Examples with StandaloneLeaderRetrievalService org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService used on opensource projects

Example 1 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class JobManagerTest method testKvStateMessages.

/**
	 * Tests that the JobManager handles {@link org.apache.flink.runtime.query.KvStateMessage}
	 * instances as expected.
	 */
@Test
public void testKvStateMessages() throws Exception {
    Deadline deadline = new FiniteDuration(100, TimeUnit.SECONDS).fromNow();
    Configuration config = new Configuration();
    config.setString(ConfigConstants.AKKA_ASK_TIMEOUT, "100ms");
    UUID leaderSessionId = null;
    ActorGateway jobManager = new AkkaActorGateway(JobManager.startJobManagerActors(config, system, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), TestingJobManager.class, MemoryArchivist.class)._1(), leaderSessionId);
    LeaderRetrievalService leaderRetrievalService = new StandaloneLeaderRetrievalService(AkkaUtils.getAkkaURL(system, jobManager.actor()));
    Configuration tmConfig = new Configuration();
    tmConfig.setInteger(ConfigConstants.TASK_MANAGER_MEMORY_SIZE_KEY, 4);
    tmConfig.setInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, 8);
    ActorRef taskManager = TaskManager.startTaskManagerComponentsAndActor(tmConfig, ResourceID.generate(), system, "localhost", scala.Option.<String>empty(), scala.Option.apply(leaderRetrievalService), true, TestingTaskManager.class);
    Future<Object> registrationFuture = jobManager.ask(new NotifyWhenAtLeastNumTaskManagerAreRegistered(1), deadline.timeLeft());
    Await.ready(registrationFuture, deadline.timeLeft());
    //
    // Location lookup
    //
    LookupKvStateLocation lookupNonExistingJob = new LookupKvStateLocation(new JobID(), "any-name");
    Future<KvStateLocation> lookupFuture = jobManager.ask(lookupNonExistingJob, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (IllegalStateException ignored) {
    // Expected
    }
    JobGraph jobGraph = new JobGraph("croissant");
    JobVertex jobVertex1 = new JobVertex("cappuccino");
    jobVertex1.setParallelism(4);
    jobVertex1.setMaxParallelism(16);
    jobVertex1.setInvokableClass(BlockingNoOpInvokable.class);
    JobVertex jobVertex2 = new JobVertex("americano");
    jobVertex2.setParallelism(4);
    jobVertex2.setMaxParallelism(16);
    jobVertex2.setInvokableClass(BlockingNoOpInvokable.class);
    jobGraph.addVertex(jobVertex1);
    jobGraph.addVertex(jobVertex2);
    Future<JobSubmitSuccess> submitFuture = jobManager.ask(new SubmitJob(jobGraph, ListeningBehaviour.DETACHED), deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<JobSubmitSuccess>apply(JobSubmitSuccess.class));
    Await.result(submitFuture, deadline.timeLeft());
    Object lookupUnknownRegistrationName = new LookupKvStateLocation(jobGraph.getJobID(), "unknown");
    lookupFuture = jobManager.ask(lookupUnknownRegistrationName, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (UnknownKvStateLocation ignored) {
    // Expected
    }
    //
    // Registration
    //
    NotifyKvStateRegistered registerNonExistingJob = new NotifyKvStateRegistered(new JobID(), new JobVertexID(), new KeyGroupRange(0, 0), "any-name", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1233));
    jobManager.tell(registerNonExistingJob);
    LookupKvStateLocation lookupAfterRegistration = new LookupKvStateLocation(registerNonExistingJob.getJobId(), registerNonExistingJob.getRegistrationName());
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (IllegalStateException ignored) {
    // Expected
    }
    NotifyKvStateRegistered registerForExistingJob = new NotifyKvStateRegistered(jobGraph.getJobID(), jobVertex1.getID(), new KeyGroupRange(0, 0), "register-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    jobManager.tell(registerForExistingJob);
    lookupAfterRegistration = new LookupKvStateLocation(registerForExistingJob.getJobId(), registerForExistingJob.getRegistrationName());
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    KvStateLocation location = Await.result(lookupFuture, deadline.timeLeft());
    assertNotNull(location);
    assertEquals(jobGraph.getJobID(), location.getJobId());
    assertEquals(jobVertex1.getID(), location.getJobVertexId());
    assertEquals(jobVertex1.getMaxParallelism(), location.getNumKeyGroups());
    assertEquals(1, location.getNumRegisteredKeyGroups());
    KeyGroupRange keyGroupRange = registerForExistingJob.getKeyGroupRange();
    assertEquals(1, keyGroupRange.getNumberOfKeyGroups());
    assertEquals(registerForExistingJob.getKvStateId(), location.getKvStateID(keyGroupRange.getStartKeyGroup()));
    assertEquals(registerForExistingJob.getKvStateServerAddress(), location.getKvStateServerAddress(keyGroupRange.getStartKeyGroup()));
    //
    // Unregistration
    //
    NotifyKvStateUnregistered unregister = new NotifyKvStateUnregistered(registerForExistingJob.getJobId(), registerForExistingJob.getJobVertexId(), registerForExistingJob.getKeyGroupRange(), registerForExistingJob.getRegistrationName());
    jobManager.tell(unregister);
    lookupFuture = jobManager.ask(lookupAfterRegistration, deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<KvStateLocation>apply(KvStateLocation.class));
    try {
        Await.result(lookupFuture, deadline.timeLeft());
        fail("Did not throw expected Exception");
    } catch (UnknownKvStateLocation ignored) {
    // Expected
    }
    //
    // Duplicate registration fails task
    //
    NotifyKvStateRegistered register = new NotifyKvStateRegistered(jobGraph.getJobID(), jobVertex1.getID(), new KeyGroupRange(0, 0), "duplicate-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    NotifyKvStateRegistered duplicate = new NotifyKvStateRegistered(jobGraph.getJobID(), // <--- different operator, but...
    jobVertex2.getID(), new KeyGroupRange(0, 0), // ...same name
    "duplicate-me", new KvStateID(), new KvStateServerAddress(InetAddress.getLocalHost(), 1293));
    Future<TestingJobManagerMessages.JobStatusIs> failedFuture = jobManager.ask(new NotifyWhenJobStatus(jobGraph.getJobID(), JobStatus.FAILED), deadline.timeLeft()).mapTo(ClassTag$.MODULE$.<JobStatusIs>apply(JobStatusIs.class));
    jobManager.tell(register);
    jobManager.tell(duplicate);
    // Wait for failure
    JobStatusIs jobStatus = Await.result(failedFuture, deadline.timeLeft());
    assertEquals(JobStatus.FAILED, jobStatus.state());
}

