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

Example 66 with Job

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

the class JobHistoryEventHandler method serviceInit.

/* (non-Javadoc)
   * @see org.apache.hadoop.yarn.service.AbstractService#init(org.
   * apache.hadoop.conf.Configuration)
   * Initializes the FileSystem and Path objects for the log and done directories.
   * Creates these directories if they do not already exist.
   */
@Override
protected void serviceInit(Configuration conf) throws Exception {
    String jobId = TypeConverter.fromYarn(context.getApplicationID()).toString();
    String stagingDirStr = null;
    String doneDirStr = null;
    String userDoneDirStr = null;
    try {
        stagingDirStr = JobHistoryUtils.getConfiguredHistoryStagingDirPrefix(conf, jobId);
        doneDirStr = JobHistoryUtils.getConfiguredHistoryIntermediateDoneDirPrefix(conf);
        userDoneDirStr = JobHistoryUtils.getHistoryIntermediateDoneDirForUser(conf);
    } catch (IOException e) {
        LOG.error("Failed while getting the configured log directories", e);
        throw new YarnRuntimeException(e);
    }
    //Check for the existence of the history staging dir. Maybe create it. 
    try {
        stagingDirPath = FileContext.getFileContext(conf).makeQualified(new Path(stagingDirStr));
        stagingDirFS = FileSystem.get(stagingDirPath.toUri(), conf);
        mkdir(stagingDirFS, stagingDirPath, new FsPermission(JobHistoryUtils.HISTORY_STAGING_DIR_PERMISSIONS));
    } catch (IOException e) {
        LOG.error("Failed while checking for/creating  history staging path: [" + stagingDirPath + "]", e);
        throw new YarnRuntimeException(e);
    }
    //Check for the existence of intermediate done dir.
    Path doneDirPath = null;
    try {
        doneDirPath = FileContext.getFileContext(conf).makeQualified(new Path(doneDirStr));
        doneDirFS = FileSystem.get(doneDirPath.toUri(), conf);
        // created by the JobHistoryServer or as part of deployment.
        if (!doneDirFS.exists(doneDirPath)) {
            if (JobHistoryUtils.shouldCreateNonUserDirectory(conf)) {
                LOG.info("Creating intermediate history logDir: [" + doneDirPath + "] + based on conf. Should ideally be created by the JobHistoryServer: " + MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR);
                mkdir(doneDirFS, doneDirPath, new FsPermission(JobHistoryUtils.HISTORY_INTERMEDIATE_DONE_DIR_PERMISSIONS.toShort()));
            // TODO Temporary toShort till new FsPermission(FsPermissions)
            // respects
            // sticky
            } else {
                String message = "Not creating intermediate history logDir: [" + doneDirPath + "] based on conf: " + MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR + ". Either set to true or pre-create this directory with" + " appropriate permissions";
                LOG.error(message);
                throw new YarnRuntimeException(message);
            }
        }
    } catch (IOException e) {
        LOG.error("Failed checking for the existance of history intermediate " + "done directory: [" + doneDirPath + "]");
        throw new YarnRuntimeException(e);
    }
