Examples with HealthServiceManager io.pravega.shared.health.HealthServiceManager used on opensource projects

Example 1 with HealthServiceManager

use of io.pravega.shared.health.HealthServiceManager in project pravega by pravega.

the class ControllerServiceStarter method startUp.

@Override
protected void startUp() {
    long traceId = LoggerHelpers.traceEnterWithContext(log, this.objectId, "startUp");
    log.info("Initiating controller service startUp");
    log.info("Controller serviceConfig = {}", serviceConfig.toString());
    log.info("Event processors enabled = {}", serviceConfig.getEventProcessorConfig().isPresent());
    log.info("Cluster listener enabled = {}", serviceConfig.isControllerClusterListenerEnabled());
    log.info("    Host monitor enabled = {}", serviceConfig.getHostMonitorConfig().isHostMonitorEnabled());
    log.info("     gRPC server enabled = {}", serviceConfig.getGRPCServerConfig().isPresent());
    log.info("     REST server enabled = {}", serviceConfig.getRestServerConfig().isPresent());
    final BucketStore bucketStore;
    final TaskMetadataStore taskMetadataStore;
    final HostControllerStore hostStore;
    final CheckpointStore checkpointStore;
    try {
        // Initialize the executor service.
        controllerExecutor = ExecutorServiceHelpers.newScheduledThreadPool(serviceConfig.getThreadPoolSize(), "controllerpool");
        eventExecutor = ExecutorServiceHelpers.newScheduledThreadPool(serviceConfig.getThreadPoolSize(), "eventprocessor");
        retentionExecutor = ExecutorServiceHelpers.newScheduledThreadPool(Config.RETENTION_THREAD_POOL_SIZE, "retentionpool");
        watermarkingExecutor = ExecutorServiceHelpers.newScheduledThreadPool(Config.WATERMARKING_THREAD_POOL_SIZE, "watermarkingpool");
        bucketStore = StreamStoreFactory.createBucketStore(storeClient, controllerExecutor);
        log.info("Created the bucket store.");
        taskMetadataStore = TaskStoreFactory.createStore(storeClient, controllerExecutor);
        log.info("Created the task store.");
        hostStore = HostStoreFactory.createStore(serviceConfig.getHostMonitorConfig(), storeClient);
        log.info("Created the host store.");
        checkpointStore = CheckpointStoreFactory.create(storeClient);
        log.info("Created the checkpoint store.");
        // Initialize Stream and Transaction metrics.
        StreamMetrics.initialize();
        TransactionMetrics.initialize();
        // On each controller process restart, we use a fresh hostId,
        // which is a combination of hostname and random GUID.
        String hostName = getHostName();
        Host host = new Host(hostName, getPort(), UUID.randomUUID().toString());
        // Create a RequestTracker instance to trace client requests end-to-end.
        GRPCServerConfig grpcServerConfig = serviceConfig.getGRPCServerConfig().get();
        RequestTracker requestTracker = new RequestTracker(grpcServerConfig.isRequestTracingEnabled());
        // Create a Health Service Manager instance.
        healthServiceManager = new HealthServiceManager(serviceConfig.getHealthCheckFrequency());
        if (serviceConfig.getHostMonitorConfig().isHostMonitorEnabled()) {
            // Start the Segment Container Monitor.
            monitor = new SegmentContainerMonitor(hostStore, (CuratorFramework) storeClient.getClient(), new UniformContainerBalancer(), serviceConfig.getHostMonitorConfig().getHostMonitorMinRebalanceInterval());
            monitor.startAsync();
            log.info("Started Segment Container Monitor service.");
            SegmentContainerMonitorHealthContributor segmentContainerMonitorHC = new SegmentContainerMonitorHealthContributor("segmentContainerMonitor", monitor);
            healthServiceManager.register(segmentContainerMonitorHC);
        }
        // This client config is used by the segment store helper (SegmentHelper) to connect to the segment store.
        ClientConfig.ClientConfigBuilder clientConfigBuilder = ClientConfig.builder().controllerURI(URI.create((grpcServerConfig.isTlsEnabled() ? "tls://" : "tcp://") + "localhost:" + grpcServerConfig.getPort())).trustStore(grpcServerConfig.getTlsTrustStore()).validateHostName(false);
        Optional<Boolean> tlsEnabledForSegmentStore = BooleanUtils.extract(serviceConfig.getTlsEnabledForSegmentStore());
        if (tlsEnabledForSegmentStore.isPresent()) {
            clientConfigBuilder.enableTlsToSegmentStore(tlsEnabledForSegmentStore.get());
        }
        // Use one connection per Segment Store to save up resources.
        ClientConfig clientConfig = clientConfigBuilder.maxConnectionsPerSegmentStore(1).build();