Example 2 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class TaskManagerMetricsTest method testMetricRegistryLifeCycle.

/**
	 * Tests the metric registry life cycle on JobManager re-connects.
	 */
@Test
public void testMetricRegistryLifeCycle() throws Exception {
    ActorSystem actorSystem = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(new Configuration());
        // ================================================================
        // Start JobManager
        // ================================================================
        final ActorRef jobManager = JobManager.startJobManagerActors(new Configuration(), actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), JobManager.class, MemoryArchivist.class)._1();
        LeaderRetrievalService leaderRetrievalService = new StandaloneLeaderRetrievalService(jobManager.path().toString());
        // ================================================================
        // Start TaskManager
        // ================================================================
        final Configuration config = new Configuration();
        final ResourceID tmResourceID = ResourceID.generate();
        TaskManagerServicesConfiguration taskManagerServicesConfiguration = TaskManagerServicesConfiguration.fromConfiguration(config, InetAddress.getLocalHost(), false);
        TaskManagerConfiguration taskManagerConfiguration = TaskManagerConfiguration.fromConfiguration(config);
        TaskManagerServices taskManagerServices = TaskManagerServices.fromConfiguration(taskManagerServicesConfiguration, tmResourceID);
        final MetricRegistry tmRegistry = taskManagerServices.getMetricRegistry();
        // create the task manager
        final Props tmProps = TaskManager.getTaskManagerProps(TaskManager.class, taskManagerConfiguration, tmResourceID, taskManagerServices.getTaskManagerLocation(), taskManagerServices.getMemoryManager(), taskManagerServices.getIOManager(), taskManagerServices.getNetworkEnvironment(), leaderRetrievalService, tmRegistry);
        final ActorRef taskManager = actorSystem.actorOf(tmProps);
        new JavaTestKit(actorSystem) {

