use of io.reactivex.mantis.remote.observable.RemoteRxServer 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();
}
}
use of io.reactivex.mantis.remote.observable.RemoteRxServer in project mantis by Netflix.
the class StageExecutorsGroupByTest method testExecuteIntermediatStage.
@SuppressWarnings({ "rawtypes", "unchecked" })
@Test
public void testExecuteIntermediatStage() throws InterruptedException {
// Note, this test has a timing issue, client starts
// sending data before server is ready, resulting
// in a RST (connection reset by peer)
TestGroupByJob provider = new TestGroupByJob();
Job<Pair> job = provider.getJobInstance();
List<StageConfig<?, ?>> stages = job.getStages();
PortSelectorWithinRange portSelector = new PortSelectorWithinRange(8000, 9000);
final int publishPort = portSelector.acquirePort();
final int consumerPort = portSelector.acquirePort();
Observable<Observable<GroupedObservable<String, Integer>>> go = Observable.just(Observable.range(0, 10).groupBy(new Func1<Integer, String>() {
@Override
public String call(Integer t1) {
if ((t1 % 2) == 0) {
return "even";
} else {
return "odd";
}
}
}));
// mimic previous stage with a server
ServeGroupedObservable<String, Integer> config = new ServeGroupedObservable.Builder<String, Integer>().keyEncoder(Codecs.string()).valueEncoder(Codecs.integer()).observable(go).build();
RemoteRxServer server = new RemoteRxServer.Builder().addObservable(config).port(consumerPort).build();
server.start();
EndpointInjector staticEndpoints = new EndpointInjector() {
@Override
public Observable<EndpointChange> deltas() {
return Observable.create(new OnSubscribe<EndpointChange>() {
@Override
public void call(Subscriber<? super EndpointChange> subscriber) {
subscriber.onNext(new EndpointChange(EndpointChange.Type.add, new Endpoint("localhost", consumerPort, "0")));
subscriber.onNext(new EndpointChange(EndpointChange.Type.add, new Endpoint("localhost", consumerPort, "1")));
subscriber.onCompleted();
}
});
}
};
WorkerConsumer consumer = new WorkerConsumerRemoteObservable(null, staticEndpoints);
WorkerPublisher producer = new WorkerPublisherRemoteObservable(publishPort, null, Observable.just(1), null);
// execute source
StageExecutors.executeIntermediate(consumer, stages.get(1), producer, new Context());
ConnectToGroupedObservable<String, Integer> connectConfig = new ConnectToGroupedObservable.Builder<String, Integer>().host("localhost").port(publishPort).keyDecoder(Codecs.string()).valueDecoder(Codecs.integer()).build();
Iterator<GroupedObservable<String, Integer>> iter = RemoteObservable.connect(connectConfig).getObservable().toBlocking().getIterator();
// verify numbers are grouped by even/odd
// even is first due to zero
GroupedObservable<String, Integer> even = iter.next();
Assert.assertEquals("even", even.getKey());
Iterator<Integer> evenIter = even.toBlocking().getIterator();
Assert.assertEquals(0, evenIter.next().intValue());
Assert.assertEquals(4, evenIter.next().intValue());
Assert.assertEquals(16, evenIter.next().intValue());
Assert.assertEquals(36, evenIter.next().intValue());
GroupedObservable<String, Integer> odd = iter.next();
Assert.assertEquals("odd", odd.getKey());
Iterator<Integer> oddIter = odd.toBlocking().getIterator();
Assert.assertEquals(1, oddIter.next().intValue());
Assert.assertEquals(9, oddIter.next().intValue());
Assert.assertEquals(25, oddIter.next().intValue());
Assert.assertEquals(49, oddIter.next().intValue());
// should only have two groups
Assert.assertEquals(false, iter.hasNext());
}
use of io.reactivex.mantis.remote.observable.RemoteRxServer in project mantis by Netflix.
the class StageExecutorsTest method testExecuteIntermediatStage.
