Add cert health info to checks

proxy/internal/acme/manager.go

@@ -8,12 +8,15 @@ import (
 	"encoding/binary"
 	"fmt"
 	"net"
+	"slices"
 	"sync"
 	"time"
 
 	log "github.com/sirupsen/logrus"
 	"golang.org/x/crypto/acme"
 	"golang.org/x/crypto/acme/autocert"
+
+	"github.com/netbirdio/netbird/shared/management/domain"
 )
 
 // OID for the SCT list extension (1.3.6.1.4.1.11129.2.4.2)
@@ -23,19 +26,31 @@ type certificateNotifier interface {
 	NotifyCertificateIssued(ctx context.Context, accountID, reverseProxyID, domain string) error
 }
 
+type domainState int
+
+const (
+	domainPending domainState = iota
+	domainReady
+	domainFailed
+)
+
+type domainInfo struct {
+	accountID      string
+	reverseProxyID string
+	state          domainState
+	err            string
+}
+
 // Manager wraps autocert.Manager with domain tracking and cross-replica
 // coordination via a pluggable locking strategy. The locker prevents
 // duplicate ACME requests when multiple replicas share a certificate cache.
 type Manager struct {
 	*autocert.Manager
 
-	certDir    string
-	locker     certLocker
-	domainsMux sync.RWMutex
-	domains    map[string]struct {
-		accountID      string
-		reverseProxyID string
-	}
+	certDir string
+	locker  certLocker
+	mu      sync.RWMutex
+	domains map[domain.Domain]*domainInfo
 
 	certNotifier certificateNotifier
 	logger       *log.Logger
@@ -49,12 +64,9 @@ func NewManager(certDir, acmeURL string, notifier certificateNotifier, logger *l
 		logger = log.StandardLogger()
 	}
 	mgr := &Manager{
-		certDir: certDir,
-		locker:  newCertLocker(lockMethod, certDir, logger),
-		domains: make(map[string]struct {
-			accountID      string
-			reverseProxyID string
-		}),
+		certDir:      certDir,
+		locker:       newCertLocker(lockMethod, certDir, logger),
+		domains:      make(map[domain.Domain]*domainInfo),
 		certNotifier: notifier,
 		logger:       logger,
 	}
@@ -73,49 +85,50 @@ func (mgr *Manager) hostPolicy(_ context.Context, host string) error {
 	if h, _, err := net.SplitHostPort(host); err == nil {
 		host = h
 	}
-	mgr.domainsMux.RLock()
-	_, exists := mgr.domains[host]
-	mgr.domainsMux.RUnlock()
+	mgr.mu.RLock()
+	_, exists := mgr.domains[domain.Domain(host)]
+	mgr.mu.RUnlock()
 	if !exists {
 		return fmt.Errorf("unknown domain %q", host)
 	}
 	return nil
 }
 
-func (mgr *Manager) AddDomain(domain, accountID, reverseProxyID string) {
-	mgr.domainsMux.Lock()
-	mgr.domains[domain] = struct {
-		accountID      string
-		reverseProxyID string
-	}{
+// AddDomain registers a domain for ACME certificate prefetching.
+func (mgr *Manager) AddDomain(d domain.Domain, accountID, reverseProxyID string) {
+	mgr.mu.Lock()
+	mgr.domains[d] = &domainInfo{
 		accountID:      accountID,
 		reverseProxyID: reverseProxyID,
+		state:          domainPending,
 	}
-	mgr.domainsMux.Unlock()
+	mgr.mu.Unlock()
 
-	go mgr.prefetchCertificate(domain)
+	go mgr.prefetchCertificate(d)
 }
 
 // prefetchCertificate proactively triggers certificate generation for a domain.
 // It acquires a distributed lock to prevent multiple replicas from issuing
 // duplicate ACME requests. The second replica will block until the first
 // finishes, then find the certificate in the cache.
-func (mgr *Manager) prefetchCertificate(domain string) {
+func (mgr *Manager) prefetchCertificate(d domain.Domain) {
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
 	defer cancel()
 
