Examples with HeartbeatServices org.apache.flink.runtime.heartbeat.HeartbeatServices used on opensource projects

Example 11 with HeartbeatServices

use of org.apache.flink.runtime.heartbeat.HeartbeatServices in project flink by apache.

the class JobMasterTest method testReleasingTaskExecutorIfNoMoreSlotsRegistered.

/**
 * Tests that the TaskExecutor is released if all of its slots have been freed.
 */
@Test
public void testReleasingTaskExecutorIfNoMoreSlotsRegistered() throws Exception {
    final JobGraph jobGraph = createSingleVertexJobWithRestartStrategy();
    final JobMaster jobMaster = new JobMasterBuilder(jobGraph, rpcService).withConfiguration(configuration).withHighAvailabilityServices(haServices).withHeartbeatServices(heartbeatServices).createJobMaster();
    final CompletableFuture<JobID> disconnectTaskExecutorFuture = new CompletableFuture<>();
    final CompletableFuture<AllocationID> freedSlotFuture = new CompletableFuture<>();
    final TestingTaskExecutorGateway testingTaskExecutorGateway = new TestingTaskExecutorGatewayBuilder().setFreeSlotFunction((allocationID, throwable) -> {
        freedSlotFuture.complete(allocationID);
        return CompletableFuture.completedFuture(Acknowledge.get());
    }).setDisconnectJobManagerConsumer((jobID, throwable) -> disconnectTaskExecutorFuture.complete(jobID)).createTestingTaskExecutorGateway();
    final LocalUnresolvedTaskManagerLocation taskManagerLocation = new LocalUnresolvedTaskManagerLocation();
    try {
        jobMaster.start();
        final JobMasterGateway jobMasterGateway = jobMaster.getSelfGateway(JobMasterGateway.class);
        final Collection<SlotOffer> slotOffers = registerSlotsAtJobMaster(1, jobMasterGateway, jobGraph.getJobID(), testingTaskExecutorGateway, taskManagerLocation);
        // check that we accepted the offered slot
        assertThat(slotOffers, hasSize(1));
        final AllocationID allocationId = slotOffers.iterator().next().getAllocationId();
        // now fail the allocation and check that we close the connection to the TaskExecutor
        jobMasterGateway.failSlot(taskManagerLocation.getResourceID(), allocationId, new FlinkException("Fail allocation test exception"));
        // we should free the slot and then disconnect from the TaskExecutor because we use no
        // longer slots from it
        assertThat(freedSlotFuture.get(), equalTo(allocationId));
        assertThat(disconnectTaskExecutorFuture.get(), equalTo(jobGraph.getJobID()));
    } finally {
        RpcUtils.terminateRpcEndpoint(jobMaster, testingTimeout);
    }
}

Example 12 with HeartbeatServices

use of org.apache.flink.runtime.heartbeat.HeartbeatServices in project flink by apache.

the class DispatcherTest method testRetrieveJobResultAfterSubmissionOfFailedJob.

@Test
public void testRetrieveJobResultAfterSubmissionOfFailedJob() throws Exception {
    dispatcher = createAndStartDispatcher(heartbeatServices, haServices, new ExpectedJobIdJobManagerRunnerFactory(jobId, createdJobManagerRunnerLatch));
    final DispatcherGateway dispatcherGateway = dispatcher.getSelfGateway(DispatcherGateway.class);
    final JobID failedJobId = new JobID();
    final String failedJobName = "test";
    final CompletableFuture<Acknowledge> submitFuture = dispatcherGateway.submitFailedJob(failedJobId, failedJobName, new RuntimeException("Test exception."));
    submitFuture.get();
    final ArchivedExecutionGraph archivedExecutionGraph = dispatcherGateway.requestJob(failedJobId, TIMEOUT).get();
    Assertions.assertThat(archivedExecutionGraph.getJobID()).isEqualTo(failedJobId);
    Assertions.assertThat(archivedExecutionGraph.getJobName()).isEqualTo(failedJobName);
    Assertions.assertThat(archivedExecutionGraph.getState()).isEqualTo(JobStatus.FAILED);
    Assertions.assertThat(archivedExecutionGraph.getFailureInfo()).isNotNull().extracting(ErrorInfo::getException).extracting(e -> e.deserializeError(Thread.currentThread().getContextClassLoader())).satisfies(exception -> Assertions.assertThat(exception).isInstanceOf(RuntimeException.class).hasMessage("Test exception."));
}

Example 13 with HeartbeatServices

use of org.apache.flink.runtime.heartbeat.HeartbeatServices in project flink by apache.

the class TaskExecutorTest method runJobManagerHeartbeatTest.