        connectionFactory = connectionFactoryRef.orElseGet(() -> new SocketConnectionFactoryImpl(clientConfig));
        connectionPool = new ConnectionPoolImpl(clientConfig, connectionFactory);
        segmentHelper = segmentHelperRef.orElseGet(() -> new SegmentHelper(connectionPool, hostStore, controllerExecutor));
        GrpcAuthHelper authHelper = new GrpcAuthHelper(serviceConfig.getGRPCServerConfig().get().isAuthorizationEnabled(), grpcServerConfig.getTokenSigningKey(), grpcServerConfig.getAccessTokenTTLInSeconds());
        streamStore = streamMetadataStoreRef.orElseGet(() -> StreamStoreFactory.createStore(storeClient, segmentHelper, authHelper, controllerExecutor));
        log.info("Created the stream store.");
        streamMetadataTasks = new StreamMetadataTasks(streamStore, bucketStore, taskMetadataStore, segmentHelper, controllerExecutor, eventExecutor, host.getHostId(), authHelper, serviceConfig.getRetentionFrequency().toMillis());
        streamTransactionMetadataTasks = new StreamTransactionMetadataTasks(streamStore, segmentHelper, controllerExecutor, eventExecutor, host.getHostId(), serviceConfig.getTimeoutServiceConfig(), authHelper);
        BucketServiceFactory bucketServiceFactory = new BucketServiceFactory(host.getHostId(), bucketStore, 1000);
        Duration executionDurationRetention = serviceConfig.getRetentionFrequency();
        PeriodicRetention retentionWork = new PeriodicRetention(streamStore, streamMetadataTasks, retentionExecutor, requestTracker);
        retentionService = bucketServiceFactory.createRetentionService(executionDurationRetention, retentionWork::retention, retentionExecutor);
        retentionService.startAsync();
        retentionService.awaitRunning();
        log.info("Started background periodic service for Retention.");
        RetentionServiceHealthContributor retentionServiceHC = new RetentionServiceHealthContributor("retentionService", retentionService);
        healthServiceManager.register(retentionServiceHC);
        Duration executionDurationWatermarking = Duration.ofSeconds(Config.MINIMUM_WATERMARKING_FREQUENCY_IN_SECONDS);
        watermarkingWork = new PeriodicWatermarking(streamStore, bucketStore, clientConfig, watermarkingExecutor, requestTracker);
        watermarkingService = bucketServiceFactory.createWatermarkingService(executionDurationWatermarking, watermarkingWork::watermark, watermarkingExecutor);
        watermarkingService.startAsync();
        watermarkingService.awaitRunning();
        log.info("Started background periodic service for Watermarking.");
        WatermarkingServiceHealthContributor watermarkingServiceHC = new WatermarkingServiceHealthContributor("watermarkingService", watermarkingService);
        healthServiceManager.register(watermarkingServiceHC);
        // Controller has a mechanism to track the currently active controller host instances. On detecting a failure of
        // any controller instance, the failure detector stores the failed HostId in a failed hosts directory (FH), and
        // invokes the taskSweeper.sweepOrphanedTasks for each failed host. When all resources under the failed hostId
        // are processed and deleted, that failed HostId is removed from FH folder.
        // Moreover, on controller process startup, it detects any hostIds not in the currently active set of
        // controllers and starts sweeping tasks orphaned by those hostIds.
        TaskSweeper taskSweeper = new TaskSweeper(taskMetadataStore, host.getHostId(), controllerExecutor, streamMetadataTasks);
        TxnSweeper txnSweeper = new TxnSweeper(streamStore, streamTransactionMetadataTasks, serviceConfig.getTimeoutServiceConfig().getMaxLeaseValue(), controllerExecutor);
        RequestSweeper requestSweeper = new RequestSweeper(streamStore, controllerExecutor, streamMetadataTasks);
        if (serviceConfig.isControllerClusterListenerEnabled()) {
            cluster = new ClusterZKImpl((CuratorFramework) storeClient.getClient(), ClusterType.CONTROLLER);
        }
        kvtMetadataStore = kvtMetaStoreRef.orElseGet(() -> KVTableStoreFactory.createStore(storeClient, segmentHelper, authHelper, controllerExecutor, streamStore));
        kvtMetadataTasks = new TableMetadataTasks(kvtMetadataStore, segmentHelper, controllerExecutor, eventExecutor, host.getHostId(), authHelper);
        controllerService = new ControllerService(kvtMetadataStore, kvtMetadataTasks, streamStore, bucketStore, streamMetadataTasks, streamTransactionMetadataTasks, segmentHelper, controllerExecutor, cluster, requestTracker);