            {
                new Within(new FiniteDuration(5000, TimeUnit.SECONDS)) {

                    @Override
                    protected void run() {
                        taskManager.tell(TaskManagerMessages.getNotifyWhenRegisteredAtJobManagerMessage(), getTestActor());
                        // wait for the TM to be registered
                        expectMsgEquals(TaskManagerMessages.getRegisteredAtJobManagerMessage());
                        // trigger re-registration of TM; this should include a disconnect from the current JM
                        taskManager.tell(new TaskManagerMessages.JobManagerLeaderAddress(jobManager.path().toString(), null), jobManager);
                        // wait for re-registration to be completed
                        taskManager.tell(TaskManagerMessages.getNotifyWhenRegisteredAtJobManagerMessage(), getTestActor());
                        expectMsgEquals(TaskManagerMessages.getRegisteredAtJobManagerMessage());
                    }
                };
            }
        };
        // verify that the registry was not shutdown due to the disconnect
        Assert.assertFalse(tmRegistry.isShutdown());
        // shut down the actors and the actor system
        actorSystem.shutdown();
        actorSystem.awaitTermination();
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
    }
}

Example 3 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class TaskManagerProcessReapingTestBase method testReapProcessOnFailure.

@Test
public void testReapProcessOnFailure() {
    Process taskManagerProcess = null;
    ActorSystem jmActorSystem = null;
    final StringWriter processOutput = new StringWriter();
    try {
        String javaCommand = getJavaCommandPath();
        // is available on this machine
        if (javaCommand == null) {
            System.out.println("---- Skipping TaskManagerProcessReapingTest : Could not find java executable ----");
            return;
        }
        // create a logging file for the process
        File tempLogFile = File.createTempFile("testlogconfig", "properties");
        tempLogFile.deleteOnExit();
        CommonTestUtils.printLog4jDebugConfig(tempLogFile);
        final int jobManagerPort = NetUtils.getAvailablePort();
        // start a JobManager
        Tuple2<String, Object> localAddress = new Tuple2<String, Object>("localhost", jobManagerPort);
        jmActorSystem = AkkaUtils.createActorSystem(new Configuration(), new Some<Tuple2<String, Object>>(localAddress));
        ActorRef jmActor = JobManager.startJobManagerActors(new Configuration(), jmActorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), JobManager.class, MemoryArchivist.class)._1;
        // start a ResourceManager
        StandaloneLeaderRetrievalService standaloneLeaderRetrievalService = new StandaloneLeaderRetrievalService(AkkaUtils.getAkkaURL(jmActorSystem, jmActor));
        FlinkResourceManager.startResourceManagerActors(new Configuration(), jmActorSystem, standaloneLeaderRetrievalService, StandaloneResourceManager.class);
        final int taskManagerPort = NetUtils.getAvailablePort();
        // start the task manager process
        String[] command = new String[] { javaCommand, "-Dlog.level=DEBUG", "-Dlog4j.configuration=file:" + tempLogFile.getAbsolutePath(), "-Xms256m", "-Xmx256m", "-classpath", getCurrentClasspath(), TaskManagerTestEntryPoint.class.getName(), String.valueOf(jobManagerPort), String.valueOf(taskManagerPort) };
        ProcessBuilder bld = new ProcessBuilder(command);
        taskManagerProcess = bld.start();
        new PipeForwarder(taskManagerProcess.getErrorStream(), processOutput);
        // grab the reference to the TaskManager. try multiple times, until the process
        // is started and the TaskManager is up
        String taskManagerActorName = String.format("akka.tcp://flink@%s/user/%s", "localhost:" + taskManagerPort, TaskManager.TASK_MANAGER_NAME());
        ActorRef taskManagerRef = null;
        Throwable lastError = null;
        for (int i = 0; i < 40; i++) {
            try {
                taskManagerRef = TaskManager.getTaskManagerRemoteReference(taskManagerActorName, jmActorSystem, new FiniteDuration(25, TimeUnit.SECONDS));
                break;
            } catch (Throwable t) {
                // TaskManager probably not ready yet
                lastError = t;
            }
            Thread.sleep(500);
        }
        assertTrue("TaskManager process died", isProcessAlive(taskManagerProcess));
        if (taskManagerRef == null) {
            if (lastError != null) {
                lastError.printStackTrace();
            }
            fail("TaskManager process did not launch the TaskManager properly. Failed to look up " + taskManagerActorName);
        }
        // kill the TaskManager actor
        onTaskManagerProcessRunning(taskManagerRef);
        // wait for max 5 seconds for the process to terminate
        {
            long now = System.currentTimeMillis();
            long deadline = now + 10000;
            while (now < deadline && isProcessAlive(taskManagerProcess)) {
                Thread.sleep(100);
                now = System.currentTimeMillis();
            }
        }
        assertFalse("TaskManager process did not terminate upon actor death", isProcessAlive(taskManagerProcess));
        int returnCode = taskManagerProcess.exitValue();
        assertEquals("TaskManager died, but not because of the process reaper", TaskManager.RUNTIME_FAILURE_RETURN_CODE(), returnCode);
        onTaskManagerProcessTerminated(processOutput.toString());
    } catch (Exception e) {
        e.printStackTrace();
        printProcessLog(processOutput.toString());
        fail(e.getMessage());
    } catch (Error e) {
        e.printStackTrace();
        printProcessLog(processOutput.toString());
        throw e;
    } finally {
        if (taskManagerProcess != null) {
            taskManagerProcess.destroy();
        }
        if (jmActorSystem != null) {
            jmActorSystem.shutdown();
        }
    }
}

