Examples with TestingLeaderElectionService org.apache.flink.runtime.leaderelection.TestingLeaderElectionService used on opensource projects

Example 16 with TestingLeaderElectionService

use of org.apache.flink.runtime.leaderelection.TestingLeaderElectionService in project flink by apache.

the class ResourceManagerHATest method testGrantAndRevokeLeadership.

@Test
public void testGrantAndRevokeLeadership() throws Exception {
    final TestingLeaderElectionService leaderElectionService = new TestingLeaderElectionService();
    final TestingResourceManagerService resourceManagerService = TestingResourceManagerService.newBuilder().setRmLeaderElectionService(leaderElectionService).build();
    try {
        resourceManagerService.start();
        final UUID leaderId = UUID.randomUUID();
        resourceManagerService.isLeader(leaderId);
        // after grant leadership, verify resource manager is started with the fencing token
        assertEquals(leaderId, leaderElectionService.getConfirmationFuture().get().getLeaderSessionId());
        assertTrue(resourceManagerService.getResourceManagerFencingToken().isPresent());
        assertEquals(leaderId, resourceManagerService.getResourceManagerFencingToken().get().toUUID());
        // then revoke leadership, verify resource manager is closed
        final Optional<CompletableFuture<Void>> rmTerminationFutureOpt = resourceManagerService.getResourceManagerTerminationFuture();
        assertTrue(rmTerminationFutureOpt.isPresent());
        resourceManagerService.notLeader();
        rmTerminationFutureOpt.get().get();
        resourceManagerService.rethrowFatalErrorIfAny();
    } finally {
        resourceManagerService.cleanUp();
    }
}

Example 17 with TestingLeaderElectionService

use of org.apache.flink.runtime.leaderelection.TestingLeaderElectionService in project flink by apache.

the class ResourceManagerTaskExecutorTest method createAndStartResourceManager.

private void createAndStartResourceManager() throws Exception {
    final TestingLeaderElectionService leaderElectionService = new TestingLeaderElectionService();
    rmService = TestingResourceManagerService.newBuilder().setRpcService(rpcService).setRmLeaderElectionService(leaderElectionService).build();
    rmService.start();
    rmService.isLeader(UUID.randomUUID());
    leaderElectionService.getConfirmationFuture().thenRun(() -> {
        rmGateway = rmService.getResourceManagerGateway().orElseThrow(() -> new AssertionError("RM not available after confirming leadership."));
    }).get(TIMEOUT.getSize(), TIMEOUT.getUnit());
}

Example 18 with TestingLeaderElectionService

use of org.apache.flink.runtime.leaderelection.TestingLeaderElectionService 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();
}

Example 19 with TestingLeaderElectionService

use of org.apache.flink.runtime.leaderelection.TestingLeaderElectionService in project flink by apache.

the class DispatcherCleanupITCase method testCleanupThroughRetries.

@Test
public void testCleanupThroughRetries() throws Exception {
    final JobGraph jobGraph = createJobGraph();
    final JobID jobId = jobGraph.getJobID();
    // JobGraphStore
    final AtomicInteger actualGlobalCleanupCallCount = new AtomicInteger();
    final OneShotLatch successfulCleanupLatch = new OneShotLatch();
    final int numberOfErrors = 5;
    final RuntimeException temporaryError = new RuntimeException("Expected RuntimeException: Unable to remove job graph.");
    final JobGraphStore jobGraphStore = createAndStartJobGraphStoreWithCleanupFailures(numberOfErrors, temporaryError, actualGlobalCleanupCallCount, successfulCleanupLatch);
    haServices.setJobGraphStore(jobGraphStore);
    // Construct leader election service.
    final TestingLeaderElectionService leaderElectionService = new TestingLeaderElectionService();
    haServices.setJobMasterLeaderElectionService(jobId, leaderElectionService);
    // start the dispatcher with enough retries on cleanup
    final JobManagerRunnerRegistry jobManagerRunnerRegistry = new DefaultJobManagerRunnerRegistry(2);
    final Dispatcher dispatcher = createTestingDispatcherBuilder().setResourceCleanerFactory(new DispatcherResourceCleanerFactory(ForkJoinPool.commonPool(), TestingRetryStrategies.createWithNumberOfRetries(numberOfErrors), jobManagerRunnerRegistry, haServices.getJobGraphStore(), blobServer, haServices, UnregisteredMetricGroups.createUnregisteredJobManagerMetricGroup())).build();
    dispatcher.start();
    toTerminate.add(dispatcher);
    leaderElectionService.isLeader(UUID.randomUUID());
    final DispatcherGateway dispatcherGateway = dispatcher.getSelfGateway(DispatcherGateway.class);
    dispatcherGateway.submitJob(jobGraph, TIMEOUT).get();
    waitForJobToFinish(leaderElectionService, dispatcherGateway, jobId);
    successfulCleanupLatch.await();
    assertThat(actualGlobalCleanupCallCount.get(), equalTo(numberOfErrors + 1));
    assertThat("The JobGraph should be removed from JobGraphStore.", haServices.getJobGraphStore().getJobIds(), IsEmptyCollection.empty());
    CommonTestUtils.waitUntilCondition(() -> haServices.getJobResultStore().hasJobResultEntry(jobId), Deadline.fromNow(Duration.ofMinutes(5)), "The JobResultStore should have this job marked as clean.");
}