        // Setup event processors.
        setController(new LocalController(controllerService, grpcServerConfig.isAuthorizationEnabled(), grpcServerConfig.getTokenSigningKey()));
        CompletableFuture<Void> eventProcessorFuture = CompletableFuture.completedFuture(null);
        if (serviceConfig.getEventProcessorConfig().isPresent()) {
            // Create ControllerEventProcessor object.
            controllerEventProcessors = new ControllerEventProcessors(host.getHostId(), serviceConfig.getEventProcessorConfig().get(), localController, checkpointStore, streamStore, bucketStore, connectionPool, streamMetadataTasks, streamTransactionMetadataTasks, kvtMetadataStore, kvtMetadataTasks, eventExecutor);
            // Bootstrap and start it asynchronously.
            eventProcessorFuture = controllerEventProcessors.bootstrap(streamTransactionMetadataTasks, streamMetadataTasks, kvtMetadataTasks).thenAcceptAsync(x -> controllerEventProcessors.startAsync(), eventExecutor);
            EventProcessorHealthContributor eventProcessorHC = new EventProcessorHealthContributor("eventProcessor", controllerEventProcessors);
            healthServiceManager.register(eventProcessorHC);
        }
        // Setup and start controller cluster listener after all sweepers have been initialized.
        if (serviceConfig.isControllerClusterListenerEnabled()) {
            List<FailoverSweeper> failoverSweepers = new ArrayList<>();
            failoverSweepers.add(taskSweeper);
            failoverSweepers.add(txnSweeper);
            failoverSweepers.add(requestSweeper);
            if (serviceConfig.getEventProcessorConfig().isPresent()) {
                assert controllerEventProcessors != null;
                failoverSweepers.add(controllerEventProcessors);
            }
            controllerClusterListener = new ControllerClusterListener(host, cluster, controllerExecutor, failoverSweepers);
            controllerClusterListener.startAsync();
            ClusterListenerHealthContributor clusterListenerHC = new ClusterListenerHealthContributor("clusterListener", controllerClusterListener);
            healthServiceManager.register(clusterListenerHC);
        }
        // Start the Health Service.
        healthServiceManager.start();
        // Start RPC server.
        if (serviceConfig.getGRPCServerConfig().isPresent()) {
            grpcServer = new GRPCServer(controllerService, grpcServerConfig, requestTracker);
            grpcServer.startAsync();
            grpcServer.awaitRunning();
            GRPCServerHealthContributor grpcServerHC = new GRPCServerHealthContributor("GRPCServer", grpcServer);
            healthServiceManager.register(grpcServerHC);
        }
        // Start REST server.
        if (serviceConfig.getRestServerConfig().isPresent()) {
            List<Object> resources = new ArrayList<>();
            resources.add(new StreamMetadataResourceImpl(this.localController, controllerService, grpcServer.getAuthHandlerManager(), connectionFactory, clientConfig));
            resources.add(new HealthImpl(grpcServer.getAuthHandlerManager(), healthServiceManager.getEndpoint()));
            resources.add(new PingImpl());
            MetricsProvider.getMetricsProvider().prometheusResource().ifPresent(resources::add);
            restServer = new RESTServer(serviceConfig.getRestServerConfig().get(), Set.copyOf(resources));
            restServer.startAsync();
            restServer.awaitRunning();
        }
        // Wait for controller event processors to start.
        if (serviceConfig.getEventProcessorConfig().isPresent()) {
            // if store client has failed because of session expiration, there are two possibilities where
            // controllerEventProcessors.awaitRunning may be stuck forever -
            // 1. stream creation is retried indefinitely and cannot complete because of zk session expiration
            // 2. event writer after stream creation throws exception.
            // In both of above cases controllerEventProcessors.startAsync may not get called.
            CompletableFuture.anyOf(storeClientFailureFuture, eventProcessorFuture.thenAccept(x -> controllerEventProcessors.awaitRunning())).join();
        }
        // Wait for controller cluster listeners to start.
        if (serviceConfig.isControllerClusterListenerEnabled()) {
            controllerClusterListener.awaitRunning();
        }
    } catch (Exception e) {
        log.error("Failed trying to start controller services", e);
        throw e;
    } finally {
        LoggerHelpers.traceLeave(log, this.objectId, "startUp", traceId);
    }
}

