Improve signal handling for shutdown

- Initiates graceful shutdown upon receiving SIGINT or SIGTERM.
- Logs "Initiating graceful shutdown, repeat signal to force shutdown"
  after canceling the context.
- Adds logic to handle repeated signals to trigger a forced shutdown.
- Implements a 30-second timeout for graceful shutdown before exiting
  with an error.
- Ensures clarity in logs and robust shutdown handling.

cmd/scality-cosi-driver/main.go

@@ -36,19 +36,23 @@ func main() {
 	go func() {
 		sig := <-sigs
 		klog.InfoS("Signal received", "type", sig)
-		cancel() // Trigger context cancellation
+		cancel()
+
+		klog.InfoS("Initiating graceful shutdown, repeat signal to force shutdown")
 
 		select {
-		case <-ctx.Done():
-			klog.InfoS("Scality COSI driver shutdown initiated successfully, context canceled")
+		case sig = <-sigs:
+			klog.ErrorS(nil, "Force shutdown due to repeated signal", "type", sig)
+			os.Exit(1)
 		case <-time.After(30 * time.Second):
-			klog.ErrorS(nil, "Scality COSI driver graceful shutdown timed out, forcing application exit after 30 seconds")
+			klog.ErrorS(nil, "Force shutdown due to timeout", "timeout", 30*time.Second)
 			os.Exit(1)
 		}
 	}()
 
 	// Call the run function (defined in cmd.go)
 	if err := run(ctx); err != nil {
 		klog.ErrorS(err, "Scality COSI driver encountered an error, shutting down")
+		os.Exit(1)
 	}
 }

pkg/grpcfactory/server_test.go

@@ -11,6 +11,7 @@ import (
 
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/scality/cosi-driver/pkg/grpcfactory"
 	cosi "sigs.k8s.io/container-object-storage-interface-spec"
 )
@@ -59,7 +60,7 @@ var _ = Describe("gRPC Factory Server", Ordered, func() {
 			Expect(server).NotTo(BeNil())
 
 			go func() {
-				err := server.Run(ctx)
+				err := server.Run(ctx, prometheus.NewRegistry())
 				if errors.Is(err, context.Canceled) {
 					return // Expected when the context is canceled
 				}
@@ -81,7 +82,7 @@ var _ = Describe("gRPC Factory Server", Ordered, func() {
 			Expect(server2).NotTo(BeNil())
 
 			// Run the second server and expect it to fail
-			err = server2.Run(ctx)
+			err = server2.Run(ctx, prometheus.NewRegistry())
 			Expect(err).To(HaveOccurred())
 			Expect(err.Error()).To(ContainSubstring("address already in use"))
 		}, SpecTimeout(1*time.Second))
@@ -94,7 +95,7 @@ var _ = Describe("gRPC Factory Server", Ordered, func() {
 			Expect(server).NotTo(BeNil())
 
 			// Wait for server.Run to return an error
-			err = server.Run(ctx)
+			err = server.Run(ctx, prometheus.NewRegistry())
 			Expect(err).To(HaveOccurred())
 			Expect(err.Error()).To(ContainSubstring("unsupported scheme: expected 'unix'"))
 		}, SpecTimeout(1*time.Second))