private void runJobManagerHeartbeatTest(ResourceID jmResourceId, HeartbeatServices heartbeatServices, Consumer<TestingJobMasterGatewayBuilder> jobMasterGatewayBuilderConsumer, TriConsumer<ResourceID, TaskExecutorGateway, AllocationID> heartbeatAction) throws IOException, InterruptedException, ExecutionException, TimeoutException {
    final JobLeaderService jobLeaderService = new DefaultJobLeaderService(unresolvedTaskManagerLocation, RetryingRegistrationConfiguration.defaultConfiguration());
    final String jobMasterAddress = "jm";
    final UUID jmLeaderId = UUID.randomUUID();
    final CountDownLatch registrationAttempts = new CountDownLatch(2);
    final OneShotLatch slotOfferedLatch = new OneShotLatch();
    final CompletableFuture<ResourceID> disconnectTaskManagerFuture = new CompletableFuture<>();
    final TestingJobMasterGatewayBuilder testingJobMasterGatewayBuilder = new TestingJobMasterGatewayBuilder().setRegisterTaskManagerFunction((ignoredJobId, ignoredTaskManagerRegistrationInformation) -> {
        registrationAttempts.countDown();
        return CompletableFuture.completedFuture(new JMTMRegistrationSuccess(jmResourceId));
    }).setDisconnectTaskManagerFunction(resourceID -> {
        disconnectTaskManagerFuture.complete(resourceID);
        return CompletableFuture.completedFuture(Acknowledge.get());
    }).setOfferSlotsFunction((resourceID, slotOffers) -> {
        slotOfferedLatch.trigger();
        return CompletableFuture.completedFuture(slotOffers);
    });
    jobMasterGatewayBuilderConsumer.accept(testingJobMasterGatewayBuilder);
    final TestingJobMasterGateway jobMasterGateway = testingJobMasterGatewayBuilder.build();
    final TaskExecutorLocalStateStoresManager localStateStoresManager = createTaskExecutorLocalStateStoresManager();
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).setTaskSlotTable(TaskSlotUtils.createTaskSlotTable(1)).setJobLeaderService(jobLeaderService).setTaskStateManager(localStateStoresManager).build();
    final TestingTaskExecutor taskManager = createTestingTaskExecutor(taskManagerServices, heartbeatServices);
    final OneShotLatch slotReportReceived = new OneShotLatch();
    final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
    testingResourceManagerGateway.setSendSlotReportFunction(ignored -> {
        slotReportReceived.trigger();
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    final Queue<CompletableFuture<RegistrationResponse>> registrationResponses = new ArrayDeque<>();
    registrationResponses.add(CompletableFuture.completedFuture(new TaskExecutorRegistrationSuccess(new InstanceID(), testingResourceManagerGateway.getOwnResourceId(), new ClusterInformation("foobar", 1234))));
    registrationResponses.add(new CompletableFuture<>());
    testingResourceManagerGateway.setRegisterTaskExecutorFunction(taskExecutorRegistration -> registrationResponses.poll());
    rpc.registerGateway(jobMasterAddress, jobMasterGateway);
    rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
    try {
        taskManager.start();
        taskManager.waitUntilStarted();
        final TaskExecutorGateway taskExecutorGateway = taskManager.getSelfGateway(TaskExecutorGateway.class);
        resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
        slotReportReceived.await();
        final AllocationID allocationId = new AllocationID();
        requestSlot(taskExecutorGateway, jobId, allocationId, buildSlotID(0), ResourceProfile.UNKNOWN, jobMasterAddress, testingResourceManagerGateway.getFencingToken());
        // now inform the task manager about the new job leader
        jobManagerLeaderRetriever.notifyListener(jobMasterAddress, jmLeaderId);
        // register task manager success will trigger monitoring heartbeat target between tm and
        // jm
        slotOfferedLatch.await();
        heartbeatAction.accept(unresolvedTaskManagerLocation.getResourceID(), taskExecutorGateway, allocationId);
        // the timeout should trigger disconnecting from the JobManager
        final ResourceID resourceID = disconnectTaskManagerFuture.get();
        assertThat(resourceID, equalTo(unresolvedTaskManagerLocation.getResourceID()));
        assertTrue("The TaskExecutor should try to reconnect to the JM", registrationAttempts.await(timeout.toMilliseconds(), TimeUnit.SECONDS));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskManager, timeout);
    }
}

Example 14 with HeartbeatServices

use of org.apache.flink.runtime.heartbeat.HeartbeatServices in project flink by apache.

the class AbstractDispatcherTest method setUp.

@Before
public void setUp() throws Exception {
    heartbeatServices = new HeartbeatServices(1000L, 10000L);
    haServices = new TestingHighAvailabilityServices();
    haServices.setCheckpointRecoveryFactory(new StandaloneCheckpointRecoveryFactory());
    haServices.setResourceManagerLeaderRetriever(new SettableLeaderRetrievalService());
    haServices.setJobGraphStore(new StandaloneJobGraphStore());
    haServices.setJobResultStore(new EmbeddedJobResultStore());
    configuration = new Configuration();
    blobServer = new BlobServer(configuration, temporaryFolder.newFolder(), new VoidBlobStore());
}

Example 15 with HeartbeatServices

use of org.apache.flink.runtime.heartbeat.HeartbeatServices in project flink by apache.

the class JobMasterTest method setupClass.

@BeforeClass
public static void setupClass() {
    rpcService = new TestingRpcService();
    fastHeartbeatServices = new HeartbeatServices(fastHeartbeatInterval, fastHeartbeatTimeout, -1);
    heartbeatServices = new HeartbeatServices(heartbeatInterval, heartbeatTimeout, 1);
}