Examples with TestingHighAvailabilityServices org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices used on opensource projects

Example 21 with TestingHighAvailabilityServices

use of org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices in project flink by apache.

the class SlotProtocolTest method testSlotsUnavailableRequest.

/**
	 * Tests whether
	 * 1) SlotRequest is routed to the SlotManager
	 * 2) SlotRequest is confirmed
	 * 3) SlotRequest leads to a container allocation
	 * 4) Slot becomes available and TaskExecutor gets a SlotRequest
	 */
@Test
public void testSlotsUnavailableRequest() throws Exception {
    final String rmAddress = "/rm1";
    final String jmAddress = "/jm1";
    final JobID jobID = new JobID();
    testRpcService.registerGateway(jmAddress, mock(JobMasterGateway.class));
    final TestingHighAvailabilityServices testingHaServices = new TestingHighAvailabilityServices();
    final UUID rmLeaderID = UUID.randomUUID();
    final UUID jmLeaderID = UUID.randomUUID();
    TestingLeaderElectionService rmLeaderElectionService = configureHA(testingHaServices, jobID, rmAddress, rmLeaderID, jmAddress, jmLeaderID);
    ResourceManagerConfiguration resourceManagerConfiguration = new ResourceManagerConfiguration(Time.seconds(5L), Time.seconds(5L), Time.minutes(5L));
    JobLeaderIdService jobLeaderIdService = new JobLeaderIdService(testingHaServices, testRpcService.getScheduledExecutor(), resourceManagerConfiguration.getJobTimeout());
    final TestingSlotManagerFactory slotManagerFactory = new TestingSlotManagerFactory();
    SpiedResourceManager resourceManager = new SpiedResourceManager(testRpcService, resourceManagerConfiguration, testingHaServices, slotManagerFactory, mock(MetricRegistry.class), jobLeaderIdService, mock(FatalErrorHandler.class));
    resourceManager.start();
    rmLeaderElectionService.isLeader(rmLeaderID);
    Future<RegistrationResponse> registrationFuture = resourceManager.registerJobManager(rmLeaderID, jmLeaderID, jmAddress, jobID);
    try {
        registrationFuture.get(5, TimeUnit.SECONDS);
    } catch (Exception e) {
        Assert.fail("JobManager registration Future didn't become ready.");
    }
    final SlotManager slotManager = slotManagerFactory.slotManager;
    final AllocationID allocationID = new AllocationID();
    final ResourceProfile resourceProfile = new ResourceProfile(1.0, 100);
    SlotRequest slotRequest = new SlotRequest(jobID, allocationID, resourceProfile);
    RMSlotRequestReply slotRequestReply = resourceManager.requestSlot(jmLeaderID, rmLeaderID, slotRequest);
    // 1) SlotRequest is routed to the SlotManager
    verify(slotManager).requestSlot(slotRequest);
    // 2) SlotRequest is confirmed
    Assert.assertEquals(slotRequestReply.getAllocationID(), allocationID);
    // 3) SlotRequest leads to a container allocation
    Assert.assertEquals(1, resourceManager.startNewWorkerCalled);
    Assert.assertFalse(slotManager.isAllocated(allocationID));
    // slot becomes available
    final String tmAddress = "/tm1";
    TaskExecutorGateway taskExecutorGateway = mock(TaskExecutorGateway.class);
    Mockito.when(taskExecutorGateway.requestSlot(any(SlotID.class), any(JobID.class), any(AllocationID.class), any(String.class), any(UUID.class), any(Time.class))).thenReturn(new FlinkCompletableFuture<TMSlotRequestReply>());
    testRpcService.registerGateway(tmAddress, taskExecutorGateway);
    final ResourceID resourceID = ResourceID.generate();
    final SlotID slotID = new SlotID(resourceID, 0);
    final SlotStatus slotStatus = new SlotStatus(slotID, resourceProfile);
    final SlotReport slotReport = new SlotReport(Collections.singletonList(slotStatus));
    // register slot at SlotManager
    slotManager.registerTaskExecutor(resourceID, new TaskExecutorRegistration(taskExecutorGateway), slotReport);
    // 4) Slot becomes available and TaskExecutor gets a SlotRequest
    verify(taskExecutorGateway, timeout(5000)).requestSlot(eq(slotID), eq(jobID), eq(allocationID), any(String.class), any(UUID.class), any(Time.class));
}

Example 22 with TestingHighAvailabilityServices

use of org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices in project flink by apache.

the class JobMasterTest method testHeartbeatTimeoutWithTaskManager.