Example 2 with HealthServiceManager

use of io.pravega.shared.health.HealthServiceManager in project pravega by pravega.

the class ServiceStarter method start.

// endregion
// region Service Operation
public void start() throws Exception {
    Exceptions.checkNotClosed(this.closed, this);
    healthServiceManager = new HealthServiceManager(serviceConfig.getHealthCheckInterval());
    healthServiceManager.start();
    log.info("Initializing HealthService ...");
    MetricsConfig metricsConfig = builderConfig.getConfig(MetricsConfig::builder);
    if (metricsConfig.isEnableStatistics()) {
        log.info("Initializing metrics provider ...");
        MetricsProvider.initialize(metricsConfig);
        statsProvider = MetricsProvider.getMetricsProvider();
        statsProvider.start();
    }
    log.info("Initializing ZooKeeper Client ...");
    this.zkClient = createZKClient();
    log.info("Initializing Service Builder ...");
    this.serviceBuilder.initialize();
    log.info("Creating StreamSegmentService ...");
    StreamSegmentStore service = this.serviceBuilder.createStreamSegmentService();
    log.info("Creating TableStoreService ...");
    TableStore tableStoreService = this.serviceBuilder.createTableStoreService();
    log.info("Creating Segment Stats recorder ...");
    autoScaleMonitor = new AutoScaleMonitor(service, builderConfig.getConfig(AutoScalerConfig::builder));
    AutoScalerConfig autoScalerConfig = builderConfig.getConfig(AutoScalerConfig::builder);
    TokenVerifierImpl tokenVerifier = null;
    if (autoScalerConfig.isAuthEnabled()) {
        tokenVerifier = new TokenVerifierImpl(autoScalerConfig.getTokenSigningKey());
    }
    // Log the configuration
    log.info(serviceConfig.toString());
    log.info(autoScalerConfig.toString());
    this.listener = new PravegaConnectionListener(this.serviceConfig.isEnableTls(), this.serviceConfig.isEnableTlsReload(), this.serviceConfig.getListeningIPAddress(), this.serviceConfig.getListeningPort(), service, tableStoreService, autoScaleMonitor.getStatsRecorder(), autoScaleMonitor.getTableSegmentStatsRecorder(), tokenVerifier, this.serviceConfig.getCertFile(), this.serviceConfig.getKeyFile(), this.serviceConfig.isReplyWithStackTraceOnError(), serviceBuilder.getLowPriorityExecutor(), this.serviceConfig.getTlsProtocolVersion(), healthServiceManager);
    this.listener.startListening();
    log.info("PravegaConnectionListener started successfully.");
    if (serviceConfig.isEnableAdminGateway()) {
        this.adminListener = new AdminConnectionListener(this.serviceConfig.isEnableTls(), this.serviceConfig.isEnableTlsReload(), this.serviceConfig.getListeningIPAddress(), this.serviceConfig.getAdminGatewayPort(), service, tableStoreService, tokenVerifier, this.serviceConfig.getCertFile(), this.serviceConfig.getKeyFile(), this.serviceConfig.getTlsProtocolVersion(), healthServiceManager);
