Examples with TaskExecutorLocalStateStoresManager org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager used on opensource projects

Example 1 with TaskExecutorLocalStateStoresManager

use of org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager in project flink by apache.

the class TaskExecutorPartitionLifecycleTest method internalTestPartitionRelease.

private void internalTestPartitionRelease(TaskExecutorPartitionTracker partitionTracker, ShuffleEnvironment<?, ?> shuffleEnvironment, CompletableFuture<ResultPartitionID> startTrackingFuture, TestAction testAction) throws Exception {
    final ResultPartitionDeploymentDescriptor taskResultPartitionDescriptor = PartitionTestUtils.createPartitionDeploymentDescriptor(ResultPartitionType.BLOCKING);
    final ExecutionAttemptID eid1 = taskResultPartitionDescriptor.getShuffleDescriptor().getResultPartitionID().getProducerId();
    final TaskDeploymentDescriptor taskDeploymentDescriptor = TaskExecutorSubmissionTest.createTaskDeploymentDescriptor(jobId, "job", eid1, new SerializedValue<>(new ExecutionConfig()), "Sender", 1, 0, 1, 0, new Configuration(), new Configuration(), TestingInvokable.class.getName(), Collections.singletonList(taskResultPartitionDescriptor), Collections.emptyList(), Collections.emptyList(), Collections.emptyList());
    final TaskSlotTable<Task> taskSlotTable = createTaskSlotTable();
    final TaskExecutorLocalStateStoresManager localStateStoresManager = new TaskExecutorLocalStateStoresManager(false, Reference.owned(new File[] { tmp.newFolder() }), Executors.directExecutor());
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setTaskSlotTable(taskSlotTable).setTaskStateManager(localStateStoresManager).setShuffleEnvironment(shuffleEnvironment).build();
    final CompletableFuture<Void> taskFinishedFuture = new CompletableFuture<>();
    final OneShotLatch slotOfferedLatch = new OneShotLatch();
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setRegisterTaskManagerFunction((ignoredJobId, ignoredTaskManagerRegistrationInformation) -> CompletableFuture.completedFuture(new JMTMRegistrationSuccess(ResourceID.generate()))).setOfferSlotsFunction((resourceID, slotOffers) -> {
        slotOfferedLatch.trigger();
        return CompletableFuture.completedFuture(slotOffers);
    }).setUpdateTaskExecutionStateFunction(taskExecutionState -> {
        if (taskExecutionState.getExecutionState() == ExecutionState.FINISHED) {
            taskFinishedFuture.complete(null);
        }
        return CompletableFuture.completedFuture(Acknowledge.get());
    }).build();
    final TestingTaskExecutor taskExecutor = createTestingTaskExecutor(taskManagerServices, partitionTracker);
    final CompletableFuture<SlotReport> initialSlotReportFuture = new CompletableFuture<>();
    final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
    testingResourceManagerGateway.setSendSlotReportFunction(resourceIDInstanceIDSlotReportTuple3 -> {
        initialSlotReportFuture.complete(resourceIDInstanceIDSlotReportTuple3.f2);
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    testingResourceManagerGateway.setRegisterTaskExecutorFunction(input -> CompletableFuture.completedFuture(new TaskExecutorRegistrationSuccess(new InstanceID(), testingResourceManagerGateway.getOwnResourceId(), new ClusterInformation("blobServerHost", 55555))));
    try {
        taskExecutor.start();
        taskExecutor.waitUntilStarted();
        final TaskExecutorGateway taskExecutorGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        final String jobMasterAddress = "jm";
        rpc.registerGateway(jobMasterAddress, jobMasterGateway);
        rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
        // inform the task manager about the job leader
        taskManagerServices.getJobLeaderService().addJob(jobId, jobMasterAddress);
        jobManagerLeaderRetriever.notifyListener(jobMasterAddress, UUID.randomUUID());
        resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
        final Optional<SlotStatus> slotStatusOptional = StreamSupport.stream(initialSlotReportFuture.get().spliterator(), false).findAny();
        assertTrue(slotStatusOptional.isPresent());
        final SlotStatus slotStatus = slotStatusOptional.get();
        while (true) {
            try {
                taskExecutorGateway.requestSlot(slotStatus.getSlotID(), jobId, taskDeploymentDescriptor.getAllocationId(), ResourceProfile.ZERO, jobMasterAddress, testingResourceManagerGateway.getFencingToken(), timeout).get();
                break;
            } catch (Exception e) {
                // the proper establishment of the RM connection is tracked
                // asynchronously, so we have to poll here until it went through
                // until then, slot requests will fail with an exception
                Thread.sleep(50);
            }
        }
        TestingInvokable.sync = new BlockerSync();
        // Wait till the slot has been successfully offered before submitting the task.
        // This ensures TM has been successfully registered to JM.
        slotOfferedLatch.await();
        taskExecutorGateway.submitTask(taskDeploymentDescriptor, jobMasterGateway.getFencingToken(), timeout).get();
        TestingInvokable.sync.awaitBlocker();
        // the task is still running => the partition is in in-progress and should be tracked
        assertThat(startTrackingFuture.get(), equalTo(taskResultPartitionDescriptor.getShuffleDescriptor().getResultPartitionID()));
        TestingInvokable.sync.releaseBlocker();
        taskFinishedFuture.get(timeout.getSize(), timeout.getUnit());
        testAction.accept(jobId, taskResultPartitionDescriptor, taskExecutor, taskExecutorGateway);
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
    // the shutdown of the backing shuffle environment releases all partitions
    // the book-keeping is not aware of this
    assertTrue(shuffleEnvironment.getPartitionsOccupyingLocalResources().isEmpty());
}