-	mgr.logger.Infof("acquiring cert lock for domain %q", domain)
+	name := d.PunycodeString()
+
+	mgr.logger.Infof("acquiring cert lock for domain %q", name)
 	lockStart := time.Now()
-	unlock, err := mgr.locker.Lock(ctx, domain)
+	unlock, err := mgr.locker.Lock(ctx, name)
 	if err != nil {
-		mgr.logger.Warnf("acquire cert lock for domain %q, proceeding without lock: %v", domain, err)
+		mgr.logger.Warnf("acquire cert lock for domain %q, proceeding without lock: %v", name, err)
 	} else {
-		mgr.logger.Infof("acquired cert lock for domain %q in %s", domain, time.Since(lockStart))
+		mgr.logger.Infof("acquired cert lock for domain %q in %s", name, time.Since(lockStart))
 		defer unlock()
 	}
 
 	hello := &tls.ClientHelloInfo{
-		ServerName: domain,
+		ServerName: name,
 		Conn:       &dummyConn{ctx: ctx},
 	}
 
@@ -123,35 +136,47 @@ func (mgr *Manager) prefetchCertificate(domain string) {
 	cert, err := mgr.GetCertificate(hello)
 	elapsed := time.Since(start)
 	if err != nil {
-		mgr.logger.Warnf("prefetch certificate for domain %q: %v", domain, err)
+		mgr.logger.Warnf("prefetch certificate for domain %q: %v", name, err)
+		mgr.setDomainState(d, domainFailed, err.Error())
 		return
 	}
 
+	mgr.setDomainState(d, domainReady, "")
+
 	now := time.Now()
 	if cert != nil && cert.Leaf != nil {
 		leaf := cert.Leaf
 		mgr.logger.Infof("certificate for domain %q ready in %s: serial=%s SANs=%v notAfter=%s",
-			domain, elapsed.Round(time.Millisecond),
+			name, elapsed.Round(time.Millisecond),
 			leaf.SerialNumber.Text(16),
 			leaf.DNSNames,
 			leaf.NotAfter.UTC().Format(time.RFC3339),
 		)
-		mgr.logCertificateDetails(domain, leaf, now)
+		mgr.logCertificateDetails(name, leaf, now)
 	} else {
-		mgr.logger.Infof("certificate for domain %q ready in %s", domain, elapsed.Round(time.Millisecond))
+		mgr.logger.Infof("certificate for domain %q ready in %s", name, elapsed.Round(time.Millisecond))
 	}
 
-	mgr.domainsMux.RLock()
-	info, exists := mgr.domains[domain]
-	mgr.domainsMux.RUnlock()
+	mgr.mu.RLock()
+	info := mgr.domains[d]
+	mgr.mu.RUnlock()
 
-	if exists && mgr.certNotifier != nil {
-		if err := mgr.certNotifier.NotifyCertificateIssued(ctx, info.accountID, info.reverseProxyID, domain); err != nil {
-			mgr.logger.Warnf("notify certificate ready for domain %q: %v", domain, err)
+	if info != nil && mgr.certNotifier != nil {
+		if err := mgr.certNotifier.NotifyCertificateIssued(ctx, info.accountID, info.reverseProxyID, name); err != nil {
+			mgr.logger.Warnf("notify certificate ready for domain %q: %v", name, err)
 		}
 	}
 }
 
+func (mgr *Manager) setDomainState(d domain.Domain, state domainState, errMsg string) {
+	mgr.mu.Lock()
+	defer mgr.mu.Unlock()
+	if info, ok := mgr.domains[d]; ok {
+		info.state = state
+		info.err = errMsg
+	}
+}
+
 // logCertificateDetails logs certificate validity and SCT timestamps.
 func (mgr *Manager) logCertificateDetails(domain string, cert *x509.Certificate, now time.Time) {
 	if cert.NotBefore.After(now) {
@@ -245,8 +270,65 @@ func (c *dummyConn) SetDeadline(t time.Time) error      { return nil }
 func (c *dummyConn) SetReadDeadline(t time.Time) error  { return nil }
 func (c *dummyConn) SetWriteDeadline(t time.Time) error { return nil }
 
