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

Example 16 with HeartbeatServices

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

the class JobMasterTest method testAllocatedSlotReportDoesNotContainStaleInformation.

/**
 * Tests that the {@link AllocatedSlotReport} contains up to date information and not stale
 * information about the allocated slots on the {@link JobMaster}.
 *
 * <p>This is a probabilistic test case which only fails if executed repeatedly without the fix
 * for FLINK-12863.
 */
@Test
public void testAllocatedSlotReportDoesNotContainStaleInformation() throws Exception {
    final CompletableFuture<Void> assertionFuture = new CompletableFuture<>();
    final UnresolvedTaskManagerLocation unresolvedTaskManagerLocation = new LocalUnresolvedTaskManagerLocation();
    final AtomicBoolean terminateHeartbeatVerification = new AtomicBoolean(false);
    final OneShotLatch hasReceivedSlotOffers = new OneShotLatch();
    final TestingTaskExecutorGateway taskExecutorGateway = new TestingTaskExecutorGatewayBuilder().setHeartbeatJobManagerFunction((taskManagerId, allocatedSlotReport) -> {
        try {
            if (hasReceivedSlotOffers.isTriggered()) {
                assertThat(allocatedSlotReport.getAllocatedSlotInfos(), hasSize(1));
            } else {
                assertThat(allocatedSlotReport.getAllocatedSlotInfos(), empty());
            }
        } catch (AssertionError e) {
            assertionFuture.completeExceptionally(e);
        }
        if (terminateHeartbeatVerification.get()) {
            assertionFuture.complete(null);
        }
        return FutureUtils.completedVoidFuture();
    }).createTestingTaskExecutorGateway();
    rpcService.registerGateway(taskExecutorGateway.getAddress(), taskExecutorGateway);
    final JobManagerSharedServices jobManagerSharedServices = new TestingJobManagerSharedServicesBuilder().build();
    final JobGraph jobGraph = JobGraphTestUtils.singleNoOpJobGraph();
    final JobMaster jobMaster = new JobMasterBuilder(jobGraph, rpcService).withHeartbeatServices(new HeartbeatServices(5L, 1000L)).withSlotPoolServiceSchedulerFactory(DefaultSlotPoolServiceSchedulerFactory.create(new TestingSlotPoolFactory(hasReceivedSlotOffers), new DefaultSchedulerFactory())).createJobMaster();
    jobMaster.start();
    try {
        final JobMasterGateway jobMasterGateway = jobMaster.getSelfGateway(JobMasterGateway.class);
        // register task manager will trigger monitor heartbeat target, schedule heartbeat
        // request at interval time
        CompletableFuture<RegistrationResponse> registrationResponse = jobMasterGateway.registerTaskManager(jobGraph.getJobID(), TaskManagerRegistrationInformation.create(taskExecutorGateway.getAddress(), unresolvedTaskManagerLocation, TestingUtils.zeroUUID()), testingTimeout);
        // wait for the completion of the registration
        registrationResponse.get();
        final SlotOffer slotOffer = new SlotOffer(new AllocationID(), 0, ResourceProfile.ANY);
        final CompletableFuture<Collection<SlotOffer>> slotOfferFuture = jobMasterGateway.offerSlots(unresolvedTaskManagerLocation.getResourceID(), Collections.singleton(slotOffer), testingTimeout);
        assertThat(slotOfferFuture.get(), containsInAnyOrder(slotOffer));
        terminateHeartbeatVerification.set(true);
        // make sure that no assertion has been violated
        assertionFuture.get();
    } finally {
        RpcUtils.terminateRpcEndpoint(jobMaster, testingTimeout);
        jobManagerSharedServices.shutdown();
    }
}

Example 17 with HeartbeatServices

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

the class TaskExecutorBuilder method build.

