Examples with TestingJobMasterGateway org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway used on opensource projects

Example 21 with TestingJobMasterGateway

use of org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway in project flink by apache.

the class TaskExecutorTest method testReleaseInactiveSlots.

@Test
public void testReleaseInactiveSlots() throws Exception {
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setTaskSlotTable(TaskSlotUtils.createTaskSlotTable(1)).build();
    final TaskExecutor taskExecutor = createTaskExecutor(taskManagerServices, HEARTBEAT_SERVICES);
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setRegisterTaskManagerFunction((ignoredJobId, ignoredTaskManagerRegistrationInformation) -> new CompletableFuture<>()).build();
    rpc.registerGateway(jobMasterGateway.getAddress(), jobMasterGateway);
    final InstanceID registrationId = new InstanceID();
    final OneShotLatch taskExecutorIsRegistered = new OneShotLatch();
    final CompletableFuture<Tuple3<InstanceID, SlotID, AllocationID>> availableSlotFuture = new CompletableFuture<>();
    final TestingResourceManagerGateway resourceManagerGateway = createRmWithTmRegisterAndNotifySlotHooks(registrationId, taskExecutorIsRegistered, availableSlotFuture);
    rpc.registerGateway(resourceManagerGateway.getAddress(), resourceManagerGateway);
    resourceManagerLeaderRetriever.notifyListener(resourceManagerGateway.getAddress(), resourceManagerGateway.getFencingToken().toUUID());
    try {
        taskExecutor.start();
        final TaskExecutorGateway taskExecutorGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        taskExecutorIsRegistered.await();
        final AllocationID allocationId = new AllocationID();
        final SlotID slotId = new SlotID(taskExecutor.getResourceID(), 0);
        requestSlot(taskExecutorGateway, jobId, allocationId, slotId, ResourceProfile.UNKNOWN, jobMasterGateway.getAddress(), resourceManagerGateway.getFencingToken());
        taskExecutorGateway.freeInactiveSlots(jobId, timeout);
        // the slot should be freed
        assertThat(availableSlotFuture.get().f1, is(slotId));
        assertThat(availableSlotFuture.get().f2, is(allocationId));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 22 with TestingJobMasterGateway

use of org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway in project flink by apache.

the class TaskExecutorTest method testSlotOfferResponseWithPendingSlotOffer.