-func (mgr *Manager) RemoveDomain(domain string) {
-	mgr.domainsMux.Lock()
-	defer mgr.domainsMux.Unlock()
-	delete(mgr.domains, domain)
+// RemoveDomain removes a domain from tracking.
+func (mgr *Manager) RemoveDomain(d domain.Domain) {
+	mgr.mu.Lock()
+	defer mgr.mu.Unlock()
+	delete(mgr.domains, d)
+}
+
+// PendingCerts returns the number of certificates currently being prefetched.
+func (mgr *Manager) PendingCerts() int {
+	mgr.mu.RLock()
+	defer mgr.mu.RUnlock()
+	var n int
+	for _, info := range mgr.domains {
+		if info.state == domainPending {
+			n++
+		}
+	}
+	return n
+}
+
+// TotalDomains returns the total number of registered domains.
+func (mgr *Manager) TotalDomains() int {
+	mgr.mu.RLock()
+	defer mgr.mu.RUnlock()
+	return len(mgr.domains)
+}
+
+// PendingDomains returns the domain names currently being prefetched.
+func (mgr *Manager) PendingDomains() []string {
+	return mgr.domainsByState(domainPending)
+}
+
+// ReadyDomains returns domain names that have successfully obtained certificates.
+func (mgr *Manager) ReadyDomains() []string {
+	return mgr.domainsByState(domainReady)
+}
+
+// FailedDomains returns domain names that failed certificate prefetch, mapped to their error.
+func (mgr *Manager) FailedDomains() map[string]string {
+	mgr.mu.RLock()
+	defer mgr.mu.RUnlock()
+	result := make(map[string]string)
+	for d, info := range mgr.domains {
+		if info.state == domainFailed {
+			result[d.PunycodeString()] = info.err
+		}
+	}
+	return result
+}
+
+func (mgr *Manager) domainsByState(state domainState) []string {
+	mgr.mu.RLock()
+	defer mgr.mu.RUnlock()
+	var domains []string
+	for d, info := range mgr.domains {
+		if info.state == state {
+			domains = append(domains, d.PunycodeString())
+		}
+	}
+	slices.Sort(domains)
+	return domains
 }

proxy/internal/acme/manager_test.go

@@ -3,6 +3,7 @@ package acme
 import (
 	"context"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -12,6 +13,14 @@ func TestHostPolicy(t *testing.T) {
 	mgr := NewManager(t.TempDir(), "https://acme.example.com/directory", nil, nil, "")
 	mgr.AddDomain("example.com", "acc1", "rp1")
 
+	// Wait for the background prefetch goroutine to finish so the temp dir
+	// can be cleaned up without a race.
+	t.Cleanup(func() {
+		assert.Eventually(t, func() bool {
+			return mgr.PendingCerts() == 0
+		}, 30*time.Second, 50*time.Millisecond)
+	})
+
 	tests := []struct {
 		name    string
 		host    string
@@ -59,3 +68,35 @@ func TestHostPolicy(t *testing.T) {
 		})
 	}
 }
+
+func TestDomainStates(t *testing.T) {
+	mgr := NewManager(t.TempDir(), "https://acme.example.com/directory", nil, nil, "")
+
+	assert.Equal(t, 0, mgr.PendingCerts(), "initially zero")
+	assert.Equal(t, 0, mgr.TotalDomains(), "initially zero domains")
+	assert.Empty(t, mgr.PendingDomains())
+	assert.Empty(t, mgr.ReadyDomains())
+	assert.Empty(t, mgr.FailedDomains())
+
+	// AddDomain starts as pending, then the prefetch goroutine will fail
+	// (no real ACME server) and transition to failed.
+	mgr.AddDomain("a.example.com", "acc1", "rp1")
+	mgr.AddDomain("b.example.com", "acc1", "rp1")
+
+	assert.Equal(t, 2, mgr.TotalDomains(), "two domains registered")
+
+	// Pending domains should eventually drain after prefetch goroutines finish.
+	assert.Eventually(t, func() bool {
+		return mgr.PendingCerts() == 0
+	}, 30*time.Second, 100*time.Millisecond, "pending certs should return to zero after prefetch completes")
+
+	assert.Empty(t, mgr.PendingDomains())
+	assert.Equal(t, 2, mgr.TotalDomains(), "total domains unchanged")
+
+	// With a fake ACME URL, both should have failed.
+	failed := mgr.FailedDomains()
+	assert.Len(t, failed, 2, "both domains should have failed")
+	assert.Contains(t, failed, "a.example.com")
+	assert.Contains(t, failed, "b.example.com")
+	assert.Empty(t, mgr.ReadyDomains())
+}

proxy/internal/debug/client.go

@@ -56,6 +56,33 @@ func (c *Client) printHealth(data map[string]any) {
 	_, _ = fmt.Fprintf(c.out, "Management Connected: %s\n", boolIcon(data["management_connected"]))
 	_, _ = fmt.Fprintf(c.out, "All Clients Healthy:  %s\n", boolIcon(data["all_clients_healthy"]))
 