Example 2 with TaskExecutorLocalStateStoresManager

use of org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager in project flink by apache.

the class TaskExecutorTest method testFreeingInactiveSlotDoesNotFail.

/**
 * Tests that freeing an inactive slot is a legal operation that does not throw an exception.
 */
@Test
public void testFreeingInactiveSlotDoesNotFail() throws Exception {
    final OneShotLatch taskExecutorIsRegistered = new OneShotLatch();
    final CompletableFuture<Tuple3<InstanceID, SlotID, AllocationID>> availableSlotFuture = new CompletableFuture<>();
    final TestingResourceManagerGateway resourceManagerGateway = createRmWithTmRegisterAndNotifySlotHooks(new InstanceID(), taskExecutorIsRegistered, availableSlotFuture);
    rpc.registerGateway(resourceManagerGateway.getAddress(), resourceManagerGateway);
    final MultiShotLatch offerSlotsLatch = new MultiShotLatch();
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setOfferSlotsFunction((resourceID, slotOffers) -> {
        offerSlotsLatch.trigger();
        return new CompletableFuture<>();
    }).build();
    rpc.registerGateway(jobMasterGateway.getAddress(), jobMasterGateway);
    final TaskSlotTable<Task> taskSlotTable = TaskSlotUtils.createTaskSlotTable(1);
    final TaskExecutorLocalStateStoresManager localStateStoresManager = createTaskExecutorLocalStateStoresManager();
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).setTaskSlotTable(taskSlotTable).setTaskStateManager(localStateStoresManager).build();
    final TestingTaskExecutor taskExecutor = createTestingTaskExecutor(taskManagerServices);
    final ThreadSafeTaskSlotTable<Task> threadSafeTaskSlotTable = new ThreadSafeTaskSlotTable<>(taskSlotTable, taskExecutor.getMainThreadExecutableForTesting());
    try {
        taskExecutor.start();
        taskExecutor.waitUntilStarted();
        final TaskExecutorGateway tmGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        taskExecutorIsRegistered.await();
        jobManagerLeaderRetriever.notifyListener(jobMasterGateway.getAddress(), jobMasterGateway.getFencingToken().toUUID());
        final AllocationID allocationId = new AllocationID();
        requestSlot(tmGateway, jobId, allocationId, buildSlotID(0), ResourceProfile.UNKNOWN, jobMasterGateway.getAddress(), resourceManagerGateway.getFencingToken());
        offerSlotsLatch.await();
        tmGateway.freeSlot(allocationId, new RuntimeException("test exception"), timeout).get();
        assertThat(availableSlotFuture.get().f2, is(allocationId));
        assertThat(threadSafeTaskSlotTable.getAllocationIdsPerJob(jobId), empty());
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 3 with TaskExecutorLocalStateStoresManager

use of org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager in project flink by apache.

the class TaskExecutorTest method testHeartbeatReporting.