    //Check/create user directory under intermediate done dir.
    try {
        doneDirPrefixPath = FileContext.getFileContext(conf).makeQualified(new Path(userDoneDirStr));
        mkdir(doneDirFS, doneDirPrefixPath, new FsPermission(JobHistoryUtils.HISTORY_INTERMEDIATE_USER_DIR_PERMISSIONS));
    } catch (IOException e) {
        LOG.error("Error creating user intermediate history done directory: [ " + doneDirPrefixPath + "]", e);
        throw new YarnRuntimeException(e);
    }
    // Maximum number of unflushed completion-events that can stay in the queue
    // before flush kicks in.
    maxUnflushedCompletionEvents = conf.getInt(MRJobConfig.MR_AM_HISTORY_MAX_UNFLUSHED_COMPLETE_EVENTS, MRJobConfig.DEFAULT_MR_AM_HISTORY_MAX_UNFLUSHED_COMPLETE_EVENTS);
    // We want to cut down flushes after job completes so as to write quicker,
    // so we increase maxUnflushedEvents post Job completion by using the
    // following multiplier.
    postJobCompletionMultiplier = conf.getInt(MRJobConfig.MR_AM_HISTORY_JOB_COMPLETE_UNFLUSHED_MULTIPLIER, MRJobConfig.DEFAULT_MR_AM_HISTORY_JOB_COMPLETE_UNFLUSHED_MULTIPLIER);
    // Max time until which flush doesn't take place.
    flushTimeout = conf.getLong(MRJobConfig.MR_AM_HISTORY_COMPLETE_EVENT_FLUSH_TIMEOUT_MS, MRJobConfig.DEFAULT_MR_AM_HISTORY_COMPLETE_EVENT_FLUSH_TIMEOUT_MS);
    minQueueSizeForBatchingFlushes = conf.getInt(MRJobConfig.MR_AM_HISTORY_USE_BATCHED_FLUSH_QUEUE_SIZE_THRESHOLD, MRJobConfig.DEFAULT_MR_AM_HISTORY_USE_BATCHED_FLUSH_QUEUE_SIZE_THRESHOLD);
    // configuration status: off, on_with_v1 or on_with_v2.
    if (conf.getBoolean(MRJobConfig.MAPREDUCE_JOB_EMIT_TIMELINE_DATA, MRJobConfig.DEFAULT_MAPREDUCE_JOB_EMIT_TIMELINE_DATA)) {
        LOG.info("Emitting job history data to the timeline service is enabled");
        if (YarnConfiguration.timelineServiceEnabled(conf)) {
            boolean timelineServiceV2Enabled = ((int) YarnConfiguration.getTimelineServiceVersion(conf) == 2);
            if (timelineServiceV2Enabled) {
                timelineV2Client = ((MRAppMaster.RunningAppContext) context).getTimelineV2Client();
                timelineV2Client.init(conf);
            } else {
                timelineClient = ((MRAppMaster.RunningAppContext) context).getTimelineClient();
                timelineClient.init(conf);
            }
            LOG.info("Timeline service is enabled; version: " + YarnConfiguration.getTimelineServiceVersion(conf));
        } else {
            LOG.info("Timeline service is not enabled");
        }
    } else {
        LOG.info("Emitting job history data to the timeline server is not " + "enabled");
    }
    // Flag for setting
    String jhistFormat = conf.get(JHAdminConfig.MR_HS_JHIST_FORMAT, JHAdminConfig.DEFAULT_MR_HS_JHIST_FORMAT);
    if (jhistFormat.equals("json")) {
        jhistMode = EventWriter.WriteMode.JSON;
    } else if (jhistFormat.equals("binary")) {
        jhistMode = EventWriter.WriteMode.BINARY;
    } else {
        LOG.warn("Unrecognized value '" + jhistFormat + "' for property " + JHAdminConfig.MR_HS_JHIST_FORMAT + ".  Valid values are " + "'json' or 'binary'.  Falling back to default value '" + JHAdminConfig.DEFAULT_MR_HS_JHIST_FORMAT + "'.");
    }
    super.serviceInit(conf);
}

