Examples with TestingResourceManagerGateway org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway used on opensource projects

Example 36 with TestingResourceManagerGateway

use of org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway in project flink by apache.

the class TaskExecutorTest method testSlotReportDoesNotContainStaleInformation.

/**
 * Tests that the {@link SlotReport} sent to the RM does not contain out dated/stale information
 * as slots are being requested from the TM.
 *
 * <p>This is a probabilistic test case and needs to be executed several times to produce a
 * failure without the fix for FLINK-12865.
 */
@Test
public void testSlotReportDoesNotContainStaleInformation() throws Exception {
    final OneShotLatch receivedSlotRequest = new OneShotLatch();
    final CompletableFuture<Void> verifySlotReportFuture = new CompletableFuture<>();
    final OneShotLatch terminateSlotReportVerification = new OneShotLatch();
    final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
    // Assertions for this test
    testingResourceManagerGateway.setTaskExecutorHeartbeatFunction((ignored, heartbeatPayload) -> {
        try {
            final ArrayList<SlotStatus> slots = Lists.newArrayList(heartbeatPayload.getSlotReport());
            assertThat(slots, hasSize(1));
            final SlotStatus slotStatus = slots.get(0);
            log.info("Received SlotStatus: {}", slotStatus);
            if (receivedSlotRequest.isTriggered()) {
                assertThat(slotStatus.getAllocationID(), is(notNullValue()));
            } else {
                assertThat(slotStatus.getAllocationID(), is(nullValue()));
            }
        } catch (AssertionError e) {
            verifySlotReportFuture.completeExceptionally(e);
        }
        if (terminateSlotReportVerification.isTriggered()) {
            verifySlotReportFuture.complete(null);
        }
        return FutureUtils.completedVoidFuture();
    });
    final CompletableFuture<ResourceID> taskExecutorRegistrationFuture = new CompletableFuture<>();
    testingResourceManagerGateway.setSendSlotReportFunction(ignored -> {
        taskExecutorRegistrationFuture.complete(null);
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
    resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setTaskSlotTable(new AllocateSlotNotifyingTaskSlotTable(receivedSlotRequest)).build();
    final TaskExecutor taskExecutor = createTaskExecutor(taskManagerServices);
    final ResourceID taskExecutorResourceId = taskManagerServices.getUnresolvedTaskManagerLocation().getResourceID();
    taskExecutor.start();
    final TaskExecutorGateway taskExecutorGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
    final ScheduledExecutorService heartbeatExecutor = java.util.concurrent.Executors.newSingleThreadScheduledExecutor();
    try {
        taskExecutorRegistrationFuture.get();
        final OneShotLatch scheduleFirstHeartbeat = new OneShotLatch();
        final ResourceID resourceManagerResourceId = testingResourceManagerGateway.getOwnResourceId();
        final long heartbeatInterval = 5L;
        heartbeatExecutor.scheduleWithFixedDelay(() -> {
            scheduleFirstHeartbeat.trigger();
            taskExecutorGateway.heartbeatFromResourceManager(resourceManagerResourceId);
        }, 0L, heartbeatInterval, TimeUnit.MILLISECONDS);
        scheduleFirstHeartbeat.await();
        taskExecutorGateway.requestSlot(new SlotID(taskExecutorResourceId, 0), jobId, new AllocationID(), ResourceProfile.ZERO, "foobar", testingResourceManagerGateway.getFencingToken(), timeout).get();
        terminateSlotReportVerification.trigger();
        verifySlotReportFuture.get();
    } finally {
        ExecutorUtils.gracefulShutdown(timeout.toMilliseconds(), TimeUnit.MILLISECONDS, heartbeatExecutor);
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 37 with TestingResourceManagerGateway

use of org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway in project flink by apache.

the class TaskExecutorTest method runJobManagerHeartbeatTest.

