Examples with ResourceID org.apache.flink.runtime.clusterframework.types.ResourceID used on opensource projects

Example 6 with ResourceID

use of org.apache.flink.runtime.clusterframework.types.ResourceID in project flink by apache.

the class TaskExecutorTest method testTriggerRegistrationOnLeaderChange.

@Test
public void testTriggerRegistrationOnLeaderChange() throws Exception {
    final ResourceID resourceID = ResourceID.generate();
    final String address1 = "/resource/manager/address/one";
    final String address2 = "/resource/manager/address/two";
    final UUID leaderId1 = UUID.randomUUID();
    final UUID leaderId2 = UUID.randomUUID();
    final TestingSerialRpcService rpc = new TestingSerialRpcService();
    try {
        // register the mock resource manager gateways
        ResourceManagerGateway rmGateway1 = mock(ResourceManagerGateway.class);
        ResourceManagerGateway rmGateway2 = mock(ResourceManagerGateway.class);
        when(rmGateway1.registerTaskExecutor(any(UUID.class), anyString(), any(ResourceID.class), any(SlotReport.class), any(Time.class))).thenReturn(FlinkCompletableFuture.<RegistrationResponse>completed(new RegistrationResponse.Success()));
        when(rmGateway2.registerTaskExecutor(any(UUID.class), anyString(), any(ResourceID.class), any(SlotReport.class), any(Time.class))).thenReturn(FlinkCompletableFuture.<RegistrationResponse>completed(new RegistrationResponse.Success()));
        rpc.registerGateway(address1, rmGateway1);
        rpc.registerGateway(address2, rmGateway2);
        TestingLeaderRetrievalService testLeaderService = new TestingLeaderRetrievalService();
        TestingHighAvailabilityServices haServices = new TestingHighAvailabilityServices();
        haServices.setResourceManagerLeaderRetriever(testLeaderService);
        TaskManagerConfiguration taskManagerServicesConfiguration = mock(TaskManagerConfiguration.class);
        when(taskManagerServicesConfiguration.getNumberSlots()).thenReturn(1);
        when(taskManagerServicesConfiguration.getConfiguration()).thenReturn(new Configuration());
        when(taskManagerServicesConfiguration.getTmpDirectories()).thenReturn(new String[1]);
        TaskManagerLocation taskManagerLocation = mock(TaskManagerLocation.class);
        when(taskManagerLocation.getResourceID()).thenReturn(resourceID);
        when(taskManagerLocation.getHostname()).thenReturn("foobar");
        final TaskSlotTable taskSlotTable = mock(TaskSlotTable.class);
        final SlotReport slotReport = new SlotReport();
        when(taskSlotTable.createSlotReport(any(ResourceID.class))).thenReturn(slotReport);
        final TestingFatalErrorHandler testingFatalErrorHandler = new TestingFatalErrorHandler();
        TaskExecutor taskManager = new TaskExecutor(taskManagerServicesConfiguration, 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, mock(JobManagerTable.class), mock(JobLeaderService.class), testingFatalErrorHandler);
        taskManager.start();
        String taskManagerAddress = taskManager.getAddress();
        // no connection initially, since there is no leader
        assertNull(taskManager.getResourceManagerConnection());
        // define a leader and see that a registration happens
        testLeaderService.notifyListener(address1, leaderId1);
        verify(rmGateway1).registerTaskExecutor(eq(leaderId1), eq(taskManagerAddress), eq(resourceID), any(SlotReport.class), any(Time.class));
        assertNotNull(taskManager.getResourceManagerConnection());
        // cancel the leader 
        testLeaderService.notifyListener(null, null);
        // set a new leader, see that a registration happens 
        testLeaderService.notifyListener(address2, leaderId2);
        verify(rmGateway2).registerTaskExecutor(eq(leaderId2), eq(taskManagerAddress), eq(resourceID), eq(slotReport), any(Time.class));
        assertNotNull(taskManager.getResourceManagerConnection());
        // check if a concurrent error occurred
        testingFatalErrorHandler.rethrowError();
    } finally {
        rpc.stopService();
    }
}

Example 7 with ResourceID

use of org.apache.flink.runtime.clusterframework.types.ResourceID 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();
    }
}

Example 8 with ResourceID

use of org.apache.flink.runtime.clusterframework.types.ResourceID in project flink by apache.

the class SlotProtocolTest method testSlotAvailableRequest.

