Examples with Lifecycle io.mantisrx.runtime.lifecycle.Lifecycle used on opensource projects

Example 1 with Lifecycle

use of io.mantisrx.runtime.lifecycle.Lifecycle in project mantis by Netflix.

the class WorkerExecutionOperationsNetworkStage method executeStage.

@SuppressWarnings({ "rawtypes", "unchecked" })
@Override
public void executeStage(final ExecutionDetails setup) {
    ExecuteStageRequest executionRequest = setup.getExecuteStageRequest().getRequest();
    // Initialize the schedulingInfo observable for current job and mark it shareable to be reused by anyone interested in this data.
    // Observable<JobSchedulingInfo> selfSchedulingInfo = mantisMasterApi.schedulingChanges(executionRequest.getJobId()).switchMap((e) -> Observable.just(e).repeatWhen(x -> x.delay(5 , TimeUnit.SECONDS))).subscribeOn(Schedulers.io()).share();
    Observable<JobSchedulingInfo> selfSchedulingInfo = mantisMasterApi.schedulingChanges(executionRequest.getJobId()).subscribeOn(Schedulers.io()).share();
    WorkerInfo workerInfo = generateWorkerInfo(executionRequest.getJobName(), executionRequest.getJobId(), executionRequest.getStage(), executionRequest.getWorkerIndex(), executionRequest.getWorkerNumber(), executionRequest.getDurationType(), "host", executionRequest.getWorkerPorts());
    final Observable<Integer> sourceStageTotalWorkersObs = createSourceStageTotalWorkersObservable(selfSchedulingInfo);
    RunningWorker.Builder rwBuilder = new RunningWorker.Builder().job(setup.getMantisJob()).schedulingInfo(executionRequest.getSchedulingInfo()).stageTotalWorkersObservable(sourceStageTotalWorkersObs).jobName(executionRequest.getJobName()).stageNum(executionRequest.getStage()).workerIndex(executionRequest.getWorkerIndex()).workerNum(executionRequest.getWorkerNumber()).totalStages(executionRequest.getTotalNumStages()).metricsPort(executionRequest.getMetricsPort()).ports(executionRequest.getPorts().iterator()).jobStatusObserver(setup.getStatus()).requestSubject(setup.getExecuteStageRequest().getRequestSubject()).workerInfo(workerInfo).vmTaskStatusObservable(vmTaskStatusObserver).hasJobMaster(executionRequest.getHasJobMaster()).jobId(executionRequest.getJobId());
    if (executionRequest.getStage() == 0) {
        rwBuilder = rwBuilder.stage(new JobMasterStageConfig("jobmasterconfig"));
    } else {
        rwBuilder = rwBuilder.stage((StageConfig) setup.getMantisJob().getStages().get(executionRequest.getStage() - 1));
    }
    final RunningWorker rw = rwBuilder.build();
    AtomicReference<SubscriptionStateHandler> subscriptionStateHandlerRef = new AtomicReference<>();
    if (rw.getStageNum() == rw.getTotalStagesNet()) {
        // set up subscription state handler only for sink (last) stage
        subscriptionStateHandlerRef.set(setupSubscriptionStateHandler(setup.getExecuteStageRequest().getRequest().getJobId(), mantisMasterApi, setup.getExecuteStageRequest().getRequest().getSubscriptionTimeoutSecs(), setup.getExecuteStageRequest().getRequest().getMinRuntimeSecs()));
    }
    logger.info("Running worker info: " + rw);
    rw.signalStartedInitiated();
    try {
        logger.info(">>>>>>>>>>>>>>>>Calling lifecycle.startup()");
        Lifecycle lifecycle = rw.getJob().getLifecycle();
        lifecycle.startup();
        ServiceLocator serviceLocator = lifecycle.getServiceLocator();
        if (lookupSpectatorRegistry) {
            try {
                final Registry spectatorRegistry = serviceLocator.service(Registry.class);
                SpectatorRegistryFactory.setRegistry(spectatorRegistry);
            } catch (Throwable t) {
                logger.error("failed to init spectator registry using service locator, falling back to {}", SpectatorRegistryFactory.getRegistry().getClass().getCanonicalName());
            }
        }
        // create job context
        Parameters parameters = ParameterUtils.createContextParameters(rw.getJob().getParameterDefinitions(), setup.getParameters());
        final Context context = generateContext(parameters, serviceLocator, workerInfo, MetricsRegistry.getInstance(), () -> {
            rw.signalCompleted();
            // wait for completion signal to go to the master and us getting killed. Upon timeout, exit.
