Remove fatal logging in executable code

All Fatal log call are removed. Also all runnable components
are now managed by controller-runtime, implementing
manager.Runnable interface.

Author: tchap

pkg/ext-proc/internal/runnable/grpc.go

@@ -0,0 +1,52 @@
+package runnable
+
+import (
+	"context"
+	"fmt"
+	"net"
+
+	"google.golang.org/grpc"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/manager"
+)
+
+// GRPCServer converts the given gRPC server into a runnable.
+// The server name is just being used for logging.
+func GRPCServer(name string, srv *grpc.Server, port int) manager.Runnable {
+	return manager.RunnableFunc(func(ctx context.Context) error {
+		// Use "name" key as that is what manager.Server does as well.
+		log := ctrl.Log.WithValues("name", name)
+		log.Info("gRPC server starting")
+
+		// Start listening.
+		lis, err := net.Listen("tcp", fmt.Sprintf(":%d", port))
+		if err != nil {
+			log.Error(err, "gRPC server failed to listen")
+			return err
+		}
+
+		log.Info("gRPC server listening", "port", port)
+
+		// Shutdown on context closed.
+		// Terminate the server on context closed.
+		// Make sure the goroutine does not leak.
+		doneCh := make(chan struct{})
+		defer close(doneCh)
+		go func() {
+			select {
+			case <-ctx.Done():
+				log.Info("gRPC server shutting down")
+				srv.GracefulStop()
+			case <-doneCh:
+			}
+		}()
+
+		// Keep serving until terminated.
+		if err := srv.Serve(lis); err != nil && err != grpc.ErrServerStopped {
+			log.Error(err, "gRPC server failed")
+			return err
+		}
+		log.Info("gRPC server terminated")
+		return nil
+	})
+}

pkg/ext-proc/internal/runnable/leader_election.go

@@ -0,0 +1,31 @@
+package runnable
+
+import "sigs.k8s.io/controller-runtime/pkg/manager"
+
+type leaderElection struct {
+	manager.Runnable
+	needsLeaderElection bool
+}
+
+// LeaderElection wraps the given runnable to implement manager.LeaderElectionRunnable.
+func LeaderElection(runnable manager.Runnable, needsLeaderElection bool) manager.Runnable {
+	return &leaderElection{
+		Runnable:            runnable,
+		needsLeaderElection: needsLeaderElection,
+	}
+}
+
+// RequireLeaderElection wraps the given runnable, marking it as requiring leader election.
+func RequireLeaderElection(runnable manager.Runnable) manager.Runnable {
+	return LeaderElection(runnable, true)
+}
+
+// RequireLeaderElection wraps the given runnable, marking it as not requiring leader election.
+func NoLeaderElection(runnable manager.Runnable) manager.Runnable {
+	return LeaderElection(runnable, false)
+}
+
+// NeedLeaderElection implements manager.NeedLeaderElection interface.
+func (r *leaderElection) NeedLeaderElection() bool {
+	return r.needsLeaderElection
+}

pkg/ext-proc/main.go

@@ -1,11 +1,11 @@
 package main
 
 import (
-	"context"
 	"flag"
 	"fmt"
 	"net"
 	"net/http"
+	"os"
 	"strconv"
 
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -14,6 +14,7 @@ import (
 	"inference.networking.x-k8s.io/gateway-api-inference-extension/api/v1alpha1"
 	"inference.networking.x-k8s.io/gateway-api-inference-extension/pkg/ext-proc/backend"
 	"inference.networking.x-k8s.io/gateway-api-inference-extension/pkg/ext-proc/backend/vllm"
+	"inference.networking.x-k8s.io/gateway-api-inference-extension/pkg/ext-proc/internal/runnable"
 	"inference.networking.x-k8s.io/gateway-api-inference-extension/pkg/ext-proc/metrics"
 	runserver "inference.networking.x-k8s.io/gateway-api-inference-extension/pkg/ext-proc/server"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -23,6 +24,7 @@ import (
 	"k8s.io/component-base/metrics/legacyregistry"
 	klog "k8s.io/klog/v2"
 	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/manager"
 	"sigs.k8s.io/controller-runtime/pkg/metrics/filters"
 )
 
@@ -79,17 +81,25 @@ func init() {
 }
 
 func main() {
+	if err := run(); err != nil {
+		os.Exit(1)
+	}
+}
+
+func run() error {
 	klog.InitFlags(nil)
 	flag.Parse()
 
 	ctrl.SetLogger(klog.TODO())
 	cfg, err := ctrl.GetConfig()
 	if err != nil {
-		klog.Fatalf("Failed to get rest config: %v", err)
+		klog.ErrorS(err, "Failed to get rest config")
+		return err
 	}
 	// Validate flags
 	if err := validateFlags(); err != nil {
-		klog.Fatalf("Failed to validate flags: %v", err)
+		klog.ErrorS(err, "Failed to validate flags")
+		return err
 	}
 
 	// Print all flag values
@@ -114,101 +124,95 @@ func main() {
 		Config:                 ctrl.GetConfigOrDie(),
 		Datastore:              datastore,
 	}
-	serverRunner.Setup()
-
-	// Start health and ext-proc servers in goroutines
-	healthSvr := startHealthServer(datastore, *grpcHealthPort)
-	extProcSvr := serverRunner.Start(
-		datastore,
-		&vllm.PodMetricsClientImpl{},
-	)
-	// Start metrics handler
-	metricsSvr := startMetricsHandler(*metricsPort, cfg)
-
-	// Start manager, blocking
-	serverRunner.StartManager()
+	if err := serverRunner.Setup(); err != nil {
+		klog.ErrorS(err, "Failed to setup ext-proc server")
+		return err
+	}
+	mgr := serverRunner.Manager
 