/**
 * Tests that the correct partition/slot report is sent as part of the heartbeat response.
 */
@Test
public void testHeartbeatReporting() throws Exception {
    final String rmAddress = "rm";
    final UUID rmLeaderId = UUID.randomUUID();
    // register the mock resource manager gateway
    final TestingResourceManagerGateway rmGateway = new TestingResourceManagerGateway();
    final CompletableFuture<ResourceID> taskExecutorRegistrationFuture = new CompletableFuture<>();
    final ResourceID rmResourceId = rmGateway.getOwnResourceId();
    final CompletableFuture<RegistrationResponse> registrationResponse = CompletableFuture.completedFuture(new TaskExecutorRegistrationSuccess(new InstanceID(), rmResourceId, new ClusterInformation("localhost", 1234)));
    rmGateway.setRegisterTaskExecutorFunction(taskExecutorRegistration -> {
        taskExecutorRegistrationFuture.complete(taskExecutorRegistration.getResourceId());
        return registrationResponse;
    });
    final CompletableFuture<SlotReport> initialSlotReportFuture = new CompletableFuture<>();
    rmGateway.setSendSlotReportFunction(resourceIDInstanceIDSlotReportTuple3 -> {
        initialSlotReportFuture.complete(resourceIDInstanceIDSlotReportTuple3.f2);
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    final CompletableFuture<TaskExecutorHeartbeatPayload> heartbeatPayloadCompletableFuture = new CompletableFuture<>();
    rmGateway.setTaskExecutorHeartbeatFunction((resourceID, heartbeatPayload) -> {
        heartbeatPayloadCompletableFuture.complete(heartbeatPayload);
        return FutureUtils.completedVoidFuture();
    });
    rpc.registerGateway(rmAddress, rmGateway);
    final SlotID slotId = buildSlotID(0);
    final ResourceProfile resourceProfile = ResourceProfile.fromResources(1.0, 1);
    final SlotReport slotReport1 = new SlotReport(new SlotStatus(slotId, resourceProfile));
    final SlotReport slotReport2 = new SlotReport(new SlotStatus(slotId, resourceProfile, new JobID(), new AllocationID()));
    final Queue<SlotReport> reports = new ArrayDeque<>(Arrays.asList(slotReport1, slotReport2));
    final TaskSlotTable<Task> taskSlotTable = TestingTaskSlotTable.<Task>newBuilder().createSlotReportSupplier(reports::poll).closeAsyncReturns(CompletableFuture.completedFuture(null)).build();
    final TaskExecutorLocalStateStoresManager localStateStoresManager = createTaskExecutorLocalStateStoresManager();
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).setTaskSlotTable(taskSlotTable).setTaskStateManager(localStateStoresManager).build();
    final TaskExecutorPartitionTracker partitionTracker = createPartitionTrackerWithFixedPartitionReport(taskManagerServices.getShuffleEnvironment());
    final TaskExecutor taskManager = createTaskExecutor(taskManagerServices, HEARTBEAT_SERVICES, partitionTracker);
    try {
        taskManager.start();
        // define a leader and see that a registration happens
        resourceManagerLeaderRetriever.notifyListener(rmAddress, rmLeaderId);
        // register resource manager success will trigger monitoring heartbeat target between tm
        // and rm
        assertThat(taskExecutorRegistrationFuture.get(), equalTo(unresolvedTaskManagerLocation.getResourceID()));
        assertThat(initialSlotReportFuture.get(), equalTo(slotReport1));
        TaskExecutorGateway taskExecutorGateway = taskManager.getSelfGateway(TaskExecutorGateway.class);
        // trigger the heartbeat asynchronously
        taskExecutorGateway.heartbeatFromResourceManager(rmResourceId);
        // wait for heartbeat response
        SlotReport actualSlotReport = heartbeatPayloadCompletableFuture.get().getSlotReport();
        // the new slot report should be reported
        assertEquals(slotReport2, actualSlotReport);
        ClusterPartitionReport actualClusterPartitionReport = heartbeatPayloadCompletableFuture.get().getClusterPartitionReport();
        assertEquals(partitionTracker.createClusterPartitionReport(), actualClusterPartitionReport);
    } finally {
        RpcUtils.terminateRpcEndpoint(taskManager, timeout);
    }
}

Example 4 with TaskExecutorLocalStateStoresManager

use of org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager in project flink by apache.

the class TaskExecutorTest method testTriggerRegistrationOnLeaderChange.

