use of org.apache.flink.runtime.heartbeat.TestingHeartbeatServices in project flink by apache.
the class JobMasterTest method testJobFailureWhenTaskExecutorHeartbeatTimeout.
@Test
public void testJobFailureWhenTaskExecutorHeartbeatTimeout() throws Exception {
final TestingHeartbeatServices testingHeartbeatService = new TestingHeartbeatServices(heartbeatInterval, heartbeatTimeout);
runJobFailureWhenTaskExecutorTerminatesTest(testingHeartbeatService, (localTaskManagerLocation, jobMasterGateway) -> testingHeartbeatService.triggerHeartbeatTimeout(jmResourceId, localTaskManagerLocation.getResourceID()), (jobMasterGateway, taskManagerResourceId) -> (resourceId, ignored) -> {
jobMasterGateway.heartbeatFromTaskManager(taskManagerResourceId, TaskExecutorToJobManagerHeartbeatPayload.empty());
return FutureUtils.completedVoidFuture();
});
}
use of org.apache.flink.runtime.heartbeat.TestingHeartbeatServices in project flink by apache.
the class TaskExecutorTest method testHeartbeatTimeoutWithJobManager.
@Test
public void testHeartbeatTimeoutWithJobManager() throws Exception {
final TestingHeartbeatServices heartbeatServices = new TestingHeartbeatServices();
final ResourceID jmResourceId = ResourceID.generate();
runJobManagerHeartbeatTest(jmResourceId, heartbeatServices, ignored -> {
}, (taskExecutorResourceId, ignoredTaskExecutorGateway, ignoredAllocationId) -> heartbeatServices.triggerHeartbeatTimeout(taskExecutorResourceId, jmResourceId));
}
use of org.apache.flink.runtime.heartbeat.TestingHeartbeatServices in project flink by apache.
the class ZooKeeperDefaultDispatcherRunnerTest method testResourceCleanupUnderLeadershipChange.
/**
* See FLINK-11665.
*/
@Test
public void testResourceCleanupUnderLeadershipChange() throws Exception {
final TestingRpcService rpcService = testingRpcServiceResource.getTestingRpcService();
final TestingLeaderElectionService dispatcherLeaderElectionService = new TestingLeaderElectionService();
final CuratorFramework client = ZooKeeperUtils.startCuratorFramework(configuration, fatalErrorHandler).asCuratorFramework();
try (final TestingHighAvailabilityServices highAvailabilityServices = new TestingHighAvailabilityServicesBuilder().setDispatcherLeaderElectionService(dispatcherLeaderElectionService).setJobMasterLeaderRetrieverFunction(jobId -> ZooKeeperUtils.createLeaderRetrievalService(client)).build()) {
final PartialDispatcherServices partialDispatcherServices = new PartialDispatcherServices(configuration, highAvailabilityServices, CompletableFuture::new, blobServer, new TestingHeartbeatServices(), UnregisteredMetricGroups::createUnregisteredJobManagerMetricGroup, new MemoryExecutionGraphInfoStore(), fatalErrorHandler, VoidHistoryServerArchivist.INSTANCE, null, ForkJoinPool.commonPool(), new DispatcherOperationCaches());
final DefaultDispatcherRunnerFactory defaultDispatcherRunnerFactory = DefaultDispatcherRunnerFactory.createSessionRunner(SessionDispatcherFactory.INSTANCE);
try (final DispatcherRunner dispatcherRunner = createDispatcherRunner(rpcService, dispatcherLeaderElectionService, new JobPersistenceComponentFactory() {
@Override
public JobGraphStore createJobGraphStore() {
return createZooKeeperJobGraphStore(client);
}
@Override
public JobResultStore createJobResultStore() {
return new EmbeddedJobResultStore();
}
}, partialDispatcherServices, defaultDispatcherRunnerFactory)) {
// initial run
DispatcherGateway dispatcherGateway = grantLeadership(dispatcherLeaderElectionService);
final JobGraph jobGraph = createJobGraphWithBlobs();
LOG.info("Initial job submission {}.", jobGraph.getJobID());
dispatcherGateway.submitJob(jobGraph, TESTING_TIMEOUT).get();
dispatcherLeaderElectionService.notLeader();
// recovering submitted jobs
LOG.info("Re-grant leadership first time.");
dispatcherGateway = grantLeadership(dispatcherLeaderElectionService);
LOG.info("Cancel recovered job {}.", jobGraph.getJobID());
// cancellation of the job should remove everything
final CompletableFuture<JobResult> jobResultFuture = dispatcherGateway.requestJobResult(jobGraph.getJobID(), TESTING_TIMEOUT);
dispatcherGateway.cancelJob(jobGraph.getJobID(), TESTING_TIMEOUT).get();
// a successful cancellation should eventually remove all job information
final JobResult jobResult = jobResultFuture.get();
assertThat(jobResult.getApplicationStatus(), is(ApplicationStatus.CANCELED));
dispatcherLeaderElectionService.notLeader();
// check that the job has been removed from ZooKeeper
final JobGraphStore submittedJobGraphStore = createZooKeeperJobGraphStore(client);
CommonTestUtils.waitUntilCondition(() -> submittedJobGraphStore.getJobIds().isEmpty(), Deadline.fromNow(VERIFICATION_TIMEOUT), 20L);
}
}
// check resource clean up
assertThat(clusterHaStorageDir.listFiles(), is(emptyArray()));
}
Aggregations