Examples with JobDataReplicator com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator used on opensource projects

Example 1 with JobDataReplicator

use of com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator in project titus-control-plane by Netflix.

the class DefaultNodeConditionControllerTest method checkTasksTerminatedDueToBadNodeConditions.

@Test
public void checkTasksTerminatedDueToBadNodeConditions() {
    // Mock jobs, tasks & nodes
    Map<String, TitusNode> nodeMap = buildNodes();
    List<Job<BatchJobExt>> jobs = getJobs(true);
    Map<String, List<Task>> tasksByJobIdMap = buildTasksForJobAndNodeAssignment(new ArrayList<>(nodeMap.values()), jobs);
    TitusRuntime titusRuntime = mock(TitusRuntime.class);
    when(titusRuntime.getRegistry()).thenReturn(new DefaultRegistry());
    RelocationConfiguration configuration = mock(RelocationConfiguration.class);
    when(configuration.getBadNodeConditionPattern()).thenReturn(".*Failure");
    when(configuration.isTaskTerminationOnBadNodeConditionEnabled()).thenReturn(true);
    NodeDataResolver nodeDataResolver = mock(NodeDataResolver.class);
    when(nodeDataResolver.resolve()).thenReturn(nodeMap);
    JobDataReplicator jobDataReplicator = mock(JobDataReplicator.class);
    when(jobDataReplicator.getStalenessMs()).thenReturn(0L);
    ReadOnlyJobOperations readOnlyJobOperations = mock(ReadOnlyJobOperations.class);
    when(readOnlyJobOperations.getJobs()).thenReturn(new ArrayList<>(jobs));
    tasksByJobIdMap.forEach((key, value) -> when(readOnlyJobOperations.getTasks(key)).thenReturn(value));
    JobManagementClient jobManagementClient = mock(JobManagementClient.class);
    Set<String> terminatedTaskIds = new HashSet<>();
    when(jobManagementClient.killTask(anyString(), anyBoolean(), any())).thenAnswer(invocation -> {
        String taskIdToBeTerminated = invocation.getArgument(0);
        terminatedTaskIds.add(taskIdToBeTerminated);
        return Mono.empty();
    });
    DefaultNodeConditionController nodeConditionCtrl = new DefaultNodeConditionController(configuration, nodeDataResolver, jobDataReplicator, readOnlyJobOperations, jobManagementClient, titusRuntime);
    ExecutionContext executionContext = ExecutionContext.newBuilder().withIteration(ExecutionId.initial()).build();
    StepVerifier.create(nodeConditionCtrl.handleNodesWithBadCondition(executionContext)).verifyComplete();
    assertThat(terminatedTaskIds).isNotEmpty();
    assertThat(terminatedTaskIds.size()).isEqualTo(2);
    verifyTerminatedTasksOnBadNodes(terminatedTaskIds, tasksByJobIdMap, nodeMap);
}

Example 2 with JobDataReplicator

use of com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator in project titus-control-plane by Netflix.

the class DefaultNodeConditionControllerTest method noTerminationsOnDataStaleness.

@Test
public void noTerminationsOnDataStaleness() {
    TitusRuntime titusRuntime = mock(TitusRuntime.class);
    when(titusRuntime.getRegistry()).thenReturn(new DefaultRegistry());
    RelocationConfiguration configuration = mock(RelocationConfiguration.class);
    when(configuration.getBadNodeConditionPattern()).thenReturn(".*Problem");
    when(configuration.isTaskTerminationOnBadNodeConditionEnabled()).thenReturn(true);
    when(configuration.getDataStalenessThresholdMs()).thenReturn(8000L);
    NodeDataResolver nodeDataResolver = mock(NodeDataResolver.class);
    when(nodeDataResolver.getStalenessMs()).thenReturn(5L);
    JobDataReplicator jobDataReplicator = mock(JobDataReplicator.class);
    when(jobDataReplicator.getStalenessMs()).thenReturn(10L);
    ReadOnlyJobOperations readOnlyJobOperations = mock(ReadOnlyJobOperations.class);
    JobManagementClient jobManagementClient = mock(JobManagementClient.class);
    Set<String> terminatedTaskIds = new HashSet<>();
    when(jobManagementClient.killTask(anyString(), anyBoolean(), any())).thenAnswer(invocation -> {
        String taskIdToBeTerminated = invocation.getArgument(0);
        terminatedTaskIds.add(taskIdToBeTerminated);
        return Mono.empty();
    });
    DefaultNodeConditionController nodeConditionCtrl = new DefaultNodeConditionController(configuration, nodeDataResolver, jobDataReplicator, readOnlyJobOperations, jobManagementClient, titusRuntime);
    ExecutionContext executionContext = ExecutionContext.newBuilder().withIteration(ExecutionId.initial()).build();
    StepVerifier.create(nodeConditionCtrl.handleNodesWithBadCondition(executionContext)).verifyComplete();
    // No tasks terminated
    assertThat(terminatedTaskIds).isEmpty();
}

Example 3 with JobDataReplicator

use of com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator in project titus-control-plane by Netflix.

the class StreamDataReplicatorPerf method main.

