Examples with JobStore com.netflix.titus.api.jobmanager.store.JobStore used on opensource projects

Example 16 with JobStore

use of com.netflix.titus.api.jobmanager.store.JobStore in project titus-control-plane by Netflix.

the class KillInitiatedActions method userInitiateTaskKillAction.

/**
 * Change a task to {@link TaskState#KillInitiated} state, store it, and send the kill command to the compute provider.
 * All models are updated when both operations complete.
 * This method is used for user initiated kill operations, so the store operation happens before response is sent back to the user.
 */
public static ChangeAction userInitiateTaskKillAction(ReconciliationEngine<JobManagerReconcilerEvent> engine, JobServiceRuntime executionContext, JobStore jobStore, VersionSupplier versionSupplier, String taskId, boolean shrink, boolean preventMinSizeUpdate, String reasonCode, String reason, TitusRuntime titusRuntime, CallMetadata callMetadata) {
    return TitusChangeAction.newAction("userInitiateTaskKill").id(taskId).trigger(V3JobOperations.Trigger.API).summary(reason).callMetadata(callMetadata).changeWithModelUpdates(self -> JobEntityHolders.toTaskObservable(engine, taskId, titusRuntime).flatMap(task -> {
        TaskState taskState = task.getStatus().getState();
        if (taskState == TaskState.KillInitiated || taskState == TaskState.Finished) {
            return Observable.just(Collections.<ModelActionHolder>emptyList());
        }
        if (shrink) {
            Job<ServiceJobExt> job = engine.getReferenceView().getEntity();
            Capacity capacity = job.getJobDescriptor().getExtensions().getCapacity();
            if (preventMinSizeUpdate && capacity.getDesired() <= capacity.getMin()) {
                return Observable.<List<ModelActionHolder>>error(JobManagerException.terminateAndShrinkNotAllowed(job, task));
            }
        }
        Task taskWithKillInitiated = VersionSuppliers.nextVersion(JobFunctions.changeTaskStatus(task, TaskState.KillInitiated, reasonCode, reason, titusRuntime.getClock()), versionSupplier);
        Callable<List<ModelActionHolder>> modelUpdateActions = () -> JobEntityHolders.expectTask(engine, task.getId(), titusRuntime).map(current -> {
            List<ModelActionHolder> updateActions = new ArrayList<>();
            TitusModelAction stateUpdateAction = TitusModelAction.newModelUpdate(self).taskUpdate(taskWithKillInitiated);
            updateActions.addAll(ModelActionHolder.allModels(stateUpdateAction));
            if (shrink) {
                TitusModelAction shrinkAction = createShrinkAction(self, versionSupplier);
                updateActions.add(ModelActionHolder.reference(shrinkAction));
            }
            return updateActions;
        }).orElse(Collections.emptyList());
        return jobStore.updateTask(taskWithKillInitiated).andThen(createKillAction(executionContext, task)).andThen(Observable.fromCallable(modelUpdateActions));
    }));
}

Example 17 with JobStore

use of com.netflix.titus.api.jobmanager.store.JobStore in project titus-control-plane by Netflix.

the class DifferenceResolverUtils method findTaskStateTimeouts.