Example 4 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class TaskManagerComponentsStartupShutdownTest method testComponentsStartupShutdown.

/**
	 * Makes sure that all components are shut down when the TaskManager
	 * actor is shut down.
	 */
@Test
public void testComponentsStartupShutdown() {
    final String[] TMP_DIR = new String[] { ConfigConstants.DEFAULT_TASK_MANAGER_TMP_PATH };
    final Time timeout = Time.seconds(100);
    final int BUFFER_SIZE = 32 * 1024;
    Configuration config = new Configuration();
    config.setString(ConfigConstants.AKKA_WATCH_HEARTBEAT_INTERVAL, "200 ms");
    config.setString(ConfigConstants.AKKA_WATCH_HEARTBEAT_PAUSE, "1 s");
    config.setInteger(ConfigConstants.AKKA_WATCH_THRESHOLD, 1);
    ActorSystem actorSystem = null;
    try {
        actorSystem = AkkaUtils.createLocalActorSystem(config);
        final ActorRef jobManager = JobManager.startJobManagerActors(config, actorSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), JobManager.class, MemoryArchivist.class)._1();
        FlinkResourceManager.startResourceManagerActors(config, actorSystem, LeaderRetrievalUtils.createLeaderRetrievalService(config, jobManager), StandaloneResourceManager.class);
        final int numberOfSlots = 1;
        // create the components for the TaskManager manually
        final TaskManagerConfiguration tmConfig = new TaskManagerConfiguration(numberOfSlots, TMP_DIR, timeout, null, Time.milliseconds(500), Time.seconds(30), Time.seconds(10), // cleanup interval
        1000000, config, // exit-jvm-on-fatal-error
        false);
        final NetworkEnvironmentConfiguration netConf = new NetworkEnvironmentConfiguration(32, BUFFER_SIZE, MemoryType.HEAP, IOManager.IOMode.SYNC, 0, 0, 2, 8, null);
        ResourceID taskManagerId = ResourceID.generate();
        final TaskManagerLocation connectionInfo = new TaskManagerLocation(taskManagerId, InetAddress.getLocalHost(), 10000);
        final MemoryManager memManager = new MemoryManager(32 * BUFFER_SIZE, 1, BUFFER_SIZE, MemoryType.HEAP, false);
        final IOManager ioManager = new IOManagerAsync(TMP_DIR);
        final NetworkEnvironment network = new NetworkEnvironment(new NetworkBufferPool(netConf.numNetworkBuffers(), netConf.networkBufferSize(), netConf.memoryType()), new LocalConnectionManager(), new ResultPartitionManager(), new TaskEventDispatcher(), new KvStateRegistry(), null, netConf.ioMode(), netConf.partitionRequestInitialBackoff(), netConf.partitionRequestMaxBackoff(), netConf.networkBuffersPerChannel(), netConf.extraNetworkBuffersPerGate());
        network.start();
        LeaderRetrievalService leaderRetrievalService = new StandaloneLeaderRetrievalService(jobManager.path().toString());
        MetricRegistryConfiguration metricRegistryConfiguration = MetricRegistryConfiguration.fromConfiguration(config);
        // create the task manager
        final Props tmProps = Props.create(TaskManager.class, tmConfig, taskManagerId, connectionInfo, memManager, ioManager, network, numberOfSlots, leaderRetrievalService, new MetricRegistry(metricRegistryConfiguration));
        final ActorRef taskManager = actorSystem.actorOf(tmProps);
        new JavaTestKit(actorSystem) {