@SuppressWarnings({ "rawtypes", "unchecked" })
@Test
public void testExecuteIntermediatStage() throws InterruptedException {
TestJob provider = new TestJob();
Job<Integer> job = provider.getJobInstance();
List<StageConfig<?, ?>> stages = job.getStages();
PortSelectorWithinRange portSelector = new PortSelectorWithinRange(8000, 9000);
final int publishPort = portSelector.acquirePort();
final int consumerPort = portSelector.acquirePort();
// mimic previous stage with a server
RemoteRxServer server1 = RemoteObservable.serve(consumerPort, Observable.range(0, 10), Codecs.integer());
server1.start();
EndpointInjector staticEndpoints = new EndpointInjector() {
@Override
public Observable<EndpointChange> deltas() {
return Observable.create(new OnSubscribe<EndpointChange>() {
@Override
public void call(Subscriber<? super EndpointChange> subscriber) {
subscriber.onNext(new EndpointChange(EndpointChange.Type.add, new Endpoint("localhost", consumerPort, "1")));
subscriber.onCompleted();
}
});
}
};
WorkerConsumer consumer = new WorkerConsumerRemoteObservable(null, staticEndpoints);
WorkerPublisher producer = new WorkerPublisherRemoteObservable(publishPort, null, Observable.just(1), null);
// execute intermediate, flatten results
StageExecutors.executeIntermediate(consumer, stages.get(1), producer, new Context());
Iterator<Integer> iter = RemoteObservable.connect(new ConnectToObservable.Builder<Integer>().host("localhost").slotId("0").port(publishPort).decoder(Codecs.integer()).build()).getObservable().toBlocking().getIterator();
// verify numbers are even
Assert.assertEquals(0, iter.next().intValue());
Assert.assertEquals(2, iter.next().intValue());
Assert.assertEquals(4, iter.next().intValue());
}
use of io.reactivex.mantis.remote.observable.RemoteRxServer in project mantis by Netflix.
the class StageExecutorsTest method testExecuteSink.
@SuppressWarnings({ "rawtypes", "unchecked" })
@Test
public void testExecuteSink() throws InterruptedException {
TestJob provider = new TestJob();
Job<Integer> job = provider.getJobInstance();
List<StageConfig<?, ?>> stages = job.getStages();
PortSelectorWithinRange portSelector = new PortSelectorWithinRange(8000, 9000);
final int consumerPort = portSelector.acquirePort();
// mimic previous stage with a server
RemoteRxServer server1 = RemoteObservable.serve(consumerPort, Observable.range(0, 10), Codecs.integer());
server1.start();
EndpointInjector staticEndpoints = new EndpointInjector() {
@Override
public Observable<EndpointChange> deltas() {
return Observable.create(new OnSubscribe<EndpointChange>() {
@Override
public void call(Subscriber<? super EndpointChange> subscriber) {
subscriber.onNext(new EndpointChange(EndpointChange.Type.add, new Endpoint("localhost", consumerPort, "1")));
subscriber.onCompleted();
}
});
}
};
Action0 noOpAction = new Action0() {
@Override
public void call() {
}
};
Action1<Throwable> noOpError = new Action1<Throwable>() {
@Override
public void call(Throwable t) {
}
};
WorkerConsumer consumer = new WorkerConsumerRemoteObservable(null, staticEndpoints);
// execute source
StageExecutors.executeSink(consumer, stages.get(1), job.getSink(), new TestPortSelector(), new RxMetrics(), new Context(), noOpAction, null, null, noOpAction, noOpError);
Iterator<Integer> iter = provider.getItemsWritten().iterator();
// verify numbers are even
Assert.assertEquals(0, iter.next().intValue());
Assert.assertEquals(2, iter.next().intValue());
Assert.assertEquals(4, iter.next().intValue());
}
use of io.reactivex.mantis.remote.observable.RemoteRxServer in project mantis by Netflix.
the class WorkerPublisherRemoteObservable method start.