/**
 * Find all tasks that are stuck in a specific state. The number of {@link ChangeAction changes} will be limited
 * by the {@link TokenBucket stuckInStateRateLimiter}
 */
public static List<ChangeAction> findTaskStateTimeouts(ReconciliationEngine<JobManagerReconcilerEvent> engine, JobView runningJobView, JobManagerConfiguration configuration, JobServiceRuntime runtime, JobStore jobStore, VersionSupplier versionSupplier, TokenBucket stuckInStateRateLimiter, TitusRuntime titusRuntime) {
    Clock clock = titusRuntime.getClock();
    List<ChangeAction> actions = new ArrayList<>();
    runningJobView.getJobHolder().getChildren().forEach(taskHolder -> {
        Task task = taskHolder.getEntity();
        TaskState taskState = task.getStatus().getState();
        if (JobFunctions.isBatchJob(runningJobView.getJob()) && taskState == TaskState.Started) {
            Job<BatchJobExt> batchJob = runningJobView.getJob();
            // We expect runtime limit to be always set, so this is just extra safety measure.
            long runtimeLimitMs = Math.max(BatchJobExt.RUNTIME_LIMIT_MIN, batchJob.getJobDescriptor().getExtensions().getRuntimeLimitMs());
            long deadline = task.getStatus().getTimestamp() + runtimeLimitMs;
            if (deadline < clock.wallTime()) {
                actions.add(KillInitiatedActions.reconcilerInitiatedTaskKillInitiated(engine, task, runtime, jobStore, versionSupplier, TaskStatus.REASON_RUNTIME_LIMIT_EXCEEDED, "Task running too long (runtimeLimit=" + runtimeLimitMs + "ms)", titusRuntime));
            }
            return;
        }
        TaskTimeoutChangeActions.TimeoutStatus timeoutStatus = TaskTimeoutChangeActions.getTimeoutStatus(taskHolder, clock);
        switch(timeoutStatus) {
            case Ignore:
            case Pending:
                break;
            case NotSet:
                long timeoutMs = -1;
                switch(taskState) {
                    case Launched:
                        timeoutMs = configuration.getTaskInLaunchedStateTimeoutMs();
                        break;
                    case StartInitiated:
                        timeoutMs = isBatch(runningJobView.getJob()) ? configuration.getBatchTaskInStartInitiatedStateTimeoutMs() : configuration.getServiceTaskInStartInitiatedStateTimeoutMs();
                        break;
                    case KillInitiated:
                        timeoutMs = configuration.getTaskInKillInitiatedStateTimeoutMs();
                        break;
                }
                if (timeoutMs > 0) {
                    actions.add(TaskTimeoutChangeActions.setTimeout(taskHolder.getId(), task.getStatus().getState(), timeoutMs, clock));
                }
                break;
            case TimedOut:
                if (!stuckInStateRateLimiter.tryTake()) {
                    break;
                }
                if (task.getStatus().getState() == TaskState.KillInitiated) {
                    int attempts = TaskTimeoutChangeActions.getKillInitiatedAttempts(taskHolder) + 1;
                    if (attempts >= configuration.getTaskKillAttempts()) {
                        actions.add(BasicTaskActions.updateTaskInRunningModel(task.getId(), V3JobOperations.Trigger.Reconciler, configuration, engine, taskParam -> Optional.of(taskParam.toBuilder().withStatus(taskParam.getStatus().toBuilder().withState(TaskState.Finished).withReasonCode(TaskStatus.REASON_STUCK_IN_KILLING_STATE).withReasonMessage("stuck in " + taskState + "state").build()).build()), "TimedOut in KillInitiated state", versionSupplier, titusRuntime, JobManagerConstants.RECONCILER_CALLMETADATA.toBuilder().withCallReason("Kill initiated").build()));
                    } else {
                        actions.add(TaskTimeoutChangeActions.incrementTaskKillAttempt(task.getId(), configuration.getTaskInKillInitiatedStateTimeoutMs(), clock));
                        actions.add(KillInitiatedActions.reconcilerInitiatedTaskKillInitiated(engine, task, runtime, jobStore, versionSupplier, TaskStatus.REASON_STUCK_IN_KILLING_STATE, "Another kill attempt (" + (attempts + 1) + ')', titusRuntime));
                    }
                } else {
                    actions.add(KillInitiatedActions.reconcilerInitiatedTaskKillInitiated(engine, task, runtime, jobStore, versionSupplier, TaskStatus.REASON_STUCK_IN_STATE, "Task stuck in " + taskState + " state", titusRuntime));
                }
                break;
        }
    });
    return actions;
}

Example 18 with JobStore

use of com.netflix.titus.api.jobmanager.store.JobStore in project titus-control-plane by Netflix.

the class EmbeddedTitusMaster method boot.

public EmbeddedTitusMaster boot() {
    Stopwatch timer = Stopwatch.createStarted();
    logger.info("Starting Titus Master");
    Module embeddedKubeModule;
    if (embeddedKubeCluster == null) {
        embeddedKubeModule = new AbstractModule() {

            @Override
            protected void configure() {
            }
        };
    } else {
        embeddedKubeModule = new EmbeddedKubeModule(embeddedKubeCluster);
    }
    injector = InjectorBuilder.fromModules(Modules.override(new TitusRuntimeModule(false)).with(new AbstractModule() {

        @Override
        protected void configure() {
            bind(Archaius2ConfigurationLogger.class).asEagerSingleton();
            bind(Registry.class).toInstance(new DefaultRegistry());
        }
    }), embeddedKubeModule, Modules.override(new TitusMasterModule(enableREST, TitusMasterModule.Mode.EMBEDDED_KUBE)).with(new AbstractModule() {