/**
 * Tests the behavior of the task executor when a slot offer response is received while a newer
 * slot offer is in progress.
 */
private void testSlotOfferResponseWithPendingSlotOffer(final ResponseOrder responseOrder) throws Exception {
    final OneShotLatch taskExecutorIsRegistered = new OneShotLatch();
    final TestingResourceManagerGateway resourceManagerGateway = createRmWithTmRegisterAndNotifySlotHooks(new InstanceID(), taskExecutorIsRegistered, new CompletableFuture<>());
    final CompletableFuture<Collection<SlotOffer>> firstOfferResponseFuture = new CompletableFuture<>();
    final CompletableFuture<Collection<SlotOffer>> secondOfferResponseFuture = new CompletableFuture<>();
    final Queue<CompletableFuture<Collection<SlotOffer>>> slotOfferResponses = new ArrayDeque<>(Arrays.asList(firstOfferResponseFuture, secondOfferResponseFuture));
    final MultiShotLatch offerSlotsLatch = new MultiShotLatch();
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setOfferSlotsFunction((resourceID, slotOffers) -> {
        offerSlotsLatch.trigger();
        return slotOfferResponses.remove();
    }).build();
    rpc.registerGateway(resourceManagerGateway.getAddress(), resourceManagerGateway);
    rpc.registerGateway(jobMasterGateway.getAddress(), jobMasterGateway);
    final TaskSlotTable<Task> taskSlotTable = TaskSlotUtils.createTaskSlotTable(2);
    final TaskManagerServices taskManagerServices = createTaskManagerServicesWithTaskSlotTable(taskSlotTable);
    final TestingTaskExecutor taskExecutor = createTestingTaskExecutor(taskManagerServices);
    final ThreadSafeTaskSlotTable<Task> threadSafeTaskSlotTable = new ThreadSafeTaskSlotTable<>(taskSlotTable, taskExecutor.getMainThreadExecutableForTesting());
    final SlotOffer slotOffer1 = new SlotOffer(new AllocationID(), 0, ResourceProfile.ANY);
    final SlotOffer slotOffer2 = new SlotOffer(new AllocationID(), 1, ResourceProfile.ANY);
    try {
        taskExecutor.start();
        taskExecutor.waitUntilStarted();
        final TaskExecutorGateway tmGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        // wait until task executor registered at the RM
        taskExecutorIsRegistered.await();
        // notify job leader to start slot offering
        jobManagerLeaderRetriever.notifyListener(jobMasterGateway.getAddress(), jobMasterGateway.getFencingToken().toUUID());
        // request the first slot
        requestSlot(tmGateway, jobId, slotOffer1.getAllocationId(), buildSlotID(slotOffer1.getSlotIndex()), ResourceProfile.UNKNOWN, jobMasterGateway.getAddress(), resourceManagerGateway.getFencingToken());
        // wait until first slot offer as arrived
        offerSlotsLatch.await();
        // request second slot, triggering another offer containing both slots
        int slotIndex = slotOffer2.getSlotIndex();
        requestSlot(tmGateway, jobId, slotOffer2.getAllocationId(), buildSlotID(slotIndex), ResourceProfile.UNKNOWN, jobMasterGateway.getAddress(), resourceManagerGateway.getFencingToken());
        // wait until second slot offer as arrived
        offerSlotsLatch.await();
        switch(responseOrder) {
            case ACCEPT_THEN_REJECT:
                // accept the first offer, but reject both slots for the second offer
                firstOfferResponseFuture.complete(Collections.singletonList(slotOffer1));
                assertThat(threadSafeTaskSlotTable.getActiveTaskSlotAllocationIdsPerJob(jobId), empty());
                secondOfferResponseFuture.complete(Collections.emptyList());
                assertThat(threadSafeTaskSlotTable.getAllocationIdsPerJob(jobId), empty());
                return;
            case REJECT_THEN_ACCEPT:
                // fail the first offer, but accept both slots for the second offer
                // in the past the rejection of the first offer freed the slot; when the slot is
                // accepted from the second offer the activation of said slot then failed
                firstOfferResponseFuture.complete(Collections.emptyList());
                secondOfferResponseFuture.complete(Arrays.asList(slotOffer1, slotOffer2));
                assertThat(threadSafeTaskSlotTable.getAllocationIdsPerJob(jobId), containsInAnyOrder(slotOffer1.getAllocationId(), slotOffer2.getAllocationId()));
                return;
        }
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 23 with TestingJobMasterGateway

use of org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway in project flink by apache.

the class TaskExecutorTest method testSlotOfferCounterIsSeparatedByJob.