private void runJobManagerHeartbeatTest(ResourceID jmResourceId, HeartbeatServices heartbeatServices, Consumer<TestingJobMasterGatewayBuilder> jobMasterGatewayBuilderConsumer, TriConsumer<ResourceID, TaskExecutorGateway, AllocationID> heartbeatAction) throws IOException, InterruptedException, ExecutionException, TimeoutException {
    final JobLeaderService jobLeaderService = new DefaultJobLeaderService(unresolvedTaskManagerLocation, RetryingRegistrationConfiguration.defaultConfiguration());
    final String jobMasterAddress = "jm";
    final UUID jmLeaderId = UUID.randomUUID();
    final CountDownLatch registrationAttempts = new CountDownLatch(2);
    final OneShotLatch slotOfferedLatch = new OneShotLatch();
    final CompletableFuture<ResourceID> disconnectTaskManagerFuture = new CompletableFuture<>();
    final TestingJobMasterGatewayBuilder testingJobMasterGatewayBuilder = new TestingJobMasterGatewayBuilder().setRegisterTaskManagerFunction((ignoredJobId, ignoredTaskManagerRegistrationInformation) -> {
        registrationAttempts.countDown();
        return CompletableFuture.completedFuture(new JMTMRegistrationSuccess(jmResourceId));
    }).setDisconnectTaskManagerFunction(resourceID -> {
        disconnectTaskManagerFuture.complete(resourceID);
        return CompletableFuture.completedFuture(Acknowledge.get());
    }).setOfferSlotsFunction((resourceID, slotOffers) -> {
        slotOfferedLatch.trigger();
        return CompletableFuture.completedFuture(slotOffers);
    });
    jobMasterGatewayBuilderConsumer.accept(testingJobMasterGatewayBuilder);
    final TestingJobMasterGateway jobMasterGateway = testingJobMasterGatewayBuilder.build();
    final TaskExecutorLocalStateStoresManager localStateStoresManager = createTaskExecutorLocalStateStoresManager();
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).setTaskSlotTable(TaskSlotUtils.createTaskSlotTable(1)).setJobLeaderService(jobLeaderService).setTaskStateManager(localStateStoresManager).build();
    final TestingTaskExecutor taskManager = createTestingTaskExecutor(taskManagerServices, heartbeatServices);
    final OneShotLatch slotReportReceived = new OneShotLatch();
    final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
    testingResourceManagerGateway.setSendSlotReportFunction(ignored -> {
        slotReportReceived.trigger();
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    final Queue<CompletableFuture<RegistrationResponse>> registrationResponses = new ArrayDeque<>();
    registrationResponses.add(CompletableFuture.completedFuture(new TaskExecutorRegistrationSuccess(new InstanceID(), testingResourceManagerGateway.getOwnResourceId(), new ClusterInformation("foobar", 1234))));
    registrationResponses.add(new CompletableFuture<>());
    testingResourceManagerGateway.setRegisterTaskExecutorFunction(taskExecutorRegistration -> registrationResponses.poll());
    rpc.registerGateway(jobMasterAddress, jobMasterGateway);
    rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
    try {
        taskManager.start();
        taskManager.waitUntilStarted();
        final TaskExecutorGateway taskExecutorGateway = taskManager.getSelfGateway(TaskExecutorGateway.class);
        resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
        slotReportReceived.await();
        final AllocationID allocationId = new AllocationID();
        requestSlot(taskExecutorGateway, jobId, allocationId, buildSlotID(0), ResourceProfile.UNKNOWN, jobMasterAddress, testingResourceManagerGateway.getFencingToken());
        // now inform the task manager about the new job leader
        jobManagerLeaderRetriever.notifyListener(jobMasterAddress, jmLeaderId);
        // register task manager success will trigger monitoring heartbeat target between tm and
        // jm
        slotOfferedLatch.await();
        heartbeatAction.accept(unresolvedTaskManagerLocation.getResourceID(), taskExecutorGateway, allocationId);
        // the timeout should trigger disconnecting from the JobManager
        final ResourceID resourceID = disconnectTaskManagerFuture.get();
        assertThat(resourceID, equalTo(unresolvedTaskManagerLocation.getResourceID()));
        assertTrue("The TaskExecutor should try to reconnect to the JM", registrationAttempts.await(timeout.toMilliseconds(), TimeUnit.SECONDS));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskManager, timeout);
    }
}

Example 38 with TestingResourceManagerGateway

use of org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway in project flink by apache.

the class TaskExecutorTest method testRegisterWithDefaultSlotResourceProfile.

@Test
public void testRegisterWithDefaultSlotResourceProfile() throws Exception {
    final int numberOfSlots = 2;
    final TaskExecutor taskExecutor = createTaskExecutor(numberOfSlots);
    taskExecutor.start();
    try {
        final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
        final CompletableFuture<ResourceProfile> registeredDefaultSlotResourceProfileFuture = new CompletableFuture<>();
        final ResourceID ownResourceId = testingResourceManagerGateway.getOwnResourceId();
        testingResourceManagerGateway.setRegisterTaskExecutorFunction(taskExecutorRegistration -> {
            registeredDefaultSlotResourceProfileFuture.complete(taskExecutorRegistration.getDefaultSlotResourceProfile());
            return CompletableFuture.completedFuture(new TaskExecutorRegistrationSuccess(new InstanceID(), ownResourceId, new ClusterInformation("localhost", 1234)));
        });
        rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
        resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
        assertThat(registeredDefaultSlotResourceProfileFuture.get(), equalTo(TaskExecutorResourceUtils.generateDefaultSlotResourceProfile(TM_RESOURCE_SPEC, numberOfSlots)));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 39 with TestingResourceManagerGateway

use of org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway in project flink by apache.