            try {
                Thread.sleep(60000);
            } catch (InterruptedException ie) {
                logger.warn("Unexpected exception sleeping: " + ie.getMessage());
            }
            System.exit(0);
        }, createWorkerMapObservable(selfSchedulingInfo, executionRequest.getJobName(), executionRequest.getJobId(), executionRequest.getDurationType()));
        // context.setPrevStageCompletedObservable(createPrevStageCompletedObservable(selfSchedulingInfo, rw.getJobId(), rw.getStageNum()));
        rw.setContext(context);
        // setup heartbeats
        heartbeatRef.set(new Heartbeat(rw.getJobId(), rw.getStageNum(), rw.getWorkerIndex(), rw.getWorkerNum()));
        final double networkMbps = executionRequest.getSchedulingInfo().forStage(rw.getStageNum()).getMachineDefinition().getNetworkMbps();
        startSendingHeartbeats(rw.getJobStatus(), new WorkerId(executionRequest.getJobId(), executionRequest.getWorkerIndex(), executionRequest.getWorkerNumber()).getId(), networkMbps);
        // execute stage
        if (rw.getStageNum() == 0) {
            logger.info("JobId: " + rw.getJobId() + ", executing Job Master");
            final AutoScaleMetricsConfig autoScaleMetricsConfig = new AutoScaleMetricsConfig();
            // Temporary workaround to enable auto-scaling by custom metric in Job Master. This will be revisited to get the entire autoscaling config
            // for a job as a System parameter in the JobMaster
            final String autoScaleMetricString = (String) parameters.get(JOB_MASTER_AUTOSCALE_METRIC_SYSTEM_PARAM, "");
            if (!Strings.isNullOrEmpty(autoScaleMetricString)) {
                final List<String> tokens = Splitter.on("::").omitEmptyStrings().trimResults().splitToList(autoScaleMetricString);
                if (tokens.size() == 3) {
                    final String metricGroup = tokens.get(0);
                    final String metricName = tokens.get(1);
                    final String algo = tokens.get(2);
                    try {
                        final AutoScaleMetricsConfig.AggregationAlgo aggregationAlgo = AutoScaleMetricsConfig.AggregationAlgo.valueOf(algo);
                        logger.info("registered UserDefined auto scale metric {}:{} algo {}", metricGroup, metricName, aggregationAlgo);
                        autoScaleMetricsConfig.addUserDefinedMetric(metricGroup, metricName, aggregationAlgo);
                    } catch (IllegalArgumentException e) {
                        final String errorMsg = String.format("ERROR: Invalid algorithm value %s for param %s (algo should be one of %s)", autoScaleMetricsConfig, JOB_MASTER_AUTOSCALE_METRIC_SYSTEM_PARAM, Arrays.stream(AutoScaleMetricsConfig.AggregationAlgo.values()).map(a -> a.name()).collect(Collectors.toList()));
                        logger.error(errorMsg);
                        throw new RuntimeException(errorMsg);
                    }
                } else {
                    final String errorMsg = String.format("ERROR: Invalid value %s for param %s", autoScaleMetricString, JOB_MASTER_AUTOSCALE_METRIC_SYSTEM_PARAM);
                    logger.error(errorMsg);
                    throw new RuntimeException(errorMsg);
                }
            } else {
                logger.info("param {} is null or empty", JOB_MASTER_AUTOSCALE_METRIC_SYSTEM_PARAM);
            }
            final JobMasterService jobMasterService = new JobMasterService(rw.getJobId(), rw.getSchedulingInfo(), workerMetricsClient, autoScaleMetricsConfig, mantisMasterApi, rw.getContext(), rw.getOnCompleteCallback(), rw.getOnErrorCallback(), rw.getOnTerminateCallback());
            jobMasterService.start();
            signalStarted(rw, subscriptionStateHandlerRef);
            // block until worker terminates
            rw.waitUntilTerminate();
        } else if (rw.getStageNum() == 1 && rw.getTotalStagesNet() == 1) {
            logger.info("JobId: " + rw.getJobId() + ", single stage job, executing entire job");
            // single stage, execute entire job on this machine
            PortSelector portSelector = new PortSelector() {

