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

Example 1 with LeaderRetrievalService

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

the class JobLeaderService method stop.

/**
	 * Stop the job leader services. This implies stopping all leader retrieval services for the
	 * different jobs and their leader retrieval listeners.
	 *
	 * @throws Exception if an error occurs while stopping the service
	 */
public void stop() throws Exception {
    LOG.info("Stop job leader service.");
    if (JobLeaderService.State.STARTED == state) {
        for (Tuple2<LeaderRetrievalService, JobLeaderService.JobManagerLeaderListener> leaderRetrievalServiceEntry : jobLeaderServices.values()) {
            LeaderRetrievalService leaderRetrievalService = leaderRetrievalServiceEntry.f0;
            JobLeaderService.JobManagerLeaderListener jobManagerLeaderListener = leaderRetrievalServiceEntry.f1;
            jobManagerLeaderListener.stop();
            leaderRetrievalService.stop();
        }
        jobLeaderServices.clear();
    }
    state = JobLeaderService.State.STOPPED;
}

Example 2 with LeaderRetrievalService

use of org.apache.flink.runtime.leaderretrieval.LeaderRetrievalService 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 3 with LeaderRetrievalService

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

the class JobSubmitTest method setupJobManager.

@BeforeClass
public static void setupJobManager() {
    jmConfig = new Configuration();
    int port = NetUtils.getAvailablePort();
    jmConfig.setString(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY, "localhost");
    jmConfig.setInteger(ConfigConstants.JOB_MANAGER_IPC_PORT_KEY, port);
    scala.Option<Tuple2<String, Object>> listeningAddress = scala.Option.apply(new Tuple2<String, Object>("localhost", port));
    jobManagerSystem = AkkaUtils.createActorSystem(jmConfig, listeningAddress);
    // only start JobManager (no ResourceManager)
    JobManager.startJobManagerActors(jmConfig, jobManagerSystem, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), JobManager.class, MemoryArchivist.class)._1();
    try {
        LeaderRetrievalService lrs = LeaderRetrievalUtils.createLeaderRetrievalService(jmConfig);
        jmGateway = LeaderRetrievalUtils.retrieveLeaderGateway(lrs, jobManagerSystem, timeout);
    } catch (Exception e) {
        fail("Could not retrieve the JobManager gateway. " + e.getMessage());
    }
}

Example 4 with LeaderRetrievalService

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

the class ZooKeeperLeaderRetrievalTest method testTimeoutOfFindConnectingAddress.

/**
	 * Tests that the LeaderRetrievalUtils.findConnectingAddress stops trying to find the
	 * connecting address if no leader address has been specified. The call should return
	 * then InetAddress.getLocalHost().
	 */
@Test
public void testTimeoutOfFindConnectingAddress() throws Exception {
    Configuration config = new Configuration();
    config.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    config.setString(HighAvailabilityOptions.HA_ZOOKEEPER_QUORUM, testingServer.getConnectString());
    FiniteDuration timeout = new FiniteDuration(10, TimeUnit.SECONDS);
    LeaderRetrievalService leaderRetrievalService = LeaderRetrievalUtils.createLeaderRetrievalService(config);
    InetAddress result = LeaderRetrievalUtils.findConnectingAddress(leaderRetrievalService, timeout);
    assertEquals(InetAddress.getLocalHost(), result);
}

Example 5 with LeaderRetrievalService

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

the class TaskExecutorTest method testSubmitTaskBeforeAcceptSlot.

