fix(metrics): release listener on TLS startup failure (#8527)

plugin/metrics/metrics.go

@@ -40,15 +40,17 @@ type Metrics struct {
 	plugins map[string]struct{} // all available plugins, used to determine which plugin made the client write
 
 	tlsConfigPath string
+	serveTLS      func(net.Listener, *http.Server, *web.FlagConfig, *slog.Logger) error
 }
 
 // New returns a new instance of Metrics with the given address.
 func New(addr string) *Metrics {
 	met := &Metrics{
-		Addr:    addr,
-		Reg:     prometheus.DefaultRegisterer.(*prometheus.Registry),
-		zoneMap: make(map[string]struct{}),
-		plugins: pluginList(caddy.ListPlugins()),
+		Addr:     addr,
+		Reg:      prometheus.DefaultRegisterer.(*prometheus.Registry),
+		zoneMap:  make(map[string]struct{}),
+		plugins:  pluginList(caddy.ListPlugins()),
+		serveTLS: web.Serve,
 	}
 
 	return met
@@ -117,6 +119,13 @@ func (sl *startupListener) Ready() <-chan struct{} {
 
 // OnStartup sets up the metrics on startup.
 func (m *Metrics) OnStartup() error {
+	if m.tlsConfigPath != "" {
+		if err := web.Validate(m.tlsConfigPath); err != nil {
+			log.Errorf("Invalid TLS config: %s", err)
+			return err
+		}
+	}
+
 	ln, err := reuseport.Listen("tcp", m.Addr)
 	if err != nil {
 		log.Errorf("Failed to start metrics handler: %s", err)
@@ -151,12 +160,6 @@ func (m *Metrics) OnStartup() error {
 		return nil
 	}
 
-	// Check TLS config file existence
-	if _, err := os.Stat(m.tlsConfigPath); os.IsNotExist(err) {
-		log.Errorf("TLS config file does not exist: %s", m.tlsConfigPath)
-		return err
-	}
-
 	// Create web config for ListenAndServe
 	webConfig := &web.FlagConfig{
 		WebListenAddresses: &[]string{m.Addr},
@@ -173,7 +176,7 @@ func (m *Metrics) OnStartup() error {
 		// web.Serve() never returns nil, it always returns a non-nil error and
 		// it doesn't retun anything if server starts successfully.
 		// startupListener handles capturing succesful startup.
-		err := web.Serve(m.ln, server, webConfig, logger)
+		err := m.serveTLS(m.ln, server, webConfig, logger)
 		if err != nil && err != http.ErrServerClosed {
 			log.Errorf("Failed to start HTTPS metrics server: %v", err)
 			startUpErr <- err
@@ -183,6 +186,10 @@ func (m *Metrics) OnStartup() error {
 	// Wait for startup errors
 	select {
 	case err := <-startUpErr:
+		if closeErr := ln.Close(); closeErr != nil {
+			log.Errorf("Failed to close metrics listener after startup error: %s", closeErr)
+		}
+		m.lnSetup = false
 		return err
 	case <-startupListener.Ready():
 		log.Infof("Server is ready and accepting connections")

plugin/metrics/metrics_test.go

@@ -8,8 +8,10 @@ import (
 	"crypto/x509"
 	"crypto/x509/pkix"
 	"encoding/pem"
+	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"math/big"
 	"net"
 	"net/http"
@@ -24,6 +26,7 @@ import (
 	"github.com/miekg/dns"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
+	"github.com/prometheus/exporter-toolkit/web"
 )
 
 const (
@@ -408,6 +411,63 @@ func TestMetricsTLS(t *testing.T) {
 	}
 }
 
+func TestMetricsTLSValidationErrorDoesNotClaimAddress(t *testing.T) {
+	probe, err := net.Listen("tcp", "localhost:0")
+	if err != nil {
+		t.Fatalf("failed to reserve test address: %v", err)
+	}
+	addr := probe.Addr().String()
+	if err := probe.Close(); err != nil {
+		t.Fatalf("failed to release test address: %v", err)
+	}
+
+	met := New(addr)
+	met.tlsConfigPath = "test_data/configs/junk.yml"
+	if err := met.OnStartup(); err == nil {
+		t.Fatal("expected invalid TLS config to fail startup")
+	}
+	if met.lnSetup {
+		t.Fatal("listener marked as set up after failed startup")
+	}
+
+	listener, err := net.Listen("tcp", addr)
+	if err != nil {
+		t.Fatalf("failed startup kept metrics address %s: %v", addr, err)
+	}
+	listener.Close()
+}
+
+func TestMetricsTLSServeErrorReleasesAddress(t *testing.T) {
+	probe, err := net.Listen("tcp", "localhost:0")
+	if err != nil {
+		t.Fatalf("failed to reserve test address: %v", err)
+	}
+	addr := probe.Addr().String()
+	if err := probe.Close(); err != nil {
+		t.Fatalf("failed to release test address: %v", err)
+	}
+
+	serveErr := errors.New("injected TLS serve failure")
+	met := New(addr)
+	met.tlsConfigPath = "test_data/configs/empty.yml"
+	met.serveTLS = func(net.Listener, *http.Server, *web.FlagConfig, *slog.Logger) error {
+		return serveErr
+	}
+
+	if err := met.OnStartup(); !errors.Is(err, serveErr) {
+		t.Fatalf("expected injected TLS serve error, got %v", err)
+	}
+	if met.lnSetup {
+		t.Fatal("listener marked as set up after TLS serve failure")
+	}
+
+	listener, err := net.Listen("tcp", addr)
+	if err != nil {
+		t.Fatalf("TLS serve failure kept metrics address %s: %v", addr, err)
+	}
+	listener.Close()
+}
+
 func TestMetrics(t *testing.T) {
 	met := New("localhost:0")
 	if err := met.OnStartup(); err != nil {