+	total, _ := data["certs_total"].(float64)
+	ready, _ := data["certs_ready"].(float64)
+	pending, _ := data["certs_pending"].(float64)
+	failed, _ := data["certs_failed"].(float64)
+	if total > 0 {
+		_, _ = fmt.Fprintf(c.out, "Certificates:         %d ready, %d pending, %d failed (%d total)\n",
+			int(ready), int(pending), int(failed), int(total))
+	}
+	if domains, ok := data["certs_ready_domains"].([]any); ok && len(domains) > 0 {
+		_, _ = fmt.Fprintf(c.out, "  Ready:\n")
+		for _, d := range domains {
+			_, _ = fmt.Fprintf(c.out, "    %v\n", d)
+		}
+	}
+	if domains, ok := data["certs_pending_domains"].([]any); ok && len(domains) > 0 {
+		_, _ = fmt.Fprintf(c.out, "  Pending:\n")
+		for _, d := range domains {
+			_, _ = fmt.Fprintf(c.out, "    %v\n", d)
+		}
+	}
+	if domains, ok := data["certs_failed_domains"].(map[string]any); ok && len(domains) > 0 {
+		_, _ = fmt.Fprintf(c.out, "  Failed:\n")
+		for d, errMsg := range domains {
+			_, _ = fmt.Fprintf(c.out, "    %s: %v\n", d, errMsg)
+		}
+	}
+
 	clients, ok := data["clients"].(map[string]any)
 	if !ok || len(clients) == 0 {
 		return
@@ -328,7 +355,7 @@ func (c *Client) fetch(ctx context.Context, path string) (map[string]any, []byte
 	if err != nil {
 		return nil, nil, fmt.Errorf("request failed: %w", err)
 	}
-	defer resp.Body.Close()
+	defer func() { _ = resp.Body.Close() }()
 
 	body, err := io.ReadAll(resp.Body)
 	if err != nil {
@@ -346,4 +373,3 @@ func (c *Client) fetch(ctx context.Context, path string) (map[string]any, []byte
 
 	return data, body, nil
 }
-

proxy/internal/debug/handler.go

@@ -2,12 +2,15 @@
 package debug
 
 import (
+	"cmp"
 	"context"
 	"embed"
 	"encoding/json"
 	"fmt"
 	"html/template"
+	"maps"
 	"net/http"
+	"slices"
 	"strconv"
 	"strings"
 	"sync"
@@ -47,6 +50,10 @@ func formatDuration(d time.Duration) string {
 	}
 }
 
+func sortedAccountIDs(m map[types.AccountID]roundtrip.ClientDebugInfo) []types.AccountID {
+	return slices.Sorted(maps.Keys(m))
+}
+
 // clientProvider provides access to NetBird clients.
 type clientProvider interface {
 	GetClient(accountID types.AccountID) (*nbembed.Client, bool)
@@ -60,10 +67,18 @@ type healthChecker interface {
 	CheckClientsConnected(ctx context.Context) (bool, map[types.AccountID]health.ClientHealth)
 }
 
+type certStatus interface {
+	TotalDomains() int
+	PendingDomains() []string
+	ReadyDomains() []string
+	FailedDomains() map[string]string
+}
+
 // Handler provides HTTP debug endpoints.
 type Handler struct {
 	provider   clientProvider
 	health     healthChecker
+	certStatus certStatus
 	logger     *log.Logger
 	startTime  time.Time
 	templates  *template.Template
@@ -87,6 +102,11 @@ func NewHandler(provider clientProvider, healthChecker healthChecker, logger *lo
 	return h
 }
 
+// SetCertStatus sets the certificate status provider for ACME prefetch observability.
+func (h *Handler) SetCertStatus(cs certStatus) {
+	h.certStatus = cs
+}
+
 func (h *Handler) loadTemplates() error {
 	tmpl, err := template.ParseFS(templateFS, "templates/*.html")
 	if err != nil {
@@ -160,12 +180,24 @@ func (h *Handler) handleClientRoutes(w http.ResponseWriter, r *http.Request, pat
 	return true
 }
 