@Test
public void testTriggerRegistrationOnLeaderChange() throws Exception {
    final UUID leaderId1 = UUID.randomUUID();
    final UUID leaderId2 = UUID.randomUUID();
    // register the mock resource manager gateways
    final CompletableFuture<TaskExecutorRegistration> rmGateway1TaskExecutorRegistration = new CompletableFuture<>();
    TestingResourceManagerGateway rmGateway1 = new TestingResourceManagerGateway();
    rmGateway1.setRegisterTaskExecutorFunction(taskExecutorRegistration -> {
        rmGateway1TaskExecutorRegistration.complete(taskExecutorRegistration);
        return createRegistrationResponse(rmGateway1);
    });
    final CompletableFuture<TaskExecutorRegistration> rmGateway2TaskExecutorRegistration = new CompletableFuture<>();
    TestingResourceManagerGateway rmGateway2 = new TestingResourceManagerGateway();
    rmGateway2.setRegisterTaskExecutorFunction(taskExecutorRegistration -> {
        rmGateway2TaskExecutorRegistration.complete(taskExecutorRegistration);
        return createRegistrationResponse(rmGateway2);
    });
    rpc.registerGateway(rmGateway1.getAddress(), rmGateway1);
    rpc.registerGateway(rmGateway2.getAddress(), rmGateway2);
    final TaskSlotTable<Task> taskSlotTable = TestingTaskSlotTable.<Task>newBuilder().createSlotReportSupplier(SlotReport::new).closeAsyncReturns(CompletableFuture.completedFuture(null)).build();
    final TaskExecutorLocalStateStoresManager localStateStoresManager = createTaskExecutorLocalStateStoresManager();
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).setTaskSlotTable(taskSlotTable).setTaskStateManager(localStateStoresManager).build();
    TaskExecutor taskManager = createTaskExecutor(taskManagerServices);
    try {
        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
        resourceManagerLeaderRetriever.notifyListener(rmGateway1.getAddress(), leaderId1);
        final TaskExecutorRegistration taskExecutorRegistration1 = rmGateway1TaskExecutorRegistration.join();
        assertThat(taskExecutorRegistration1.getTaskExecutorAddress(), is(taskManagerAddress));
        assertThat(taskExecutorRegistration1.getResourceId(), is(unresolvedTaskManagerLocation.getResourceID()));
        assertNotNull(taskManager.getResourceManagerConnection());
        // cancel the leader
        resourceManagerLeaderRetriever.notifyListener(null, null);
        // set a new leader, see that a registration happens
        resourceManagerLeaderRetriever.notifyListener(rmGateway2.getAddress(), leaderId2);
        final TaskExecutorRegistration taskExecutorRegistration2 = rmGateway2TaskExecutorRegistration.join();
        assertThat(taskExecutorRegistration2.getTaskExecutorAddress(), is(taskManagerAddress));
        assertThat(taskExecutorRegistration2.getResourceId(), is(unresolvedTaskManagerLocation.getResourceID()));
        assertNotNull(taskManager.getResourceManagerConnection());
    } finally {
        RpcUtils.terminateRpcEndpoint(taskManager, timeout);
    }
}

Example 5 with TaskExecutorLocalStateStoresManager

use of org.apache.flink.runtime.state.TaskExecutorLocalStateStoresManager in project flink by apache.

the class TaskExecutorTest method createTaskExecutorTestingContext.

private TaskExecutorTestingContext createTaskExecutorTestingContext(final TaskSlotTable<Task> taskSlotTable) throws IOException {
    final OneShotLatch offerSlotsLatch = new OneShotLatch();
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setOfferSlotsFunction((resourceID, slotOffers) -> {
        offerSlotsLatch.trigger();
        return CompletableFuture.completedFuture(slotOffers);
    }).build();
    rpc.registerGateway(jobMasterGateway.getAddress(), jobMasterGateway);
    final JobLeaderService jobLeaderService = new DefaultJobLeaderService(unresolvedTaskManagerLocation, RetryingRegistrationConfiguration.defaultConfiguration());
    TaskExecutorLocalStateStoresManager stateStoresManager = createTaskExecutorLocalStateStoresManager();
    TaskExecutorStateChangelogStoragesManager changelogStoragesManager = new TaskExecutorStateChangelogStoragesManager();
    TaskManagerMetricGroup metricGroup = TaskManagerMetricGroup.createTaskManagerMetricGroup(NoOpMetricRegistry.INSTANCE, "", ResourceID.generate());
    final TestingTaskExecutor taskExecutor = createTestingTaskExecutor(new TaskManagerServicesBuilder().setTaskSlotTable(taskSlotTable).setJobLeaderService(jobLeaderService).setTaskStateManager(stateStoresManager).setTaskChangelogStoragesManager(changelogStoragesManager).build(), HEARTBEAT_SERVICES, metricGroup);
    jobManagerLeaderRetriever.notifyListener(jobMasterGateway.getAddress(), jobMasterGateway.getFencingToken().toUUID());
    return new TaskExecutorTestingContext(jobMasterGateway, taskSlotTable, taskExecutor, changelogStoragesManager, metricGroup, offerSlotsLatch);
}