            {
                // wait for the TaskManager to be registered
                new Within(new FiniteDuration(5000, TimeUnit.SECONDS)) {

                    @Override
                    protected void run() {
                        taskManager.tell(TaskManagerMessages.getNotifyWhenRegisteredAtJobManagerMessage(), getTestActor());
                        expectMsgEquals(TaskManagerMessages.getRegisteredAtJobManagerMessage());
                    }
                };
            }
        };
        // shut down all actors and the actor system
        // Kill the Task down the JobManager
        taskManager.tell(Kill.getInstance(), ActorRef.noSender());
        jobManager.tell(Kill.getInstance(), ActorRef.noSender());
        // shut down the actors and the actor system
        actorSystem.shutdown();
        actorSystem.awaitTermination();
        actorSystem = null;
        // now that the TaskManager is shut down, the components should be shut down as well
        assertTrue(network.isShutdown());
        assertTrue(ioManager.isProperlyShutDown());
        assertTrue(memManager.isShutdown());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    } finally {
        if (actorSystem != null) {
            actorSystem.shutdown();
        }
    }
}

Example 5 with StandaloneLeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.StandaloneLeaderRetrievalService in project flink by apache.

the class StandaloneUtils method createLeaderRetrievalService.

/**
	 * Creates a {@link StandaloneLeaderRetrievalService} form the given configuration and the
	 * JobManager name. The host and port for the remote Akka URL are retrieved from the provided
	 * configuration. Instead of using the standard JobManager Akka name, the provided one is used
	 * for the remote Akka URL.
	 *
	 * @param configuration Configuration instance containing hte host and port information
	 * @param resolveInitialHostName If true, resolves the hostname of the StandaloneLeaderRetrievalService
	 * @param jobManagerName Name of the JobManager actor
	 * @return StandaloneLeaderRetrievalService
	 * @throws UnknownHostException if the host name cannot be resolved into an {@link InetAddress}
	 */
public static StandaloneLeaderRetrievalService createLeaderRetrievalService(Configuration configuration, boolean resolveInitialHostName, String jobManagerName) throws UnknownHostException {
    Tuple3<String, String, Object> stringIntPair = TaskManager.getAndCheckJobManagerAddress(configuration);
    String protocol = stringIntPair._1();
    String jobManagerHostname = stringIntPair._2();
    int jobManagerPort = (Integer) stringIntPair._3();
    // Do not try to resolve a hostname to prevent resolving to the wrong IP address
    String hostPort = NetUtils.unresolvedHostAndPortToNormalizedString(jobManagerHostname, jobManagerPort);
    if (resolveInitialHostName) {
        try {
            //noinspection ResultOfMethodCallIgnored
            InetAddress.getByName(jobManagerHostname);
        } catch (UnknownHostException e) {
            throw new UnknownHostException("Cannot resolve the JobManager hostname '" + jobManagerHostname + "' specified in the configuration");
        }
    }
    String jobManagerAkkaUrl = JobManager.getRemoteJobManagerAkkaURL(protocol, hostPort, Option.apply(jobManagerName));
    return new StandaloneLeaderRetrievalService(jobManagerAkkaUrl);
}