@Test
public void testSlotOfferCounterIsSeparatedByJob() throws Exception {
    final OneShotLatch taskExecutorIsRegistered = new OneShotLatch();
    final TestingResourceManagerGateway resourceManagerGateway = createRmWithTmRegisterAndNotifySlotHooks(new InstanceID(), taskExecutorIsRegistered, new CompletableFuture<>());
    final CompletableFuture<Collection<SlotOffer>> firstOfferResponseFuture = new CompletableFuture<>();
    final CompletableFuture<Collection<SlotOffer>> secondOfferResponseFuture = new CompletableFuture<>();
    final Queue<CompletableFuture<Collection<SlotOffer>>> slotOfferResponses = new ArrayDeque<>(Arrays.asList(firstOfferResponseFuture, secondOfferResponseFuture));
    final MultiShotLatch offerSlotsLatch = new MultiShotLatch();
    final TestingJobMasterGateway jobMasterGateway1 = new TestingJobMasterGatewayBuilder().setAddress("jm1").setOfferSlotsFunction((resourceID, slotOffers) -> {
        offerSlotsLatch.trigger();
        return slotOfferResponses.remove();
    }).build();
    final TestingJobMasterGateway jobMasterGateway2 = new TestingJobMasterGatewayBuilder().setAddress("jm2").setOfferSlotsFunction((resourceID, slotOffers) -> {
        offerSlotsLatch.trigger();
        return slotOfferResponses.remove();
    }).build();
    rpc.registerGateway(resourceManagerGateway.getAddress(), resourceManagerGateway);
    rpc.registerGateway(jobMasterGateway1.getAddress(), jobMasterGateway1);
    rpc.registerGateway(jobMasterGateway2.getAddress(), jobMasterGateway2);
    final TaskSlotTable<Task> taskSlotTable = TaskSlotUtils.createTaskSlotTable(2);
    final TaskManagerServices taskManagerServices = createTaskManagerServicesWithTaskSlotTable(taskSlotTable);
    final TestingTaskExecutor taskExecutor = createTestingTaskExecutor(taskManagerServices);
    final ThreadSafeTaskSlotTable<Task> threadSafeTaskSlotTable = new ThreadSafeTaskSlotTable<>(taskSlotTable, taskExecutor.getMainThreadExecutableForTesting());
    final SlotOffer slotOffer1 = new SlotOffer(new AllocationID(), 0, ResourceProfile.ANY);
    final SlotOffer slotOffer2 = new SlotOffer(new AllocationID(), 1, ResourceProfile.ANY);
    try {
        taskExecutor.start();
        taskExecutor.waitUntilStarted();
        final TaskExecutorGateway tmGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        // wait until task executor registered at the RM
        taskExecutorIsRegistered.await();
        // notify job leader to start slot offering
        jobManagerLeaderRetriever.notifyListener(jobMasterGateway1.getAddress(), jobMasterGateway1.getFencingToken().toUUID());
        jobManagerLeaderRetriever2.notifyListener(jobMasterGateway2.getAddress(), jobMasterGateway2.getFencingToken().toUUID());
        // request the first slot
        requestSlot(tmGateway, jobId, slotOffer1.getAllocationId(), buildSlotID(slotOffer1.getSlotIndex()), ResourceProfile.UNKNOWN, jobMasterGateway1.getAddress(), resourceManagerGateway.getFencingToken());
        // wait until first slot offer as arrived
        offerSlotsLatch.await();
        // request second slot, triggering another offer containing both slots
        requestSlot(tmGateway, jobId2, slotOffer2.getAllocationId(), buildSlotID(slotOffer2.getSlotIndex()), ResourceProfile.UNKNOWN, jobMasterGateway2.getAddress(), resourceManagerGateway.getFencingToken());
        // wait until second slot offer as arrived
        offerSlotsLatch.await();
        firstOfferResponseFuture.complete(Collections.singletonList(slotOffer1));
        secondOfferResponseFuture.complete(Collections.singletonList(slotOffer2));
        assertThat(threadSafeTaskSlotTable.getActiveTaskSlotAllocationIdsPerJob(jobId), contains(slotOffer1.getAllocationId()));
        assertThat(threadSafeTaskSlotTable.getActiveTaskSlotAllocationIdsPerJob(jobId2), contains(slotOffer2.getAllocationId()));
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 24 with TestingJobMasterGateway

use of org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway in project flink by apache.

the class TaskExecutorTest method testSyncSlotsWithJobMasterByHeartbeat.