+type failedDomain struct {
+	Domain string
+	Error  string
+}
+
 type indexData struct {
-	Version      string
-	Uptime       string
-	ClientCount  int
-	TotalDomains int
-	Clients      []clientData
+	Version             string
+	Uptime              string
+	ClientCount         int
+	TotalDomains        int
+	CertsTotal          int
+	CertsReady          int
+	CertsPending        int
+	CertsFailed         int
+	CertsPendingDomains []string
+	CertsReadyDomains   []string
+	CertsFailedDomains  []failedDomain
+	Clients             []clientData
 }
 
 type clientData struct {
@@ -177,15 +209,30 @@ type clientData struct {
 
 func (h *Handler) handleIndex(w http.ResponseWriter, _ *http.Request, wantJSON bool) {
 	clients := h.provider.ListClientsForDebug()
+	sortedIDs := sortedAccountIDs(clients)
 
 	totalDomains := 0
 	for _, info := range clients {
 		totalDomains += info.DomainCount
 	}
 
+	var certsTotal, certsReady, certsPending, certsFailed int
+	var certsPendingDomains, certsReadyDomains []string
+	var certsFailedDomains map[string]string
+	if h.certStatus != nil {
+		certsTotal = h.certStatus.TotalDomains()
+		certsPendingDomains = h.certStatus.PendingDomains()
+		certsReadyDomains = h.certStatus.ReadyDomains()
+		certsFailedDomains = h.certStatus.FailedDomains()
+		certsReady = len(certsReadyDomains)
+		certsPending = len(certsPendingDomains)
+		certsFailed = len(certsFailedDomains)
+	}
+
 	if wantJSON {
 		clientsJSON := make([]map[string]interface{}, 0, len(clients))
-		for _, info := range clients {
+		for _, id := range sortedIDs {
+			info := clients[id]
 			clientsJSON = append(clientsJSON, map[string]interface{}{
 				"account_id":   info.AccountID,
 				"domain_count": info.DomainCount,
@@ -195,25 +242,55 @@ func (h *Handler) handleIndex(w http.ResponseWriter, _ *http.Request, wantJSON b
 				"age":          time.Since(info.CreatedAt).Round(time.Second).String(),
 			})
 		}
-		h.writeJSON(w, map[string]interface{}{
+		resp := map[string]interface{}{
 			"version":       version.NetbirdVersion(),
 			"uptime":        time.Since(h.startTime).Round(time.Second).String(),
 			"client_count":  len(clients),
 			"total_domains": totalDomains,
+			"certs_total":   certsTotal,
+			"certs_ready":   certsReady,
+			"certs_pending": certsPending,
+			"certs_failed":  certsFailed,
 			"clients":       clientsJSON,
-		})
+		}
+		if len(certsPendingDomains) > 0 {
+			resp["certs_pending_domains"] = certsPendingDomains
+		}
+		if len(certsReadyDomains) > 0 {
+			resp["certs_ready_domains"] = certsReadyDomains
+		}
+		if len(certsFailedDomains) > 0 {
+			resp["certs_failed_domains"] = certsFailedDomains
+		}
+		h.writeJSON(w, resp)
 		return
 	}
 
+	sortedFailed := make([]failedDomain, 0, len(certsFailedDomains))
+	for d, e := range certsFailedDomains {
+		sortedFailed = append(sortedFailed, failedDomain{Domain: d, Error: e})
+	}
+	slices.SortFunc(sortedFailed, func(a, b failedDomain) int {
+		return cmp.Compare(a.Domain, b.Domain)
+	})
+
 	data := indexData{
-		Version:      version.NetbirdVersion(),
-		Uptime:       time.Since(h.startTime).Round(time.Second).String(),
-		ClientCount:  len(clients),
-		TotalDomains: totalDomains,
-		Clients:      make([]clientData, 0, len(clients)),
+		Version:             version.NetbirdVersion(),
+		Uptime:              time.Since(h.startTime).Round(time.Second).String(),
+		ClientCount:         len(clients),
+		TotalDomains:        totalDomains,
+		CertsTotal:          certsTotal,
+		CertsReady:          certsReady,
+		CertsPending:        certsPending,
+		CertsFailed:         certsFailed,
+		CertsPendingDomains: certsPendingDomains,
+		CertsReadyDomains:   certsReadyDomains,
+		CertsFailedDomains:  sortedFailed,
+		Clients:             make([]clientData, 0, len(clients)),
 	}
 