Example 20 with TestingLeaderElectionService

use of org.apache.flink.runtime.leaderelection.TestingLeaderElectionService in project flink by apache.

the class DispatcherCleanupITCase method testCleanupAfterLeadershipChange.

@Test
public void testCleanupAfterLeadershipChange() throws Exception {
    final JobGraph jobGraph = createJobGraph();
    final JobID jobId = jobGraph.getJobID();
    // Construct job graph store.
    final AtomicInteger actualGlobalCleanupCallCount = new AtomicInteger();
    final OneShotLatch successfulCleanupLatch = new OneShotLatch();
    final RuntimeException temporaryError = new RuntimeException("Unable to remove job graph.");
    final JobGraphStore jobGraphStore = createAndStartJobGraphStoreWithCleanupFailures(1, temporaryError, actualGlobalCleanupCallCount, successfulCleanupLatch);
    haServices.setJobGraphStore(jobGraphStore);
    // Construct leader election service.
    final TestingLeaderElectionService leaderElectionService = new TestingLeaderElectionService();
    haServices.setJobMasterLeaderElectionService(jobId, leaderElectionService);
    // start the dispatcher with no retries on cleanup
    final CountDownLatch jobGraphRemovalErrorReceived = new CountDownLatch(1);
    final Dispatcher dispatcher = createTestingDispatcherBuilder().setFatalErrorHandler(throwable -> {
        final Optional<Throwable> maybeError = ExceptionUtils.findThrowable(throwable, temporaryError::equals);
        if (maybeError.isPresent()) {
            jobGraphRemovalErrorReceived.countDown();
        } else {
            testingFatalErrorHandlerResource.getFatalErrorHandler().onFatalError(throwable);
        }
    }).build();
    dispatcher.start();
    toTerminate.add(dispatcher);
    leaderElectionService.isLeader(UUID.randomUUID());
    final DispatcherGateway dispatcherGateway = dispatcher.getSelfGateway(DispatcherGateway.class);
    dispatcherGateway.submitJob(jobGraph, TIMEOUT).get();
    waitForJobToFinish(leaderElectionService, dispatcherGateway, jobId);
    jobGraphRemovalErrorReceived.await();
    // Remove job master leadership.
    leaderElectionService.notLeader();
    // This will clear internal state of election service, so a new contender can register.
    leaderElectionService.stop();
    assertThat(successfulCleanupLatch.isTriggered(), CoreMatchers.is(false));
    assertThat("The JobGraph is still stored in the JobGraphStore.", haServices.getJobGraphStore().getJobIds(), CoreMatchers.is(Collections.singleton(jobId)));
    assertThat("The JobResultStore has this job marked as dirty.", haServices.getJobResultStore().getDirtyResults().stream().map(JobResult::getJobId).collect(Collectors.toSet()), CoreMatchers.is(Collections.singleton(jobId)));
    // Run a second dispatcher, that restores our finished job.
    final Dispatcher secondDispatcher = createTestingDispatcherBuilder().setRecoveredDirtyJobs(haServices.getJobResultStore().getDirtyResults()).build();
    secondDispatcher.start();
    toTerminate.add(secondDispatcher);
    leaderElectionService.isLeader(UUID.randomUUID());
    CommonTestUtils.waitUntilCondition(() -> haServices.getJobResultStore().getDirtyResults().isEmpty(), Deadline.fromNow(TimeUtils.toDuration(TIMEOUT)));
    assertThat("The JobGraph is not stored in the JobGraphStore.", haServices.getJobGraphStore().getJobIds(), IsEmptyCollection.empty());
    assertTrue("The JobResultStore has the job listed as clean.", haServices.getJobResultStore().hasJobResultEntry(jobId));
    // wait for the successful cleanup to be triggered
    successfulCleanupLatch.await();
    assertThat(actualGlobalCleanupCallCount.get(), equalTo(2));
}