/**
 * Tests that the TaskExecutor syncs its slots view with the JobMaster's view via the
 * AllocatedSlotReport reported by the heartbeat (See FLINK-11059).
 */
@Test
public void testSyncSlotsWithJobMasterByHeartbeat() throws Exception {
    final CountDownLatch activeSlots = new CountDownLatch(2);
    final TaskSlotTable<Task> taskSlotTable = new ActivateSlotNotifyingTaskSlotTable(2, activeSlots);
    final TaskManagerServices taskManagerServices = new TaskManagerServicesBuilder().setTaskSlotTable(taskSlotTable).build();
    final TaskExecutor taskExecutor = createTaskExecutor(taskManagerServices);
    final TestingResourceManagerGateway testingResourceManagerGateway = new TestingResourceManagerGateway();
    final BlockingQueue<AllocationID> allocationsNotifiedFree = new ArrayBlockingQueue<>(2);
    OneShotLatch initialSlotReporting = new OneShotLatch();
    testingResourceManagerGateway.setSendSlotReportFunction(resourceIDInstanceIDSlotReportTuple3 -> {
        initialSlotReporting.trigger();
        return CompletableFuture.completedFuture(Acknowledge.get());
    });
    testingResourceManagerGateway.setNotifySlotAvailableConsumer(instanceIDSlotIDAllocationIDTuple3 -> allocationsNotifiedFree.offer(instanceIDSlotIDAllocationIDTuple3.f2));
    rpc.registerGateway(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway);
    resourceManagerLeaderRetriever.notifyListener(testingResourceManagerGateway.getAddress(), testingResourceManagerGateway.getFencingToken().toUUID());
    final BlockingQueue<AllocationID> failedSlotFutures = new ArrayBlockingQueue<>(2);
    final ResourceID jobManagerResourceId = ResourceID.generate();
    final TestingJobMasterGateway jobMasterGateway = new TestingJobMasterGatewayBuilder().setFailSlotConsumer((resourceID, allocationID, throwable) -> failedSlotFutures.offer(allocationID)).setOfferSlotsFunction((resourceID, slotOffers) -> CompletableFuture.completedFuture(new ArrayList<>(slotOffers))).setRegisterTaskManagerFunction((ignoredJobId, ignoredTaskManagerRegistrationInformation) -> CompletableFuture.completedFuture(new JMTMRegistrationSuccess(jobManagerResourceId))).build();
    final String jobManagerAddress = jobMasterGateway.getAddress();
    rpc.registerGateway(jobManagerAddress, jobMasterGateway);
    jobManagerLeaderRetriever.notifyListener(jobManagerAddress, jobMasterGateway.getFencingToken().toUUID());
    taskExecutor.start();
    try {
        final TaskExecutorGateway taskExecutorGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class);
        initialSlotReporting.await();
        final AllocationID allocationIdInBoth = new AllocationID();
        final AllocationID allocationIdOnlyInJM = new AllocationID();
        final AllocationID allocationIdOnlyInTM = new AllocationID();
        taskExecutorGateway.requestSlot(new SlotID(taskExecutor.getResourceID(), 0), jobId, allocationIdInBoth, ResourceProfile.ZERO, "foobar", testingResourceManagerGateway.getFencingToken(), timeout);
        taskExecutorGateway.requestSlot(new SlotID(taskExecutor.getResourceID(), 1), jobId, allocationIdOnlyInTM, ResourceProfile.ZERO, "foobar", testingResourceManagerGateway.getFencingToken(), timeout);
        activeSlots.await();
        List<AllocatedSlotInfo> allocatedSlotInfos = Arrays.asList(new AllocatedSlotInfo(0, allocationIdInBoth), new AllocatedSlotInfo(1, allocationIdOnlyInJM));
        AllocatedSlotReport allocatedSlotReport = new AllocatedSlotReport(jobId, allocatedSlotInfos);
        taskExecutorGateway.heartbeatFromJobManager(jobManagerResourceId, allocatedSlotReport);
        assertThat(failedSlotFutures.take(), is(allocationIdOnlyInJM));
        assertThat(allocationsNotifiedFree.take(), is(allocationIdOnlyInTM));
        assertThat(failedSlotFutures.poll(5L, TimeUnit.MILLISECONDS), nullValue());
        assertThat(allocationsNotifiedFree.poll(5L, TimeUnit.MILLISECONDS), nullValue());
    } finally {
        RpcUtils.terminateRpcEndpoint(taskExecutor, timeout);
    }
}

Example 25 with TestingJobMasterGateway

use of org.apache.flink.runtime.jobmaster.utils.TestingJobMasterGateway 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);
    }
}