public TaskExecutor build() throws Exception {
    final TaskExecutorBlobService resolvedTaskExecutorBlobService;
    TaskExecutorResourceUtils.adjustForLocalExecution(configuration);
    if (taskExecutorBlobService == null) {
        resolvedTaskExecutorBlobService = NoOpTaskExecutorBlobService.INSTANCE;
    } else {
        resolvedTaskExecutorBlobService = taskExecutorBlobService;
    }
    final TaskManagerConfiguration resolvedTaskManagerConfiguration;
    if (taskManagerConfiguration == null) {
        resolvedTaskManagerConfiguration = TaskManagerConfiguration.fromConfiguration(configuration, taskExecutorResourceSpec, rpcService.getAddress(), workingDirectory.getTmpDirectory());
    } else {
        resolvedTaskManagerConfiguration = taskManagerConfiguration;
    }
    final TaskManagerServices resolvedTaskManagerServices;
    if (taskManagerServices == null) {
        final TaskManagerServicesConfiguration taskManagerServicesConfiguration = TaskManagerServicesConfiguration.fromConfiguration(configuration, resourceId, rpcService.getAddress(), true, taskExecutorResourceSpec, workingDirectory);
        resolvedTaskManagerServices = TaskManagerServices.fromConfiguration(taskManagerServicesConfiguration, VoidPermanentBlobService.INSTANCE, UnregisteredMetricGroups.createUnregisteredTaskManagerMetricGroup(), Executors.newDirectExecutorService(), throwable -> {
        }, workingDirectory);
    } else {
        resolvedTaskManagerServices = taskManagerServices;
    }
    return new TaskExecutor(rpcService, resolvedTaskManagerConfiguration, haServices, resolvedTaskManagerServices, externalResourceInfoProvider, heartbeatServices, taskManagerMetricGroup, metricQueryServiceAddress, resolvedTaskExecutorBlobService, fatalErrorHandler, partitionTracker);
}

Example 18 with HeartbeatServices

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

the class MiniCluster method setupDispatcherResourceManagerComponents.

@GuardedBy("lock")
private void setupDispatcherResourceManagerComponents(Configuration configuration, RpcServiceFactory dispatcherResourceManagerComponentRpcServiceFactory, MetricQueryServiceRetriever metricQueryServiceRetriever) throws Exception {
    dispatcherResourceManagerComponents.addAll(createDispatcherResourceManagerComponents(configuration, dispatcherResourceManagerComponentRpcServiceFactory, haServices, blobServer, heartbeatServices, metricRegistry, metricQueryServiceRetriever, new ShutDownFatalErrorHandler()));
    final Collection<CompletableFuture<ApplicationStatus>> shutDownFutures = new ArrayList<>(dispatcherResourceManagerComponents.size());
    for (DispatcherResourceManagerComponent dispatcherResourceManagerComponent : dispatcherResourceManagerComponents) {
        final CompletableFuture<ApplicationStatus> shutDownFuture = dispatcherResourceManagerComponent.getShutDownFuture();
        FutureUtils.assertNoException(shutDownFuture.thenCompose(applicationStatus -> dispatcherResourceManagerComponent.stopApplication(applicationStatus, null)));
        shutDownFutures.add(shutDownFuture);
    }
    FutureUtils.completeAll(shutDownFutures).whenComplete((ignored, exception) -> closeAsync());
}

Example 19 with HeartbeatServices

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

the class ProcessFailureCancelingITCase method testCancelingOnProcessFailure.