Example 67 with Job

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

the class JobHistoryEventHandler method serviceStop.

@Override
protected void serviceStop() throws Exception {
    LOG.info("Stopping JobHistoryEventHandler. " + "Size of the outstanding queue size is " + eventQueue.size());
    stopped = true;
    //do not interrupt while event handling is in progress
    synchronized (lock) {
        if (eventHandlingThread != null) {
            LOG.debug("Interrupting Event Handling thread");
            eventHandlingThread.interrupt();
        } else {
            LOG.debug("Null event handling thread");
        }
    }
    try {
        if (eventHandlingThread != null) {
            LOG.debug("Waiting for Event Handling thread to complete");
            eventHandlingThread.join();
        }
    } catch (InterruptedException ie) {
        LOG.info("Interrupted Exception while stopping", ie);
    }
    // the metaInfo object is wrapped up.
    for (MetaInfo mi : fileMap.values()) {
        try {
            if (LOG.isDebugEnabled()) {
                LOG.debug("Shutting down timer for " + mi);
            }
            mi.shutDownTimer();
        } catch (IOException e) {
            LOG.info("Exception while canceling delayed flush timer. " + "Likely caused by a failed flush " + e.getMessage());
        }
    }
    //write all the events remaining in queue
    Iterator<JobHistoryEvent> it = eventQueue.iterator();
    while (it.hasNext()) {
        JobHistoryEvent ev = it.next();
        LOG.info("In stop, writing event " + ev.getType());
        handleEvent(ev);
    }
    // closed their event writers
    if (forceJobCompletion) {
        for (Map.Entry<JobId, MetaInfo> jobIt : fileMap.entrySet()) {
            JobId toClose = jobIt.getKey();
            MetaInfo mi = jobIt.getValue();
            if (mi != null && mi.isWriterActive()) {
                LOG.warn("Found jobId " + toClose + " to have not been closed. Will close");
                //Create a JobFinishEvent so that it is written to the job history
                final Job job = context.getJob(toClose);
                JobUnsuccessfulCompletionEvent jucEvent = new JobUnsuccessfulCompletionEvent(TypeConverter.fromYarn(toClose), System.currentTimeMillis(), job.getCompletedMaps(), job.getCompletedReduces(), createJobStateForJobUnsuccessfulCompletionEvent(mi.getForcedJobStateOnShutDown()), job.getDiagnostics());
                JobHistoryEvent jfEvent = new JobHistoryEvent(toClose, jucEvent);
                //Bypass the queue mechanism which might wait. Call the method directly
                handleEvent(jfEvent);
            }
        }
    }
    //close all file handles
    for (MetaInfo mi : fileMap.values()) {
        try {
            mi.closeWriter();
        } catch (IOException e) {
            LOG.info("Exception while closing file " + e.getMessage());
        }
    }
    if (timelineClient != null) {
        timelineClient.stop();
    } else if (timelineV2Client != null) {
        timelineV2Client.stop();
    }
    LOG.info("Stopped JobHistoryEventHandler. super.stop()");
    super.serviceStop();
}

Example 68 with Job

use of org.apache.hadoop.mapreduce.v2.app.job.Job 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);
}

Example 69 with Job

use of org.apache.hadoop.mapreduce.v2.app.job.Job 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();
    }
}

Example 70 with Job

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

the class TestFail method testMapFailureMaxPercent.

@Test
public void testMapFailureMaxPercent() throws Exception {
    MRApp app = new MockFirstFailingTaskMRApp(4, 0);
    Configuration conf = new Configuration();
    //reduce the no of attempts so test run faster
    conf.setInt(MRJobConfig.MAP_MAX_ATTEMPTS, 2);
    conf.setInt(MRJobConfig.REDUCE_MAX_ATTEMPTS, 1);
    conf.setInt(MRJobConfig.MAP_FAILURES_MAX_PERCENT, 20);
    conf.setInt(MRJobConfig.MAP_MAX_ATTEMPTS, 1);
    Job job = app.submit(conf);
    app.waitForState(job, JobState.FAILED);
    //setting the failure percentage to 25% (1/4 is 25) will
    //make the Job successful
    app = new MockFirstFailingTaskMRApp(4, 0);
    conf = new Configuration();
    //reduce the no of attempts so test run faster
    conf.setInt(MRJobConfig.MAP_MAX_ATTEMPTS, 2);
    conf.setInt(MRJobConfig.REDUCE_MAX_ATTEMPTS, 1);
    conf.setInt(MRJobConfig.MAP_FAILURES_MAX_PERCENT, 25);
    conf.setInt(MRJobConfig.MAP_MAX_ATTEMPTS, 1);
    job = app.submit(conf);
    app.waitForState(job, JobState.SUCCEEDED);
}