public static void main(String[] args) throws InterruptedException {
    TitusRuntime titusRuntime = TitusRuntimes.internal();
    JobManagementClient client = Mockito.mock(JobManagementClient.class);
    JobConnectorConfiguration configuration = Mockito.mock(JobConnectorConfiguration.class);
    Mockito.when(client.observeJobs(ArgumentMatchers.any())).thenAnswer(invocation -> Flux.defer(() -> {
        JobManagerEvent jobUpdateEvent = JobUpdateEvent.newJob(JOB, JobManagerConstants.GRPC_REPLICATOR_CALL_METADATA);
        JobManagerEvent taskUpdateEvent = TaskUpdateEvent.newTask(JOB, TASK, JobManagerConstants.GRPC_REPLICATOR_CALL_METADATA);
        return Flux.just(jobUpdateEvent, JobManagerEvent.snapshotMarker()).concatWith(Flux.interval(Duration.ofSeconds(1)).take(1).map(tick -> taskUpdateEvent)).concatWith(Flux.interval(Duration.ofSeconds(1)).take(1).flatMap(tick -> Flux.error(new RuntimeException("Simulated error"))));
    }));
    JobDataReplicator replicator = new JobDataReplicatorProvider(configuration, client, JobSnapshotFactories.newDefault(titusRuntime), titusRuntime).get();
    replicator.events().subscribe(System.out::println);
    Thread.sleep(3600_000);
}

Example 4 with JobDataReplicator

use of com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator in project titus-control-plane by Netflix.

the class DefaultNodeConditionControllerTest method badNodeConditionsIgnoredForJobsNotOptingIn.

@Test
public void badNodeConditionsIgnoredForJobsNotOptingIn() {
    Map<String, TitusNode> nodeMap = buildNodes();
    List<Job<BatchJobExt>> jobs = getJobs(false);
    Map<String, List<Task>> stringListMap = buildTasksForJobAndNodeAssignment(new ArrayList<>(nodeMap.values()), jobs);
    TitusRuntime titusRuntime = mock(TitusRuntime.class);
    when(titusRuntime.getRegistry()).thenReturn(new DefaultRegistry());
    RelocationConfiguration configuration = mock(RelocationConfiguration.class);
    when(configuration.getBadNodeConditionPattern()).thenReturn(".*Failure");
    when(configuration.isTaskTerminationOnBadNodeConditionEnabled()).thenReturn(true);
    NodeDataResolver nodeDataResolver = mock(NodeDataResolver.class);
    when(nodeDataResolver.resolve()).thenReturn(nodeMap);
    JobDataReplicator jobDataReplicator = mock(JobDataReplicator.class);
    when(jobDataReplicator.getStalenessMs()).thenReturn(0L);
    // Job attribute "terminateContainerOnBadAgent" = False
    ReadOnlyJobOperations readOnlyJobOperations = mock(ReadOnlyJobOperations.class);
    when(readOnlyJobOperations.getJobs()).thenReturn(new ArrayList<>(jobs));
    stringListMap.forEach((key, value) -> when(readOnlyJobOperations.getTasks(key)).thenReturn(value));
    JobManagementClient jobManagementClient = mock(JobManagementClient.class);
    Set<String> terminatedTaskIds = new HashSet<>();
    when(jobManagementClient.killTask(anyString(), anyBoolean(), any())).thenAnswer(invocation -> {
        String taskIdToBeTerminated = invocation.getArgument(0);
        terminatedTaskIds.add(taskIdToBeTerminated);
        return Mono.empty();
    });
    DefaultNodeConditionController nodeConditionController = new DefaultNodeConditionController(configuration, nodeDataResolver, jobDataReplicator, readOnlyJobOperations, jobManagementClient, titusRuntime);
    ExecutionContext executionContext = ExecutionContext.newBuilder().withIteration(ExecutionId.initial()).build();
    StepVerifier.create(nodeConditionController.handleNodesWithBadCondition(executionContext)).verifyComplete();
    // no tasks should be terminated for jobs
    assertThat(terminatedTaskIds).isEmpty();
}

Example 5 with JobDataReplicator

use of com.netflix.titus.runtime.connector.jobmanager.JobDataReplicator in project titus-control-plane by Netflix.

the class RelocationConnectorStubs method getApplicationContext.

public ApplicationContext getApplicationContext() {
    StaticApplicationContext context = new StaticApplicationContext();
    context.getBeanFactory().registerSingleton("titusRuntime", titusRuntime);
    context.getBeanFactory().registerSingleton("nodeDataResolver", nodeDataResolver);
    context.getBeanFactory().registerSingleton("readOnlyJobOperations", jobOperations);
    context.getBeanFactory().registerSingleton("readOnlyEvictionOperations", evictionComponentStub.getEvictionOperations());
    context.getBeanFactory().registerSingleton("evictionServiceClient", evictionComponentStub.getEvictionServiceClient());
    context.getBeanFactory().registerSingleton("jobManagementClient", mock(JobManagementClient.class));
    JobDataReplicator jobDataReplicator = mock(JobDataReplicator.class);
    when(jobDataReplicator.getStalenessMs()).thenReturn(0L);
    context.getBeanFactory().registerSingleton("jobDataReplicator", jobDataReplicator);
    EvictionDataReplicator evictionDataReplicator = mock(EvictionDataReplicator.class);
    when(evictionDataReplicator.getStalenessMs()).thenReturn(0L);
    context.getBeanFactory().registerSingleton("evictionDataReplicator", evictionDataReplicator);
    context.refresh();
    return context;
}