                @Override
                public int acquirePort() {
                    return rw.getPorts().next();
                }
            };
            RxMetrics rxMetrics = new RxMetrics();
            StageExecutors.executeSingleStageJob(rw.getJob().getSource(), rw.getStage(), rw.getJob().getSink(), portSelector, rxMetrics, rw.getContext(), rw.getOnTerminateCallback(), rw.getWorkerIndex(), rw.getSourceStageTotalWorkersObservable(), onSinkSubscribe, onSinkUnsubscribe, rw.getOnCompleteCallback(), rw.getOnErrorCallback());
            signalStarted(rw, subscriptionStateHandlerRef);
            // block until worker terminates
            rw.waitUntilTerminate();
        } else {
            logger.info("JobId: " + rw.getJobId() + ", executing a multi-stage job, stage: " + rw.getStageNum());
            if (rw.getStageNum() == 1) {
                // execute source stage
                String remoteObservableName = rw.getJobId() + "_" + rw.getStageNum();
                StageSchedulingInfo currentStageSchedulingInfo = rw.getSchedulingInfo().forStage(1);
                WorkerPublisherRemoteObservable publisher = new WorkerPublisherRemoteObservable<>(rw.getPorts().next(), remoteObservableName, numWorkersAtStage(selfSchedulingInfo, rw.getJobId(), rw.getStageNum() + 1), rw.getJobName());
                StageExecutors.executeSource(rw.getWorkerIndex(), rw.getJob().getSource(), rw.getStage(), publisher, rw.getContext(), rw.getSourceStageTotalWorkersObservable());
                logger.info("JobId: " + rw.getJobId() + " stage: " + rw.getStageNum() + ", serving remote observable for source with name: " + remoteObservableName);
                RemoteRxServer server = publisher.getServer();
                RxMetrics rxMetrics = server.getMetrics();
                MetricsRegistry.getInstance().registerAndGet(rxMetrics.getCountersAndGauges());
                signalStarted(rw, subscriptionStateHandlerRef);
                logger.info("JobId: " + rw.getJobId() + " stage: " + rw.getStageNum() + ", blocking until source observable completes");
                server.blockUntilServerShutdown();
            } else {
                // execute intermediate stage or last stage plus sink
                executeNonSourceStage(selfSchedulingInfo, rw, subscriptionStateHandlerRef);
            }
        }
        logger.info("Calling lifecycle.shutdown()");
        lifecycle.shutdown();
    } catch (Throwable t) {
        rw.signalFailed(t);
        shutdownStage();
    }
}

Example 2 with Lifecycle

use of io.mantisrx.runtime.lifecycle.Lifecycle in project mantis by Netflix.

the class LocalJobExecutorNetworked method execute.

@SuppressWarnings({ "rawtypes", "unchecked" })
public static void execute(Job job, SchedulingInfo schedulingInfo, Parameter... parameters) throws IllegalMantisJobException {
    // validate job
    try {
        new ValidateJob(job).execute();
    } catch (CommandException e) {
        throw new IllegalMantisJobException(e);
    }
    // execute job
    List<StageConfig> stages = job.getStages();
    final SourceHolder source = job.getSource();
    final SinkHolder sink = job.getSink();
    final PortSelector portSelector = new PortSelectorInRange(8000, 9000);
    // register netty metrics
    RxNetty.useMetricListenersFactory(new MantisNettyEventsListenerFactory());
    // start our metrics server
    MetricsServer metricsServer = new MetricsServer(portSelector.acquirePort(), 1, Collections.EMPTY_MAP);
    metricsServer.start();
    Lifecycle lifecycle = job.getLifecycle();
    lifecycle.startup();
    // create job context
    Map parameterDefinitions = job.getParameterDefinitions();
    final String user = Optional.ofNullable(System.getenv("USER")).orElse("userUnknown");
    String jobId = String.format("localJob-%s-%d", user, (int) (Math.random() * 10000));
    logger.info("jobID {}", jobId);
    final ServiceLocator serviceLocator = lifecycle.getServiceLocator();
    int numInstances = schedulingInfo.forStage(1).getNumberOfInstances();
    BehaviorSubject<Integer> workersInStageOneObservable = BehaviorSubject.create(numInstances);
    BehaviorSubject<WorkerMap> workerMapObservable = BehaviorSubject.create();
    if (stages.size() == 1) {
        // single stage job
        final StageConfig stage = stages.get(0);
        // use latch to wait for all instances to complete
        final CountDownLatch waitUntilAllCompleted = new CountDownLatch(numInstances);
        Action0 countDownLatchOnComplete = new Action0() {