@Test
public void testCancelingOnProcessFailure() throws Throwable {
    Assume.assumeTrue("---- Skipping Process Failure test : Could not find java executable ----", getJavaCommandPath() != null);
    TestProcess taskManagerProcess = null;
    final TestingFatalErrorHandler fatalErrorHandler = new TestingFatalErrorHandler();
    Configuration config = new Configuration();
    config.setString(JobManagerOptions.ADDRESS, "localhost");
    config.set(AkkaOptions.ASK_TIMEOUT_DURATION, Duration.ofSeconds(100));
    config.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    config.setString(HighAvailabilityOptions.HA_ZOOKEEPER_QUORUM, zooKeeperResource.getConnectString());
    config.setString(HighAvailabilityOptions.HA_STORAGE_PATH, temporaryFolder.newFolder().getAbsolutePath());
    config.setInteger(TaskManagerOptions.NUM_TASK_SLOTS, 2);
    config.set(TaskManagerOptions.MANAGED_MEMORY_SIZE, MemorySize.parse("4m"));
    config.set(TaskManagerOptions.NETWORK_MEMORY_MIN, MemorySize.parse("3200k"));
    config.set(TaskManagerOptions.NETWORK_MEMORY_MAX, MemorySize.parse("3200k"));
    config.set(TaskManagerOptions.TASK_HEAP_MEMORY, MemorySize.parse("128m"));
    config.set(TaskManagerOptions.CPU_CORES, 1.0);
    config.setInteger(RestOptions.PORT, 0);
    final RpcService rpcService = RpcSystem.load().remoteServiceBuilder(config, "localhost", "0").createAndStart();
    final int jobManagerPort = rpcService.getPort();
    config.setInteger(JobManagerOptions.PORT, jobManagerPort);
    final DispatcherResourceManagerComponentFactory resourceManagerComponentFactory = DefaultDispatcherResourceManagerComponentFactory.createSessionComponentFactory(StandaloneResourceManagerFactory.getInstance());
    DispatcherResourceManagerComponent dispatcherResourceManagerComponent = null;
    final ScheduledExecutorService ioExecutor = TestingUtils.defaultExecutor();
    final HighAvailabilityServices haServices = HighAvailabilityServicesUtils.createHighAvailabilityServices(config, ioExecutor, AddressResolution.NO_ADDRESS_RESOLUTION, RpcSystem.load(), NoOpFatalErrorHandler.INSTANCE);
    final AtomicReference<Throwable> programException = new AtomicReference<>();
    try {
        dispatcherResourceManagerComponent = resourceManagerComponentFactory.create(config, ResourceID.generate(), ioExecutor, rpcService, haServices, blobServerResource.getBlobServer(), new HeartbeatServices(100L, 10000L, 2), NoOpMetricRegistry.INSTANCE, new MemoryExecutionGraphInfoStore(), VoidMetricQueryServiceRetriever.INSTANCE, fatalErrorHandler);
        TestProcessBuilder taskManagerProcessBuilder = new TestProcessBuilder(TaskExecutorProcessEntryPoint.class.getName());
        taskManagerProcessBuilder.addConfigAsMainClassArgs(config);
        taskManagerProcess = taskManagerProcessBuilder.start();
        // start the test program, which infinitely blocks
        Runnable programRunner = new Runnable() {

            @Override
            public void run() {
                try {
                    ExecutionEnvironment env = ExecutionEnvironment.createRemoteEnvironment("localhost", 1337, config);
                    env.setParallelism(2);
                    env.setRestartStrategy(RestartStrategies.noRestart());
                    env.generateSequence(0, Long.MAX_VALUE).map(new MapFunction<Long, Long>() {

                        @Override
                        public Long map(Long value) throws Exception {
                            synchronized (this) {
                                System.out.println(TASK_DEPLOYED_MARKER);
                                wait();
                            }
                            return 0L;
                        }
                    }).output(new DiscardingOutputFormat<>());
                    env.execute();
                } catch (Throwable t) {
                    programException.set(t);
                }
            }
        };
        Thread programThread = new Thread(programRunner);
        programThread.start();
        waitUntilAtLeastOneTaskHasBeenDeployed(taskManagerProcess);
        // kill the TaskManager after the job started to run
        taskManagerProcess.destroy();
        taskManagerProcess = null;
        // the job should fail within a few seconds due to heartbeat timeouts
        // since the CI environment is often slow, we conservatively give it up to 2 minutes
        programThread.join(TIMEOUT.toMillis());
        assertFalse("The program did not cancel in time", programThread.isAlive());
        Throwable error = programException.get();
        assertNotNull("The program did not fail properly", error);
        assertTrue(error instanceof ProgramInvocationException);
    // all seems well :-)
    } catch (Exception | Error e) {
        if (taskManagerProcess != null) {
            printOutput("TaskManager OUT", taskManagerProcess.getProcessOutput().toString());
            printOutput("TaskManager ERR", taskManagerProcess.getErrorOutput().toString());
        }
        throw ExceptionUtils.firstOrSuppressed(e, programException.get());
    } finally {
        if (taskManagerProcess != null) {
            taskManagerProcess.destroy();
        }
        if (dispatcherResourceManagerComponent != null) {
            dispatcherResourceManagerComponent.stopApplication(ApplicationStatus.SUCCEEDED, null);
        }
        fatalErrorHandler.rethrowError();
        RpcUtils.terminateRpcService(rpcService, Time.seconds(100L));
        haServices.closeAndCleanupAllData();
    }
}