the class TaskExecutorTest method runResourceManagerHeartbeatTest.

private void runResourceManagerHeartbeatTest(HeartbeatServices heartbeatServices, Consumer<TestingResourceManagerGateway> setupResourceManagerGateway, TriConsumerWithException<TaskExecutorGateway, ResourceID, CompletableFuture<ResourceID>, Exception> heartbeatAction) throws Exception {
    final String rmAddress = "rm";
    final ResourceID rmResourceId = new ResourceID(rmAddress);
    final ResourceManagerId rmLeaderId = ResourceManagerId.generate();
    TestingResourceManagerGateway rmGateway = new TestingResourceManagerGateway(rmLeaderId, rmResourceId, rmAddress, rmAddress);
    final TaskExecutorRegistrationSuccess registrationResponse = new TaskExecutorRegistrationSuccess(new InstanceID(), rmResourceId, new ClusterInformation("localhost", 1234));
    final Queue<CompletableFuture<RegistrationResponse>> registrationResponses = new ArrayDeque<>(2);
    registrationResponses.add(CompletableFuture.completedFuture(registrationResponse));
    registrationResponses.add(new CompletableFuture<>());
    final CompletableFuture<ResourceID> taskExecutorRegistrationFuture = new CompletableFuture<>();
    final CountDownLatch registrationAttempts = new CountDownLatch(2);
    rmGateway.setRegisterTaskExecutorFunction(registration -> {
        taskExecutorRegistrationFuture.complete(registration.getResourceId());
        registrationAttempts.countDown();
        return registrationResponses.poll();
    });
    setupResourceManagerGateway.accept(rmGateway);
    final CompletableFuture<ResourceID> taskExecutorDisconnectFuture = new CompletableFuture<>();
    rmGateway.setDisconnectTaskExecutorConsumer(disconnectInfo -> taskExecutorDisconnectFuture.complete(disconnectInfo.f0));
    rpc.registerGateway(rmAddress, rmGateway);
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setUnresolvedTaskManagerLocation(unresolvedTaskManagerLocation).build();
    final TaskExecutor taskManager = createTaskExecutor(taskManagerServices, heartbeatServices);
    try {
        taskManager.start();
        final TaskExecutorGateway taskExecutorGateway = taskManager.getSelfGateway(TaskExecutorGateway.class);
        // define a leader and see that a registration happens
        resourceManagerLeaderRetriever.notifyListener(rmAddress, rmLeaderId.toUUID());
        // register resource manager success will trigger monitoring heartbeat target between tm
        // and rm
        assertThat(taskExecutorRegistrationFuture.get(), equalTo(unresolvedTaskManagerLocation.getResourceID()));
        heartbeatAction.accept(taskExecutorGateway, rmGateway.getOwnResourceId(), taskExecutorDisconnectFuture);
        // heartbeat timeout should trigger disconnect TaskManager from ResourceManager
        assertThat(taskExecutorDisconnectFuture.get(timeout.toMilliseconds(), TimeUnit.MILLISECONDS), equalTo(unresolvedTaskManagerLocation.getResourceID()));
        assertTrue("The TaskExecutor should try to reconnect to the RM", registrationAttempts.await(timeout.toMilliseconds(), TimeUnit.SECONDS));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskManager, timeout);
    }
}

Example 40 with TestingResourceManagerGateway

use of org.apache.flink.runtime.resourcemanager.utils.TestingResourceManagerGateway in project flink by apache.

the class TaskExecutorTest method createAndRegisterResourceManager.

private ResourceManagerId createAndRegisterResourceManager(CompletableFuture<Tuple3<ResourceID, InstanceID, SlotReport>> initialSlotReportFuture) {
    final TestingResourceManagerGateway resourceManagerGateway = new TestingResourceManagerGateway();
    resourceManagerGateway.setSendSlotReportFunction(resourceIDInstanceIDSlotReportTuple3 -> {
        initialSlotReportFuture.complete(resourceIDInstanceIDSlotReportTuple3);
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    rpc.registerGateway(resourceManagerGateway.getAddress(), resourceManagerGateway);
    // tell the task manager about the rm leader
    resourceManagerLeaderRetriever.notifyListener(resourceManagerGateway.getAddress(), resourceManagerGateway.getFencingToken().toUUID());
    return resourceManagerGateway.getFencingToken();
}