            @Override
            public void call() {
                waitUntilAllCompleted.countDown();
            }
        };
        Action0 nullOnCompleted = new Action0() {

            @Override
            public void call() {
            }
        };
        Action1<Throwable> nullOnError = new Action1<Throwable>() {

            @Override
            public void call(Throwable t) {
            }
        };
        Map<Integer, List<WorkerInfo>> workerInfoMap = new HashMap<>();
        List<WorkerInfo> workerInfoList = new ArrayList<>();
        // run for num of instances
        for (int i = 0; i < numInstances; i++) {
            WorkerPorts workerPorts = new WorkerPorts(portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort());
            WorkerInfo workerInfo = new WorkerInfo(jobId, jobId, 1, i, i + 1, MantisJobDurationType.Perpetual, "localhost", workerPorts);
            workerInfoList.add(workerInfo);
            Context context = new Context(ParameterUtils.createContextParameters(parameterDefinitions, parameters), lifecycle.getServiceLocator(), // new WorkerInfo(jobId, jobId, 1, i, i, MantisJobDurationType.Perpetual, "localhost", new ArrayList<>(),-1,-1),
            workerInfo, MetricsRegistry.getInstance(), () -> {
                System.exit(0);
            }, workerMapObservable);
            // workers for stage 1
            workerInfoMap.put(1, workerInfoList);
            workerMapObservable.onNext(new WorkerMap(workerInfoMap));
            StageExecutors.executeSingleStageJob(source, stage, sink, () -> workerInfo.getWorkerPorts().getSinkPort(), new RxMetrics(), context, countDownLatchOnComplete, i, workersInStageOneObservable, null, null, nullOnCompleted, nullOnError);
        }
        // wait for all instances to complete
        try {
            waitUntilAllCompleted.await();
        } catch (InterruptedException e) {
            throw new RuntimeException(e);
        }
    } else {
        // multi-stage job
        int workerNumber = 0;
        // start source stages
        StageConfig currentStage = stages.get(0);
        StageConfig previousStage = null;
        StageSchedulingInfo currentStageScalingInfo = schedulingInfo.forStage(1);
        StageSchedulingInfo nextStageScalingInfo = schedulingInfo.forStage(2);
        // num ports
        int[] previousPorts = new int[currentStageScalingInfo.getNumberOfInstances()];
        Map<Integer, List<WorkerInfo>> workerInfoMap = new HashMap<>();
        List<WorkerInfo> workerInfoList = new ArrayList<>();
        for (int i = 0; i < currentStageScalingInfo.getNumberOfInstances(); i++) {
            WorkerPorts workerPorts = new WorkerPorts(portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort());
            WorkerInfo workerInfo = new WorkerInfo(jobId, jobId, 1, i, i + 1, MantisJobDurationType.Perpetual, "localhost", workerPorts);
            workerInfoList.add(workerInfo);
            // int sourcePort = portSelector.acquirePort();
            int sourcePort = workerInfo.getWorkerPorts().getSinkPort();
            previousPorts[i] = sourcePort;
            Context context = new Context(ParameterUtils.createContextParameters(parameterDefinitions, parameters), serviceLocator, workerInfo, MetricsRegistry.getInstance(), nullAction, workerMapObservable);
            startSource(i, sourcePort, nextStageScalingInfo.getNumberOfInstances(), job.getSource(), currentStage, context, workersInStageOneObservable);
        }
        // workers for stage 1
        workerInfoMap.put(1, workerInfoList);
        workerMapObservable.onNext(new WorkerMap(workerInfoMap));
        // start intermediate stages, all but last stage
        for (int i = 1; i < stages.size() - 1; i++) {
            previousStage = currentStage;
            StageSchedulingInfo previousStageScalingInfo = schedulingInfo.forStage(i);
            // stages indexed starting at 1
            currentStageScalingInfo = schedulingInfo.forStage(i + 1);
            currentStage = stages.get(i);
            // stages indexed starting at 1
            nextStageScalingInfo = schedulingInfo.forStage(i + 2);
            int[] currentPorts = new int[currentStageScalingInfo.getNumberOfInstances()];
            workerInfoList = new ArrayList<>();
            for (int j = 0; j < currentStageScalingInfo.getNumberOfInstances(); j++) {
                WorkerPorts workerPorts = new WorkerPorts(portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort());
                WorkerInfo workerInfo = new WorkerInfo(jobId, jobId, i + 1, j, workerNumber++, MantisJobDurationType.Perpetual, "localhost", workerPorts);
                workerInfoList.add(workerInfo);
                // int port = portSelector.acquirePort();
                int port = workerInfo.getWorkerPorts().getSinkPort();
                currentPorts[j] = port;
                Context context = new Context(ParameterUtils.createContextParameters(parameterDefinitions, parameters), serviceLocator, workerInfo, MetricsRegistry.getInstance(), nullAction, workerMapObservable);
                startIntermediate(previousPorts, port, currentStage, context, j, nextStageScalingInfo.getNumberOfInstances(), i, previousStageScalingInfo.getNumberOfInstances());
            }
            // workers for current stage
            workerInfoMap.put(i + 1, workerInfoList);
            workerMapObservable.onNext(new WorkerMap(workerInfoMap));
            previousPorts = currentPorts;
        }
        // start sink stage
        StageSchedulingInfo previousStageScalingInfo = schedulingInfo.forStage(stages.size() - 1);
        previousStage = stages.get(stages.size() - 2);
        currentStage = stages.get(stages.size() - 1);
        currentStageScalingInfo = schedulingInfo.forStage(stages.size());
        numInstances = currentStageScalingInfo.getNumberOfInstances();
        // use latch to wait for all instances to complete
        final CountDownLatch waitUntilAllCompleted = new CountDownLatch(numInstances);
        Action0 countDownLatchOnTerminated = new Action0() {