@Test
public void testHeartbeatTimeoutWithTaskManager() throws Exception {
    final TestingHighAvailabilityServices haServices = new TestingHighAvailabilityServices();
    final TestingLeaderRetrievalService rmLeaderRetrievalService = new TestingLeaderRetrievalService();
    haServices.setResourceManagerLeaderRetriever(rmLeaderRetrievalService);
    haServices.setCheckpointRecoveryFactory(mock(CheckpointRecoveryFactory.class));
    final TestingFatalErrorHandler testingFatalErrorHandler = new TestingFatalErrorHandler();
    final String jobManagerAddress = "jm";
    final UUID jmLeaderId = UUID.randomUUID();
    final ResourceID jmResourceId = new ResourceID(jobManagerAddress);
    final String taskManagerAddress = "tm";
    final ResourceID tmResourceId = new ResourceID(taskManagerAddress);
    final TaskManagerLocation taskManagerLocation = new TaskManagerLocation(tmResourceId, InetAddress.getLoopbackAddress(), 1234);
    final TaskExecutorGateway taskExecutorGateway = mock(TaskExecutorGateway.class);
    final TestingSerialRpcService rpc = new TestingSerialRpcService();
    rpc.registerGateway(taskManagerAddress, taskExecutorGateway);
    final long heartbeatInterval = 1L;
    final long heartbeatTimeout = 5L;
    final ScheduledExecutor scheduledExecutor = mock(ScheduledExecutor.class);
    final HeartbeatServices heartbeatServices = new TestingHeartbeatServices(heartbeatInterval, heartbeatTimeout, scheduledExecutor);
    final JobGraph jobGraph = new JobGraph();
    try {
        final JobMaster jobMaster = new JobMaster(jmResourceId, jobGraph, new Configuration(), rpc, haServices, heartbeatServices, Executors.newScheduledThreadPool(1), mock(BlobLibraryCacheManager.class), mock(RestartStrategyFactory.class), Time.of(10, TimeUnit.SECONDS), null, mock(OnCompletionActions.class), testingFatalErrorHandler, new FlinkUserCodeClassLoader(new URL[0]));
        // also start the heartbeat manager in job manager
        jobMaster.start(jmLeaderId);
        // register task manager will trigger monitoring heartbeat target, schedule heartbeat request in interval time
        jobMaster.registerTaskManager(taskManagerAddress, taskManagerLocation, jmLeaderId);
        ArgumentCaptor<Runnable> heartbeatRunnableCaptor = ArgumentCaptor.forClass(Runnable.class);
        verify(scheduledExecutor, times(1)).scheduleAtFixedRate(heartbeatRunnableCaptor.capture(), eq(0L), eq(heartbeatInterval), eq(TimeUnit.MILLISECONDS));
        Runnable heartbeatRunnable = heartbeatRunnableCaptor.getValue();
        ArgumentCaptor<Runnable> timeoutRunnableCaptor = ArgumentCaptor.forClass(Runnable.class);
        verify(scheduledExecutor).schedule(timeoutRunnableCaptor.capture(), eq(heartbeatTimeout), eq(TimeUnit.MILLISECONDS));
        Runnable timeoutRunnable = timeoutRunnableCaptor.getValue();
        // run the first heartbeat request
        heartbeatRunnable.run();
        verify(taskExecutorGateway, times(1)).heartbeatFromJobManager(eq(jmResourceId));
        // run the timeout runnable to simulate a heartbeat timeout
        timeoutRunnable.run();
        verify(taskExecutorGateway).disconnectJobManager(eq(jobGraph.getJobID()), any(TimeoutException.class));
        // check if a concurrent error occurred
        testingFatalErrorHandler.rethrowError();
    } finally {
        rpc.stopService();
    }
}

Example 23 with TestingHighAvailabilityServices

use of org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices in project flink by apache.

the class DefaultJobLeaderIdServiceTest method testInitialJobTimeout.

/**
 * Tests that the initial job registration registers a timeout which will call {@link
 * JobLeaderIdActions#notifyJobTimeout(JobID, UUID)} when executed.
 */
@Test
public void testInitialJobTimeout() throws Exception {
    final JobID jobId = new JobID();
    TestingHighAvailabilityServices highAvailabilityServices = new TestingHighAvailabilityServices();
    SettableLeaderRetrievalService leaderRetrievalService = new SettableLeaderRetrievalService(null, null);
    highAvailabilityServices.setJobMasterLeaderRetriever(jobId, leaderRetrievalService);
    ScheduledExecutor scheduledExecutor = mock(ScheduledExecutor.class);
    Time timeout = Time.milliseconds(5000L);
    JobLeaderIdActions jobLeaderIdActions = mock(JobLeaderIdActions.class);
    JobLeaderIdService jobLeaderIdService = new DefaultJobLeaderIdService(highAvailabilityServices, scheduledExecutor, timeout);
    jobLeaderIdService.start(jobLeaderIdActions);
    jobLeaderIdService.addJob(jobId);
    assertTrue(jobLeaderIdService.containsJob(jobId));
    ArgumentCaptor<Runnable> runnableArgumentCaptor = ArgumentCaptor.forClass(Runnable.class);
    verify(scheduledExecutor).schedule(runnableArgumentCaptor.capture(), anyLong(), any(TimeUnit.class));
    Runnable timeoutRunnable = runnableArgumentCaptor.getValue();
    timeoutRunnable.run();
    ArgumentCaptor<UUID> timeoutIdArgumentCaptor = ArgumentCaptor.forClass(UUID.class);
    verify(jobLeaderIdActions, times(1)).notifyJobTimeout(eq(jobId), timeoutIdArgumentCaptor.capture());
    assertTrue(jobLeaderIdService.isValidTimeout(jobId, timeoutIdArgumentCaptor.getValue()));
}

Example 24 with TestingHighAvailabilityServices

use of org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices in project flink by apache.

the class DefaultJobLeaderIdServiceTest method jobTimeoutAfterLostLeadership.