        this.adminListener.startListening();
        log.info("AdminConnectionListener started successfully.");
    }
    log.info("StreamSegmentService started.");
    healthServiceManager.register(new ZKHealthContributor(zkClient));
    healthServiceManager.register(new CacheManagerHealthContributor(serviceBuilder.getCacheManager()));
    healthServiceManager.register(new SegmentContainerRegistryHealthContributor(serviceBuilder.getSegmentContainerRegistry()));
    if (this.serviceConfig.isRestServerEnabled()) {
        log.info("Initializing RESTServer ...");
        List<Object> resources = new ArrayList<>();
        resources.add(new HealthImpl(new AuthHandlerManager(serviceConfig.getRestServerConfig()), healthServiceManager.getEndpoint()));
        MetricsProvider.getMetricsProvider().prometheusResource().ifPresent(resources::add);
        restServer = new RESTServer(serviceConfig.getRestServerConfig(), Set.copyOf(resources));
        restServer.startAsync();
        restServer.awaitRunning();
    }
}

Example 3 with HealthServiceManager

use of io.pravega.shared.health.HealthServiceManager in project pravega by pravega.

the class PravegaConnectionListenerTest method testHealth.

// Test the health status created with pravega listener.
@Test
public void testHealth() {
    @Cleanup HealthServiceManager healthServiceManager = new HealthServiceManager(Duration.ofSeconds(2));
    healthServiceManager.start();
    int port = TestUtils.getAvailableListenPort();
    @Cleanup PravegaConnectionListener listener = new PravegaConnectionListener(false, false, "localhost", port, mock(StreamSegmentStore.class), mock(TableStore.class), SegmentStatsRecorder.noOp(), TableSegmentStatsRecorder.noOp(), new PassingTokenVerifier(), null, null, true, NoOpScheduledExecutor.get(), TLS_PROTOCOL_VERSION.getDefaultValue().split(","), healthServiceManager);
    listener.startListening();
    Health health = listener.getHealthServiceManager().getHealthSnapshot();
    Assert.assertEquals("HealthContributor should report an 'UP' Status.", Status.UP, health.getStatus());
    listener.close();
    health = listener.getHealthServiceManager().getHealthSnapshot();
    Assert.assertEquals("HealthContributor should report an 'DOWN' Status.", Status.DOWN, health.getStatus());
}

Example 4 with HealthServiceManager

use of io.pravega.shared.health.HealthServiceManager in project pravega by pravega.

the class HealthTests method setup.

@Before
public void setup() throws Exception {
    serverConfig = getServerConfig();
    healthServiceManager = new HealthServiceManager(Duration.ofMillis(100));
    restServer = new RESTServer(serverConfig, Set.of(new HealthImpl(null, healthServiceManager.getEndpoint())));
    healthServiceManager.start();
    restServer.startAsync();
    restServer.awaitRunning();
    client = createJerseyClient();
}