself.
    job = createJob(getConfig(), forcedState, shutDownMessage);
    // Send out an MR AM inited event for all previous AMs.
    for (AMInfo info : amInfos) {
        dispatcher.getEventHandler().handle(new JobHistoryEvent(job.getID(), new AMStartedEvent(info.getAppAttemptId(), info.getStartTime(), info.getContainerId(), info.getNodeManagerHost(), info.getNodeManagerPort(), info.getNodeManagerHttpPort(), appSubmitTime)));
    }
    // Send out an MR AM inited event for this AM.
    dispatcher.getEventHandler().handle(new JobHistoryEvent(job.getID(), new AMStartedEvent(amInfo.getAppAttemptId(), amInfo.getStartTime(), amInfo.getContainerId(), amInfo.getNodeManagerHost(), amInfo.getNodeManagerPort(), amInfo.getNodeManagerHttpPort(), this.forcedState == null ? null : this.forcedState.toString(), appSubmitTime)));
    amInfos.add(amInfo);
    // metrics system init is really init & start.
    // It's more test friendly to put it here.
    DefaultMetricsSystem.initialize("MRAppMaster");
    boolean initFailed = false;
    if (!errorHappenedShutDown) {
        // create a job event for job intialization
        JobEvent initJobEvent = new JobEvent(job.getID(), JobEventType.JOB_INIT);
        // Send init to the job (this does NOT trigger job execution)
        // This is a synchronous call, not an event through dispatcher. We want
        // job-init to be done completely here.
        jobEventDispatcher.handle(initJobEvent);
        // If job is still not initialized, an error happened during
        // initialization. Must complete starting all of the services so failure
        // events can be processed.
        initFailed = (((JobImpl) job).getInternalState() != JobStateInternal.INITED);
        if (job.isUber()) {
            speculatorEventDispatcher.disableSpeculation();
            LOG.info("MRAppMaster uberizing job " + job.getID() + " in local container (\"uber-AM\") on node " + nmHost + ":" + nmPort + ".");
        } else {
            // send init to speculator only for non-uber jobs. 
            // This won't yet start as dispatcher isn't started yet.
            dispatcher.getEventHandler().handle(new SpeculatorEvent(job.getID(), clock.getTime()));
            LOG.info("MRAppMaster launching normal, non-uberized, multi-container " + "job " + job.getID() + ".");
        }
        // Start ClientService here, since it's not initialized if
        // errorHappenedShutDown is true
        clientService.start();
    }
    //start all the components
    super.serviceStart();
    // finally set the job classloader
    MRApps.setClassLoader(jobClassLoader, getConfig());
    // set job classloader if configured
    Limits.init(getConfig());
    if (initFailed) {
        JobEvent initFailedEvent = new JobEvent(job.getID(), JobEventType.JOB_INIT_FAILED);
        jobEventDispatcher.handle(initFailedEvent);
    } else {
        // All components have started, start the job.
        startJobs();
    }
}
Also used : AMInfo(org.apache.hadoop.mapreduce.v2.api.records.AMInfo) TaskInfo(org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskInfo) AMStartedEvent(org.apache.hadoop.mapreduce.jobhistory.AMStartedEvent) TaskId(org.apache.hadoop.mapreduce.v2.api.records.TaskId) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) SpeculatorEvent(org.apache.hadoop.mapreduce.v2.app.speculate.SpeculatorEvent) JobHistoryEvent(org.apache.hadoop.mapreduce.jobhistory.JobHistoryEvent)

Example 10 with MRAppMaster

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

the class TestJobEndNotifier method testNotificationOnLastRetryUnregistrationFailure.

@Test
public void testNotificationOnLastRetryUnregistrationFailure() throws Exception {
    HttpServer2 server = startHttpServer();
    MRApp app = spy(new MRAppWithCustomContainerAllocator(2, 2, false, this.getClass().getName(), true, 2, false));
    // Currently, we will have isLastRetry always equals to false at beginning
    // of MRAppMaster, except staging area exists or commit already started at 
    // the beginning.
    // Now manually set isLastRetry to true and this should reset to false when
    // unregister failed.
    app.isLastAMRetry = true;
    doNothing().when(app).sysexit();
    JobConf conf = new JobConf();
    conf.set(JobContext.MR_JOB_END_NOTIFICATION_URL, JobEndServlet.baseUrl + "jobend?jobid=$jobId&status=$jobStatus");
    JobImpl job = (JobImpl) app.submit(conf);
    app.waitForState(job, JobState.RUNNING);
    app.getContext().getEventHandler().handle(new JobEvent(app.getJobId(), JobEventType.JOB_AM_REBOOT));
    app.waitForInternalState(job, JobStateInternal.REBOOT);
    // Now shutdown. User should see FAILED state.
    // Unregistration fails: isLastAMRetry is recalculated, this is
    ///reboot will stop service internally, we don't need to shutdown twice
    app.waitForServiceToStop(10000);
    Assert.assertFalse(app.isLastAMRetry());
    // Since it's not last retry, JobEndServlet didn't called
    Assert.assertEquals(0, JobEndServlet.calledTimes);
    Assert.assertNull(JobEndServlet.requestUri);
    Assert.assertNull(JobEndServlet.foundJobState);
    server.stop();
}
Also used : JobImpl(org.apache.hadoop.mapreduce.v2.app.job.impl.JobImpl) JobEvent(org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent) HttpServer2(org.apache.hadoop.http.HttpServer2) JobConf(org.apache.hadoop.mapred.JobConf) Test(org.junit.Test)

Aggregations

FileSystem (org.apache.hadoop.fs.FileSystem)13 Path (org.apache.hadoop.fs.Path)13 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)12 Test (org.junit.Test)11 JobId (org.apache.hadoop.mapreduce.v2.api.records.JobId)10 ContainerAllocator (org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator)10 ApplicationId (org.apache.hadoop.yarn.api.records.ApplicationId)9 IOException (java.io.IOException)4 JobConf (org.apache.hadoop.mapred.JobConf)4 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)3 JobHistoryEvent (org.apache.hadoop.mapreduce.jobhistory.JobHistoryEvent)2 JobEvent (org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent)2 VisibleForTesting (com.google.common.annotations.VisibleForTesting)1 Configuration (org.apache.hadoop.conf.Configuration)1 HttpServer2 (org.apache.hadoop.http.HttpServer2)1 LocalContainerLauncher (org.apache.hadoop.mapred.LocalContainerLauncher)1 AMStartedEvent (org.apache.hadoop.mapreduce.jobhistory.AMStartedEvent)1 JobHistoryCopyService (org.apache.hadoop.mapreduce.jobhistory.JobHistoryCopyService)1 TaskInfo (org.apache.hadoop.mapreduce.jobhistory.JobHistoryParser.TaskInfo)1 AMInfo (org.apache.hadoop.mapreduce.v2.api.records.AMInfo)1