        @Override
        protected void configure() {
            bind(InstanceCloudConnector.class).toInstance(new NoOpInstanceCloudConnector());
            bind(MasterDescription.class).toInstance(masterDescription);
            bind(MasterMonitor.class).to(LocalMasterMonitor.class);
            bind(AppScalePolicyStore.class).to(InMemoryPolicyStore.class);
            bind(LoadBalancerStore.class).to(InMemoryLoadBalancerStore.class);
            bind(LoadBalancerConnector.class).to(NoOpLoadBalancerConnector.class);
            bind(LoadBalancerJobValidator.class).to(NoOpLoadBalancerJobValidator.class);
        }

        @Provides
        @Singleton
        public JobStore getJobStore(TitusRuntime titusRuntime) {
            if (!cassandraJobStore) {
                return jobStore;
            }
            try {
                JobStore jobStore = EmbeddedCassandraStoreFactory.newBuilder().withTitusRuntime(titusRuntime).build().getJobStore();
                return jobStore;
            } catch (Throwable e) {
                e.printStackTrace();
                return null;
            }
        }
    }), newJettyModule(), new ArchaiusModule() {

        @Override
        protected void configureArchaius() {
            bindApplicationConfigurationOverride().toInstance(config);
        }
    }).createInjector();
    if (grpcPort <= 0) {
        grpcPort = getGrpcPort();
        config.setProperty("titus.master.grpcServer.port", "" + grpcPort);
    }
    injector.getInstance(ContainerEventBus.class).submitInOrder(new ContainerEventBus.ContainerStartedEvent());
    injector.getInstance(LeaderActivator.class).becomeLeader();
    injector.getInstance(AuditLogService.class).auditLogEvents().subscribe(auditLogs::add);
    if (enableREST) {
        // Since jetty API server is run on a separate thread, it may not be ready yet
        // We do not have better way, but call it until it replies.
        getClient().findAllApplicationSLA().retryWhen(attempts -> {
            return attempts.zipWith(Observable.range(1, 5), (n, i) -> i).flatMap(i -> {
                return Observable.timer(i, TimeUnit.SECONDS);
            });
        }).timeout(30, TimeUnit.SECONDS).toBlocking().firstOrDefault(null);
    }
    logger.info("Embedded TitusMaster started in {}ms", timer.elapsed(TimeUnit.MILLISECONDS));
    return this;
}

Example 19 with JobStore

use of com.netflix.titus.api.jobmanager.store.JobStore in project titus-control-plane by Netflix.

the class CassandraJobStoreTest method testUpdateTask.

@Test
public void testUpdateTask() {
    JobStore store = getJobStore();
    Job<BatchJobExt> job = createBatchJobObject();
    store.init().await();
    store.storeJob(job).await();
    Pair<List<Job<?>>, Integer> jobsAndErrors = store.retrieveJobs().toBlocking().first();
    checkRetrievedJob(job, jobsAndErrors.getLeft().get(0));
    Task task = createTaskObject(job);
    store.storeTask(task).await();
    Task retrievedTask = store.retrieveTask(task.getId()).toBlocking().first();
    checkRetrievedTask(task, retrievedTask);
    BatchJobTask newTask = BatchJobTask.newBuilder((BatchJobTask) task).withStatus(TaskStatus.newBuilder().withState(TaskState.Finished).build()).build();
    store.updateTask(newTask).await();
    Task newRetrievedTask = store.retrieveTask(newTask.getId()).toBlocking().first();
    checkRetrievedTask(newTask, newRetrievedTask);
}

Example 20 with JobStore

use of com.netflix.titus.api.jobmanager.store.JobStore in project titus-control-plane by Netflix.

the class CassandraJobStoreTest method testUpdateJob.

@Test
public void testUpdateJob() {
    JobStore store = getJobStore();
    Job<BatchJobExt> job = createBatchJobObject();
    store.init().await();
    store.storeJob(job).await();
    Pair<List<Job<?>>, Integer> jobsAndErrors = store.retrieveJobs().toBlocking().first();
    checkRetrievedJob(job, jobsAndErrors.getLeft().get(0));
    Job<BatchJobExt> newJob = job.toBuilder().withStatus(JobStatus.newBuilder().withState(JobState.Finished).build()).build();
    store.updateJob(newJob).await();
    Pair<List<Job<?>>, Integer> newJobsAndErrors = store.retrieveJobs().toBlocking().first();
    checkRetrievedJob(newJob, newJobsAndErrors.getLeft().get(0));
}