Examples with MetricRegistry org.apache.flink.runtime.metrics.MetricRegistry used on opensource projects

Example 51 with MetricRegistry

use of org.apache.flink.runtime.metrics.MetricRegistry in project flink by apache.

the class TaskExecutorITCase method testSlotAllocation.

@Test
public void testSlotAllocation() throws Exception {
    TestingFatalErrorHandler testingFatalErrorHandler = new TestingFatalErrorHandler();
    TestingHighAvailabilityServices testingHAServices = new TestingHighAvailabilityServices();
    final Configuration configuration = new Configuration();
    final ScheduledExecutorService scheduledExecutorService = new ScheduledThreadPoolExecutor(1);
    final ResourceID taskManagerResourceId = new ResourceID("foobar");
    final UUID rmLeaderId = UUID.randomUUID();
    final TestingLeaderElectionService rmLeaderElectionService = new TestingLeaderElectionService();
    final TestingLeaderRetrievalService rmLeaderRetrievalService = new TestingLeaderRetrievalService();
    final String rmAddress = "rm";
    final String jmAddress = "jm";
    final UUID jmLeaderId = UUID.randomUUID();
    final JobID jobId = new JobID();
    final ResourceProfile resourceProfile = new ResourceProfile(1.0, 1);
    testingHAServices.setResourceManagerLeaderElectionService(rmLeaderElectionService);
    testingHAServices.setResourceManagerLeaderRetriever(rmLeaderRetrievalService);
    testingHAServices.setJobMasterLeaderRetriever(jobId, new TestingLeaderRetrievalService(jmAddress, jmLeaderId));
    TestingSerialRpcService rpcService = new TestingSerialRpcService();
    ResourceManagerConfiguration resourceManagerConfiguration = new ResourceManagerConfiguration(Time.milliseconds(500L), Time.milliseconds(500L), Time.minutes(5L));
    SlotManagerFactory slotManagerFactory = new DefaultSlotManager.Factory();
    JobLeaderIdService jobLeaderIdService = new JobLeaderIdService(testingHAServices, rpcService.getScheduledExecutor(), resourceManagerConfiguration.getJobTimeout());
    MetricRegistry metricRegistry = mock(MetricRegistry.class);
    HeartbeatServices heartbeatServices = mock(HeartbeatServices.class, RETURNS_MOCKS);
    final TaskManagerConfiguration taskManagerConfiguration = TaskManagerConfiguration.fromConfiguration(configuration);
    final TaskManagerLocation taskManagerLocation = new TaskManagerLocation(taskManagerResourceId, InetAddress.getLocalHost(), 1234);
    final MemoryManager memoryManager = mock(MemoryManager.class);
    final IOManager ioManager = mock(IOManager.class);
    final NetworkEnvironment networkEnvironment = mock(NetworkEnvironment.class);
    final TaskManagerMetricGroup taskManagerMetricGroup = mock(TaskManagerMetricGroup.class);
    final BroadcastVariableManager broadcastVariableManager = mock(BroadcastVariableManager.class);
    final FileCache fileCache = mock(FileCache.class);
    final TaskSlotTable taskSlotTable = new TaskSlotTable(Arrays.asList(resourceProfile), new TimerService<AllocationID>(scheduledExecutorService, 100L));
    final JobManagerTable jobManagerTable = new JobManagerTable();
    final JobLeaderService jobLeaderService = new JobLeaderService(taskManagerLocation);
    ResourceManager<ResourceID> resourceManager = new StandaloneResourceManager(rpcService, resourceManagerConfiguration, testingHAServices, slotManagerFactory, metricRegistry, jobLeaderIdService, testingFatalErrorHandler);
    TaskExecutor taskExecutor = new TaskExecutor(taskManagerConfiguration, taskManagerLocation, rpcService, memoryManager, ioManager, networkEnvironment, testingHAServices, heartbeatServices, metricRegistry, taskManagerMetricGroup, broadcastVariableManager, fileCache, taskSlotTable, jobManagerTable, jobLeaderService, testingFatalErrorHandler);
    JobMasterGateway jmGateway = mock(JobMasterGateway.class);
    when(jmGateway.registerTaskManager(any(String.class), any(TaskManagerLocation.class), eq(jmLeaderId), any(Time.class))).thenReturn(FlinkCompletableFuture.<RegistrationResponse>completed(new JMTMRegistrationSuccess(taskManagerResourceId, 1234)));
    when(jmGateway.getHostname()).thenReturn(jmAddress);
    rpcService.registerGateway(rmAddress, resourceManager.getSelf());
    rpcService.registerGateway(jmAddress, jmGateway);
    final AllocationID allocationId = new AllocationID();
    final SlotRequest slotRequest = new SlotRequest(jobId, allocationId, resourceProfile);
    final SlotOffer slotOffer = new SlotOffer(allocationId, 0, resourceProfile);
    try {
        resourceManager.start();
        taskExecutor.start();
        // notify the RM that it is the leader
        rmLeaderElectionService.isLeader(rmLeaderId);
        // notify the TM about the new RM leader
        rmLeaderRetrievalService.notifyListener(rmAddress, rmLeaderId);
        Future<RegistrationResponse> registrationResponseFuture = resourceManager.registerJobManager(rmLeaderId, jmLeaderId, jmAddress, jobId);
        RegistrationResponse registrationResponse = registrationResponseFuture.get();
        assertTrue(registrationResponse instanceof JobMasterRegistrationSuccess);
        resourceManager.requestSlot(jmLeaderId, rmLeaderId, slotRequest);
        verify(jmGateway).offerSlots(eq(taskManagerResourceId), (Iterable<SlotOffer>) argThat(Matchers.contains(slotOffer)), eq(jmLeaderId), any(Time.class));
    } finally {
        if (testingFatalErrorHandler.hasExceptionOccurred()) {
            testingFatalErrorHandler.rethrowError();
        }
    }
}