-	// Gracefully shutdown servers
-	if healthSvr != nil {
-		klog.Info("Health server shutting down")
-		healthSvr.GracefulStop()
+	// Register health server.
+	if err := registerHealthServer(mgr, datastore, *grpcHealthPort); err != nil {
+		return err
 	}
-	if extProcSvr != nil {
-		klog.Info("Ext-proc server shutting down")
-		extProcSvr.GracefulStop()
+
+	// Register ext-proc server.
+	if err := mgr.Add(serverRunner.AsRunnable(datastore, &vllm.PodMetricsClientImpl{})); err != nil {
+		klog.ErrorS(err, "Failed to register ext-proc server")
+		return err
 	}
-	if metricsSvr != nil {
-		klog.Info("Metrics server shutting down")
-		if err := metricsSvr.Shutdown(context.Background()); err != nil {
-			klog.Infof("Metrics server Shutdown: %v", err)
-		}
+
+	// Register metrics handler.
+	if err := registerMetricsHandler(mgr, *metricsPort, cfg); err != nil {
+		return err
 	}
 
-	klog.Info("All components shutdown")
+	// Start the manager.
+	return serverRunner.StartManager(ctrl.SetupSignalHandler())
 }
 
-// startHealthServer starts the gRPC health probe server in a goroutine.
-func startHealthServer(ds *backend.K8sDatastore, port int) *grpc.Server {
-	svr := grpc.NewServer()
-	healthPb.RegisterHealthServer(svr, &healthServer{datastore: ds})
-
-	go func() {
-		lis, err := net.Listen("tcp", fmt.Sprintf(":%d", port))
-		if err != nil {
-			klog.Fatalf("Health server failed to listen: %v", err)
-		}
-		klog.Infof("Health server listening on port: %d", port)
-
-		// Blocking and will return when shutdown is complete.
-		if err := svr.Serve(lis); err != nil && err != grpc.ErrServerStopped {
-			klog.Fatalf("Health server failed: %v", err)
-		}
-		klog.Info("Health server shutting down")
-	}()
-	return svr
+// registerHealthServer adds the Health gRPC server as a Runnable to the given manager.
+func registerHealthServer(mgr manager.Manager, ds *backend.K8sDatastore, port int) error {
+	srv := grpc.NewServer()
+	healthPb.RegisterHealthServer(srv, &healthServer{datastore: ds})
+	if err := mgr.Add(
+		runnable.NoLeaderElection(runnable.GRPCServer("health", srv, port))); err != nil {
+		klog.ErrorS(err, "Failed to register health server")
+		return err
+	}
+	return nil
 }
 
-func startMetricsHandler(port int, cfg *rest.Config) *http.Server {
+// registerMetricsHandler adds the metrics HTTP handler as a Runnable to the given manager.
+func registerMetricsHandler(mgr manager.Manager, port int, cfg *rest.Config) error {
 	metrics.Register()
 
-	var svr *http.Server
-	go func() {
-		klog.Info("Starting metrics HTTP handler ...")
-
-		mux := http.NewServeMux()
-		mux.Handle(defaultMetricsEndpoint, metricsHandlerWithAuthenticationAndAuthorization(cfg))
-
-		svr = &http.Server{
-			Addr:    net.JoinHostPort("", strconv.Itoa(port)),
-			Handler: mux,
-		}
-		if err := svr.ListenAndServe(); err != http.ErrServerClosed {
-			klog.Fatalf("failed to start metrics HTTP handler: %v", err)
-		}
-	}()
-	return svr
+	// Init HTTP server.
+	h, err := metricsHandlerWithAuthenticationAndAuthorization(cfg)
+	if err != nil {
+		return err
+	}
+
+	mux := http.NewServeMux()
+	mux.Handle(defaultMetricsEndpoint, h)
+
+	srv := &http.Server{
+		Addr:    net.JoinHostPort("", strconv.Itoa(port)),
+		Handler: mux,
+	}
+
+	if err := mgr.Add(&manager.Server{
+		Name:   "metrics",
+		Server: srv,
+	}); err != nil {
+		klog.ErrorS(err, "Failed to register metrics HTTP handler")
+		return err
+	}
+	return nil
 }
 
-func metricsHandlerWithAuthenticationAndAuthorization(cfg *rest.Config) http.Handler {
+func metricsHandlerWithAuthenticationAndAuthorization(cfg *rest.Config) (http.Handler, error) {
 	h := promhttp.HandlerFor(
 		legacyregistry.DefaultGatherer,
 		promhttp.HandlerOpts{},
 	)
 	httpClient, err := rest.HTTPClientFor(cfg)
 	if err != nil {
-		klog.Fatalf("failed to create http client for metrics auth: %v", err)
+		klog.ErrorS(err, "Failed to create http client for metrics auth")
+		return nil, err
 	}
 
 	filter, err := filters.WithAuthenticationAndAuthorization(cfg, httpClient)
 	if err != nil {
-		klog.Fatalf("failed to create metrics filter for auth: %v", err)
+		klog.ErrorS(err, "Failed to create metrics filter for auth")
+		return nil, err
 	}
 	metricsLogger := klog.LoggerWithValues(klog.NewKlogr(), "path", defaultMetricsEndpoint)
 	metricsAuthHandler, err := filter(metricsLogger, h)
 	if err != nil {
-		klog.Fatalf("failed to create metrics auth handler: %v", err)
+		klog.ErrorS(err, "Failed to create metrics auth handler")
+		return nil, err
 	}
-	return metricsAuthHandler
+	return metricsAuthHandler, nil
 }
 
 func validateFlags() error {