            @Override
            public void call() {
                waitUntilAllCompleted.countDown();
            }
        };
        Action0 nullOnCompleted = new Action0() {

            @Override
            public void call() {
            }
        };
        Action1<Throwable> nullOnError = new Action1<Throwable>() {

            @Override
            public void call(Throwable t) {
            }
        };
        workerInfoList = new ArrayList<>();
        for (int i = 0; i < numInstances; i++) {
            WorkerPorts workerPorts = new WorkerPorts(portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort(), portSelector.acquirePort());
            WorkerInfo workerInfo = new WorkerInfo(jobId, jobId, stages.size(), i, workerNumber++, MantisJobDurationType.Perpetual, "localhost", workerPorts);
            workerInfoList.add(workerInfo);
            Context context = new Context(ParameterUtils.createContextParameters(parameterDefinitions, parameters), serviceLocator, workerInfo, MetricsRegistry.getInstance(), nullAction, workerMapObservable);
            startSink(previousStage, previousPorts, currentStage, () -> workerInfo.getWorkerPorts().getSinkPort(), sink, context, countDownLatchOnTerminated, nullOnCompleted, nullOnError, stages.size(), i, previousStageScalingInfo.getNumberOfInstances());
        }
        workerInfoMap.put(stages.size(), workerInfoList);
        workerMapObservable.onNext(new WorkerMap(workerInfoMap));
        // wait for all instances to complete
        try {
            waitUntilAllCompleted.await();
        } catch (InterruptedException e) {
            throw new RuntimeException(e);
        }
    }
    lifecycle.shutdown();
    metricsServer.shutdown();
}