@SuppressWarnings({ "rawtypes", "unchecked" })
@Override
public void start(final StageConfig<T, R> stage, Observable<Observable<R>> toServe) {
RemoteRxServer.Builder serverBuilder = new RemoteRxServer.Builder();
if (stage instanceof KeyToKey || stage instanceof ScalarToKey || stage instanceof ScalarToGroup || stage instanceof GroupToGroup) {
if (runNewW2WserverGroups(jobName)) {
logger.info("Modern server setup for name: " + name + " type: Keyedstage");
long expiryTimeInSecs = Long.MAX_VALUE;
if (stage instanceof KeyToKey) {
expiryTimeInSecs = ((KeyToKey) stage).getKeyExpireTimeSeconds();
} else if (stage instanceof ScalarToKey) {
expiryTimeInSecs = ((ScalarToKey) stage).getKeyExpireTimeSeconds();
}
Func1<R, byte[]> valueEncoder = new Func1<R, byte[]>() {
@Override
public byte[] call(R t1) {
return stage.getOutputCodec().encode(t1);
}
};
Func1<String, byte[]> keyEncoder = new Func1<String, byte[]>() {
@Override
public byte[] call(String t1) {
return Codecs.string().encode(t1);
}
};
ServerConfig<KeyValuePair<String, R>> config = new ServerConfig.Builder<KeyValuePair<String, R>>().name(name).port(serverPort).metricsRegistry(MetricsRegistry.getInstance()).numQueueConsumers(numConsumerThreads()).maxChunkSize(maxChunkSize()).maxChunkTimeMSec(maxChunkTimeMSec()).bufferCapacity(bufferCapacity()).useSpscQueue(useSpsc()).router(Routers.consistentHashingLegacyTcpProtocol(jobName, keyEncoder, valueEncoder)).build();
// remove type
Observable go = toServe;
if (stage instanceof ScalarToGroup || stage instanceof GroupToGroup) {
final LegacyTcpPushServer<KeyValuePair<String, R>> modernServer = PushServers.infiniteStreamLegacyTcpNestedMantisGroup(config, go, expiryTimeInSecs, keyEncoder, io.reactivex.mantis.network.push.HashFunctions.ketama());
modernServer.start();
// support legacy server interface
this.server = new RemoteRxServer() {
@Override
public void start() {
}
@Override
public void startAndWait() {
}
@Override
public void shutdown() {
modernServer.shutdown();
}
@Override
public void blockUntilServerShutdown() {
modernServer.blockUntilShutdown();
}
};
} else {
// ScalarToKey or KeyTKey
final LegacyTcpPushServer<KeyValuePair<String, R>> modernServer = PushServers.infiniteStreamLegacyTcpNestedGroupedObservable(config, go, expiryTimeInSecs, keyEncoder, io.reactivex.mantis.network.push.HashFunctions.ketama());
modernServer.start();
// support legacy server interface
this.server = new RemoteRxServer() {
@Override
public void start() {
}
@Override
public void startAndWait() {
}
@Override
public void shutdown() {
modernServer.shutdown();
}
@Override
public void blockUntilServerShutdown() {
modernServer.blockUntilShutdown();
}
};
}
}
} else if (stage instanceof ScalarToScalar || stage instanceof KeyToScalar || stage instanceof GroupToScalar) {
if (runNewW2Wserver(jobName)) {
logger.info("Modern server setup for name: " + name + " type: Scalarstage");
Func1<R, byte[]> encoder = new Func1<R, byte[]>() {
@Override
public byte[] call(R t1) {
return stage.getOutputCodec().encode(t1);
}
};
ServerConfig<R> config = new ServerConfig.Builder<R>().name(name).port(serverPort).metricsRegistry(MetricsRegistry.getInstance()).router(Routers.roundRobinLegacyTcpProtocol(name, encoder)).build();
final LegacyTcpPushServer<R> modernServer = PushServers.infiniteStreamLegacyTcpNested(config, toServe);
modernServer.start();
// support legacy server interface
this.server = new RemoteRxServer() {
@Override
public void start() {
}
@Override
public void startAndWait() {
}
@Override
public void shutdown() {
modernServer.shutdown();
}
@Override
public void blockUntilServerShutdown() {
modernServer.blockUntilShutdown();
}
};
} else {
logger.info("Legacy server setup for name: " + name + " type: Scalarstage");
RoundRobin slotting = new RoundRobin();
serverBuilder.addObservable(new ServeNestedObservable.Builder().name(name).encoder(stage.getOutputCodec()).observable(toServe).slottingStrategy(slotting).build());
MetricsRegistry.getInstance().registerAndGet(slotting.getMetrics());
server = serverBuilder.port(serverPort).build();
server.start();
}
} else {
throw new RuntimeException("Unsupported stage type: " + stage);
}
}
Aggregations