/**
 * Tests that a timeout get cancelled once a job leader has been found. Furthermore, it tests
 * that a new timeout is registered after the jobmanager has lost leadership.
 */
@Test(timeout = 10000)
public void jobTimeoutAfterLostLeadership() throws Exception {
    final JobID jobId = new JobID();
    final String address = "foobar";
    final JobMasterId leaderId = JobMasterId.generate();
    TestingHighAvailabilityServices highAvailabilityServices = new TestingHighAvailabilityServices();
    SettableLeaderRetrievalService leaderRetrievalService = new SettableLeaderRetrievalService(null, null);
    highAvailabilityServices.setJobMasterLeaderRetriever(jobId, leaderRetrievalService);
    ScheduledFuture<?> timeout1 = mock(ScheduledFuture.class);
    ScheduledFuture<?> timeout2 = mock(ScheduledFuture.class);
    final Queue<ScheduledFuture<?>> timeoutQueue = new ArrayDeque<>(Arrays.asList(timeout1, timeout2));
    ScheduledExecutor scheduledExecutor = mock(ScheduledExecutor.class);
    final AtomicReference<Runnable> lastRunnable = new AtomicReference<>();
    doAnswer(new Answer() {

        @Override
        public Object answer(InvocationOnMock invocation) throws Throwable {
            lastRunnable.set((Runnable) invocation.getArguments()[0]);
            return timeoutQueue.poll();
        }
    }).when(scheduledExecutor).schedule(any(Runnable.class), anyLong(), any(TimeUnit.class));
    Time timeout = Time.milliseconds(5000L);
    JobLeaderIdActions jobLeaderIdActions = mock(JobLeaderIdActions.class);
    final AtomicReference<UUID> lastTimeoutId = new AtomicReference<>();
    doAnswer(new Answer() {

        @Override
        public Object answer(InvocationOnMock invocation) throws Throwable {
            lastTimeoutId.set((UUID) invocation.getArguments()[1]);
            return null;
        }
    }).when(jobLeaderIdActions).notifyJobTimeout(eq(jobId), any(UUID.class));
    JobLeaderIdService jobLeaderIdService = new DefaultJobLeaderIdService(highAvailabilityServices, scheduledExecutor, timeout);
    jobLeaderIdService.start(jobLeaderIdActions);
    jobLeaderIdService.addJob(jobId);
    CompletableFuture<JobMasterId> leaderIdFuture = jobLeaderIdService.getLeaderId(jobId);
    // notify the leader id service about the new leader
    leaderRetrievalService.notifyListener(address, leaderId.toUUID());
    assertEquals(leaderId, leaderIdFuture.get());
    assertTrue(jobLeaderIdService.containsJob(jobId));
    // check that the first timeout got cancelled
    verify(timeout1, times(1)).cancel(anyBoolean());
    verify(scheduledExecutor, times(1)).schedule(any(Runnable.class), anyLong(), any(TimeUnit.class));
    // initial timeout runnable which should no longer have an effect
    Runnable runnable = lastRunnable.get();
    assertNotNull(runnable);
    runnable.run();
    verify(jobLeaderIdActions, times(1)).notifyJobTimeout(eq(jobId), any(UUID.class));
    // the timeout should no longer be valid
    assertFalse(jobLeaderIdService.isValidTimeout(jobId, lastTimeoutId.get()));
    // lose leadership
    leaderRetrievalService.notifyListener("", null);
    verify(scheduledExecutor, times(2)).schedule(any(Runnable.class), anyLong(), any(TimeUnit.class));
    // the second runnable should be the new timeout
    runnable = lastRunnable.get();
    assertNotNull(runnable);
    runnable.run();
    verify(jobLeaderIdActions, times(2)).notifyJobTimeout(eq(jobId), any(UUID.class));
    // the new timeout should be valid
    assertTrue(jobLeaderIdService.isValidTimeout(jobId, lastTimeoutId.get()));
}

Example 25 with TestingHighAvailabilityServices

use of org.apache.flink.runtime.highavailability.TestingHighAvailabilityServices in project flink by apache.

the class ResourceManagerTest method setup.

@Before
public void setup() throws Exception {
    highAvailabilityServices = new TestingHighAvailabilityServices();
    highAvailabilityServices.setResourceManagerLeaderElectionService(new TestingLeaderElectionService());
    testingFatalErrorHandler = new TestingFatalErrorHandler();
    resourceManagerResourceId = ResourceID.generate();
}