Example 52 with MetricRegistry

use of org.apache.flink.runtime.metrics.MetricRegistry 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 53 with MetricRegistry

use of org.apache.flink.runtime.metrics.MetricRegistry in project flink by apache.

the class JobManagerHARecoveryTest method testFailingJobRecovery.

/**
	 * Tests that a failing job recovery won't cause other job recoveries to fail.
	 */
@Test
public void testFailingJobRecovery() throws Exception {
    final FiniteDuration timeout = new FiniteDuration(10, TimeUnit.SECONDS);
    final FiniteDuration jobRecoveryTimeout = new FiniteDuration(0, TimeUnit.SECONDS);
    Deadline deadline = new FiniteDuration(1, TimeUnit.MINUTES).fromNow();
    final Configuration flinkConfiguration = new Configuration();
    UUID leaderSessionID = UUID.randomUUID();
    ActorRef jobManager = null;
    JobID jobId1 = new JobID();
    JobID jobId2 = new JobID();
    // set HA mode to zookeeper so that we try to recover jobs
    flinkConfiguration.setString(HighAvailabilityOptions.HA_MODE, "zookeeper");
    try {
        final SubmittedJobGraphStore submittedJobGraphStore = mock(SubmittedJobGraphStore.class);
        SubmittedJobGraph submittedJobGraph = mock(SubmittedJobGraph.class);
        when(submittedJobGraph.getJobId()).thenReturn(jobId2);
        when(submittedJobGraphStore.getJobIds()).thenReturn(Arrays.asList(jobId1, jobId2));
        // fail the first job recovery
        when(submittedJobGraphStore.recoverJobGraph(eq(jobId1))).thenThrow(new Exception("Test exception"));
        // succeed the second job recovery
        when(submittedJobGraphStore.recoverJobGraph(eq(jobId2))).thenReturn(submittedJobGraph);
        final TestingLeaderElectionService myLeaderElectionService = new TestingLeaderElectionService();
        final Collection<JobID> recoveredJobs = new ArrayList<>(2);
        Props jobManagerProps = Props.create(TestingFailingHAJobManager.class, flinkConfiguration, TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), mock(InstanceManager.class), mock(Scheduler.class), new BlobLibraryCacheManager(mock(BlobService.class), 1 << 20), ActorRef.noSender(), new FixedDelayRestartStrategy.FixedDelayRestartStrategyFactory(Int.MaxValue(), 100), timeout, myLeaderElectionService, submittedJobGraphStore, mock(CheckpointRecoveryFactory.class), jobRecoveryTimeout, Option.<MetricRegistry>apply(null), recoveredJobs).withDispatcher(CallingThreadDispatcher.Id());
        jobManager = system.actorOf(jobManagerProps);
        Future<Object> started = Patterns.ask(jobManager, new Identify(42), deadline.timeLeft().toMillis());
        Await.ready(started, deadline.timeLeft());
        // make the job manager the leader --> this triggers the recovery of all jobs
        myLeaderElectionService.isLeader(leaderSessionID);
        // check that we have successfully recovered the second job
        assertThat(recoveredJobs, containsInAnyOrder(jobId2));
    } finally {
        TestingUtils.stopActor(jobManager);
    }
}