/**
	 * This tests task executor receive SubmitTask before OfferSlot response.
	 */
@Test
public void testSubmitTaskBeforeAcceptSlot() throws Exception {
    final JobID jobId = new JobID();
    final TestingSerialRpcService rpc = new TestingSerialRpcService();
    final Configuration configuration = new Configuration();
    final TaskManagerConfiguration taskManagerConfiguration = TaskManagerConfiguration.fromConfiguration(configuration);
    final ResourceID resourceId = new ResourceID("foobar");
    final TaskManagerLocation taskManagerLocation = new TaskManagerLocation(resourceId, InetAddress.getLoopbackAddress(), 1234);
    final TestingHighAvailabilityServices haServices = new TestingHighAvailabilityServices();
    final TimerService<AllocationID> timerService = mock(TimerService.class);
    final TaskSlotTable taskSlotTable = new TaskSlotTable(Arrays.asList(mock(ResourceProfile.class), mock(ResourceProfile.class)), timerService);
    final JobManagerTable jobManagerTable = new JobManagerTable();
    final JobLeaderService jobLeaderService = new JobLeaderService(taskManagerLocation);
    final TestingFatalErrorHandler testingFatalErrorHandler = new TestingFatalErrorHandler();
    final String resourceManagerAddress = "rm";
    final UUID resourceManagerLeaderId = UUID.randomUUID();
    final String jobManagerAddress = "jm";
    final UUID jobManagerLeaderId = UUID.randomUUID();
    final LeaderRetrievalService resourceManagerLeaderRetrievalService = new TestingLeaderRetrievalService(resourceManagerAddress, resourceManagerLeaderId);
    final LeaderRetrievalService jobManagerLeaderRetrievalService = new TestingLeaderRetrievalService(jobManagerAddress, jobManagerLeaderId);
    haServices.setResourceManagerLeaderRetriever(resourceManagerLeaderRetrievalService);
    haServices.setJobMasterLeaderRetriever(jobId, jobManagerLeaderRetrievalService);
    final ResourceManagerGateway resourceManagerGateway = mock(ResourceManagerGateway.class);
    final InstanceID registrationId = new InstanceID();
    when(resourceManagerGateway.registerTaskExecutor(eq(resourceManagerLeaderId), any(String.class), eq(resourceId), any(SlotReport.class), any(Time.class))).thenReturn(FlinkCompletableFuture.<RegistrationResponse>completed(new TaskExecutorRegistrationSuccess(registrationId, 1000L)));
    final ResourceID jmResourceId = new ResourceID(jobManagerAddress);
    final int blobPort = 42;
    final AllocationID allocationId1 = new AllocationID();
    final AllocationID allocationId2 = new AllocationID();
    final SlotOffer offer1 = new SlotOffer(allocationId1, 0, ResourceProfile.UNKNOWN);
    final JobMasterGateway jobMasterGateway = mock(JobMasterGateway.class);
    when(jobMasterGateway.registerTaskManager(any(String.class), eq(taskManagerLocation), eq(jobManagerLeaderId), any(Time.class))).thenReturn(FlinkCompletableFuture.<RegistrationResponse>completed(new JMTMRegistrationSuccess(jmResourceId, blobPort)));
    when(jobMasterGateway.getHostname()).thenReturn(jobManagerAddress);
    rpc.registerGateway(resourceManagerAddress, resourceManagerGateway);
    rpc.registerGateway(jobManagerAddress, jobMasterGateway);
    final LibraryCacheManager libraryCacheManager = mock(LibraryCacheManager.class);
    when(libraryCacheManager.getClassLoader(eq(jobId))).thenReturn(getClass().getClassLoader());
    final JobManagerConnection jobManagerConnection = new JobManagerConnection(jobId, jmResourceId, jobMasterGateway, jobManagerLeaderId, mock(TaskManagerActions.class), mock(CheckpointResponder.class), libraryCacheManager, mock(ResultPartitionConsumableNotifier.class), mock(PartitionProducerStateChecker.class));
    jobManagerTable.put(jobId, jobManagerConnection);
    try {
        final TaskExecutor taskManager = new TaskExecutor(taskManagerConfiguration, taskManagerLocation, rpc, mock(MemoryManager.class), mock(IOManager.class), mock(NetworkEnvironment.class), haServices, mock(HeartbeatServices.class, RETURNS_MOCKS), mock(MetricRegistry.class), mock(TaskManagerMetricGroup.class), mock(BroadcastVariableManager.class), mock(FileCache.class), taskSlotTable, jobManagerTable, jobLeaderService, testingFatalErrorHandler);
        taskManager.start();
        taskSlotTable.allocateSlot(0, jobId, allocationId1, Time.milliseconds(10000L));
        taskSlotTable.allocateSlot(1, jobId, allocationId2, Time.milliseconds(10000L));
        final JobVertexID jobVertexId = new JobVertexID();
        JobInformation jobInformation = new JobInformation(jobId, name.getMethodName(), new SerializedValue<>(new ExecutionConfig()), new Configuration(), Collections.<BlobKey>emptyList(), Collections.<URL>emptyList());
        TaskInformation taskInformation = new TaskInformation(jobVertexId, "test task", 1, 1, TestInvokable.class.getName(), new Configuration());
        SerializedValue<JobInformation> serializedJobInformation = new SerializedValue<>(jobInformation);
        SerializedValue<TaskInformation> serializedJobVertexInformation = new SerializedValue<>(taskInformation);
        final TaskDeploymentDescriptor tdd = new TaskDeploymentDescriptor(serializedJobInformation, serializedJobVertexInformation, new ExecutionAttemptID(), allocationId1, 0, 0, 0, null, Collections.<ResultPartitionDeploymentDescriptor>emptyList(), Collections.<InputGateDeploymentDescriptor>emptyList());
        CompletableFuture<Iterable<SlotOffer>> offerResultFuture = new FlinkCompletableFuture<>();
        // submit task first and then return acceptance response
        when(jobMasterGateway.offerSlots(any(ResourceID.class), any(Iterable.class), eq(jobManagerLeaderId), any(Time.class))).thenReturn(offerResultFuture);
        // we have to add the job after the TaskExecutor, because otherwise the service has not
        // been properly started. This will also offer the slots to the job master
        jobLeaderService.addJob(jobId, jobManagerAddress);
        verify(jobMasterGateway).offerSlots(any(ResourceID.class), any(Iterable.class), eq(jobManagerLeaderId), any(Time.class));
        // submit the task without having acknowledge the offered slots
        taskManager.submitTask(tdd, jobManagerLeaderId);
        // acknowledge the offered slots
        offerResultFuture.complete(Collections.singleton(offer1));
        verify(resourceManagerGateway).notifySlotAvailable(eq(resourceManagerLeaderId), eq(registrationId), eq(new SlotID(resourceId, 1)));
        assertTrue(taskSlotTable.existsActiveSlot(jobId, allocationId1));
        assertFalse(taskSlotTable.existsActiveSlot(jobId, allocationId2));
        assertTrue(taskSlotTable.isSlotFree(1));
        // check if a concurrent error occurred
        testingFatalErrorHandler.rethrowError();
    } finally {
        rpc.stopService();
    }
}