/**
	 * Tests whether
	 * 1) a SlotRequest is routed to the SlotManager
	 * 2) a SlotRequest is confirmed
	 * 3) a SlotRequest leads to an allocation of a registered slot
	 * 4) a SlotRequest is routed to the TaskExecutor
	 */
@Test
public void testSlotAvailableRequest() throws Exception {
    final String rmAddress = "/rm1";
    final String jmAddress = "/jm1";
    final String tmAddress = "/tm1";
    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);
    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);
    ResourceManagerConfiguration resourceManagerConfiguration = new ResourceManagerConfiguration(Time.seconds(5L), Time.seconds(5L), Time.minutes(5L));
    JobLeaderIdService jobLeaderIdService = new JobLeaderIdService(testingHaServices, testRpcService.getScheduledExecutor(), resourceManagerConfiguration.getJobTimeout());
    TestingSlotManagerFactory slotManagerFactory = new TestingSlotManagerFactory();
    ResourceManager<ResourceID> resourceManager = Mockito.spy(new StandaloneResourceManager(testRpcService, resourceManagerConfiguration, testingHaServices, slotManagerFactory, mock(MetricRegistry.class), jobLeaderIdService, mock(FatalErrorHandler.class)));
    resourceManager.start();
    rmLeaderElectionService.isLeader(rmLeaderID);
    Thread.sleep(1000);
    Future<RegistrationResponse> registrationFuture = resourceManager.registerJobManager(rmLeaderID, jmLeaderID, jmAddress, jobID);
    try {
        registrationFuture.get(5L, TimeUnit.SECONDS);
    } catch (Exception e) {
        Assert.fail("JobManager registration Future didn't become ready.");
    }
    final SlotManager slotManager = slotManagerFactory.slotManager;
    final ResourceID resourceID = ResourceID.generate();
    final AllocationID allocationID = new AllocationID();
    final ResourceProfile resourceProfile = new ResourceProfile(1.0, 100);
    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);
    SlotRequest slotRequest = new SlotRequest(jobID, allocationID, resourceProfile);
    RMSlotRequestReply slotRequestReply = resourceManager.requestSlot(jmLeaderID, rmLeaderID, slotRequest);
    // 1) a SlotRequest is routed to the SlotManager
    verify(slotManager).requestSlot(slotRequest);
    // 2) a SlotRequest is confirmed
    Assert.assertEquals(slotRequestReply.getAllocationID(), allocationID);
    // 3) a SlotRequest leads to an allocation of a registered slot
    Assert.assertTrue(slotManager.isAllocated(slotID));
    Assert.assertTrue(slotManager.isAllocated(allocationID));
    // 4) a SlotRequest is routed to the TaskExecutor
    verify(taskExecutorGateway, timeout(5000)).requestSlot(eq(slotID), eq(jobID), eq(allocationID), any(String.class), any(UUID.class), any(Time.class));
}

Example 9 with ResourceID

use of org.apache.flink.runtime.clusterframework.types.ResourceID in project flink by apache.

the class SlotManagerTest method testMultipleSlotRequestsWithOneSlot.

/**
	 * Tests multiple slot requests with one slots.
	 */
@Test
public void testMultipleSlotRequestsWithOneSlot() {
    TestingSlotManager slotManager = new TestingSlotManager();
    final AllocationID allocationID = new AllocationID();
    SlotRequest request1 = new SlotRequest(new JobID(), allocationID, DEFAULT_TESTING_PROFILE);
    slotManager.requestSlot(request1);
    final ResourceID resourceID = ResourceID.generate();
    final SlotStatus slotStatus = new SlotStatus(new SlotID(resourceID, 0), DEFAULT_TESTING_PROFILE);
    final SlotReport slotReport = new SlotReport(slotStatus);
    slotManager.registerTaskExecutor(resourceID, taskExecutorRegistration, slotReport);
    // another request pending
    SlotRequest request2 = new SlotRequest(new JobID(), new AllocationID(), DEFAULT_TESTING_PROFILE);
    slotManager.requestSlot(request2);
    assertEquals(1, slotManager.getAllocatedSlotCount());
    assertEquals(0, slotManager.getFreeSlotCount());
    assertEquals(1, slotManager.getPendingRequestCount());
    assertTrue(slotManager.isAllocated(allocationID));
    assertTrue(slotManager.isAllocated(request1.getAllocationId()));
    // but slot is reported empty in a report in the meantime which shouldn't affect the state
    slotManager.notifySlotAvailable(resourceID, slotStatus.getSlotID());
    assertEquals(1, slotManager.getAllocatedSlotCount());
    assertEquals(0, slotManager.getFreeSlotCount());
    assertEquals(0, slotManager.getPendingRequestCount());
    assertTrue(slotManager.isAllocated(slotStatus.getSlotID()));
    assertTrue(slotManager.isAllocated(request2.getAllocationId()));
    // but slot is reported empty in a report in the meantime which shouldn't affect the state
    slotManager.notifySlotAvailable(resourceID, slotStatus.getSlotID());
    assertEquals(0, slotManager.getAllocatedSlotCount());
    assertEquals(1, slotManager.getFreeSlotCount());
    assertEquals(0, slotManager.getPendingRequestCount());
}

Example 10 with ResourceID

use of org.apache.flink.runtime.clusterframework.types.ResourceID in project flink by apache.

the class MesosFlinkResourceManager method reacceptRegisteredWorkers.

/**
	 * Accept the given registered workers into the internal state.
	 *
	 * @param toConsolidate The worker IDs known previously to the JobManager.
	 * @return A collection of registered worker node records.
	 */
@Override
protected Collection<RegisteredMesosWorkerNode> reacceptRegisteredWorkers(Collection<ResourceID> toConsolidate) {
    // we check for each task manager if we recognize its Mesos task ID
    List<RegisteredMesosWorkerNode> accepted = new ArrayList<>(toConsolidate.size());
    for (ResourceID resourceID : toConsolidate) {
        MesosWorkerStore.Worker worker = workersInLaunch.remove(resourceID);
        if (worker != null) {
            LOG.info("Mesos worker consolidation recognizes TaskManager {}.", resourceID);
            accepted.add(new RegisteredMesosWorkerNode(worker));
        } else {
            if (isStarted(resourceID)) {
                LOG.info("TaskManager {} has already been registered at the resource manager.", resourceID);
            } else {
                LOG.info("Mesos worker consolidation does not recognize TaskManager {}.", resourceID);
            }
        }
    }
    return accepted;
}