-	for _, info := range clients {
+	for _, id := range sortedIDs {
+		info := clients[id]
 		domains := info.Domains.SafeString()
 		if domains == "" {
 			domains = "-"
@@ -240,10 +317,12 @@ type clientsData struct {
 
 func (h *Handler) handleListClients(w http.ResponseWriter, _ *http.Request, wantJSON bool) {
 	clients := h.provider.ListClientsForDebug()
+	sortedIDs := sortedAccountIDs(clients)
 
 	if wantJSON {
 		clientsJSON := make([]map[string]interface{}, 0, len(clients))
-		for _, info := range clients {
+		for _, id := range sortedIDs {
+			info := clients[id]
 			clientsJSON = append(clientsJSON, map[string]interface{}{
 				"account_id":   info.AccountID,
 				"domain_count": info.DomainCount,
@@ -266,7 +345,8 @@ func (h *Handler) handleListClients(w http.ResponseWriter, _ *http.Request, want
 		Clients: make([]clientData, 0, len(clients)),
 	}
 
-	for _, info := range clients {
+	for _, id := range sortedIDs {
+		info := clients[id]
 		domains := info.Domains.SafeString()
 		if domains == "" {
 			domains = "-"
@@ -556,15 +636,12 @@ func (h *Handler) handleClientStop(w http.ResponseWriter, r *http.Request, accou
 	})
 }
 
-type healthData struct {
-	Uptime              string
-	Status              string
-	ManagementReady     bool
-	AllClientsHealthy   bool
-	Clients             map[types.AccountID]health.ClientHealth
-}
-
 func (h *Handler) handleHealth(w http.ResponseWriter, r *http.Request, wantJSON bool) {
+	if !wantJSON {
+		http.Redirect(w, r, "/debug", http.StatusSeeOther)
+		return
+	}
+
 	uptime := time.Since(h.startTime).Round(10 * time.Millisecond).String()
 
 	ready := h.health.ReadinessProbe()
@@ -575,26 +652,40 @@ func (h *Handler) handleHealth(w http.ResponseWriter, r *http.Request, wantJSON
 		status = "degraded"
 	}
 
-	if wantJSON {
-		h.writeJSON(w, map[string]interface{}{
-			"status":               status,
-			"uptime":               uptime,
-			"management_connected": ready,
-			"all_clients_healthy":  allHealthy,
-			"clients":              clientHealth,
-		})
-		return
+	var certsTotal, certsReady, certsPending, certsFailed int
+	var certsPendingDomains, certsReadyDomains []string
+	var certsFailedDomains map[string]string
+	if h.certStatus != nil {
+		certsTotal = h.certStatus.TotalDomains()
+		certsPendingDomains = h.certStatus.PendingDomains()
+		certsReadyDomains = h.certStatus.ReadyDomains()
+		certsFailedDomains = h.certStatus.FailedDomains()
+		certsReady = len(certsReadyDomains)
+		certsPending = len(certsPendingDomains)
+		certsFailed = len(certsFailedDomains)
 	}
 
-	data := healthData{
-		Uptime:            time.Since(h.startTime).Round(time.Second).String(),
-		Status:            status,
-		ManagementReady:   ready,
-		AllClientsHealthy: allHealthy,
-		Clients:           clientHealth,
+	resp := map[string]any{
+		"status":               status,
+		"uptime":               uptime,
+		"management_connected": ready,
+		"all_clients_healthy":  allHealthy,
+		"certs_total":          certsTotal,
+		"certs_ready":          certsReady,
+		"certs_pending":        certsPending,
+		"certs_failed":         certsFailed,
+		"clients":              clientHealth,
 	}
-
-	h.renderTemplate(w, "health", data)
+	if len(certsPendingDomains) > 0 {
+		resp["certs_pending_domains"] = certsPendingDomains
+	}
+	if len(certsReadyDomains) > 0 {
+		resp["certs_ready_domains"] = certsReadyDomains
+	}
+	if len(certsFailedDomains) > 0 {
+		resp["certs_failed_domains"] = certsFailedDomains
+	}
+	h.writeJSON(w, resp)
 }
 
 func (h *Handler) renderTemplate(w http.ResponseWriter, name string, data interface{}) {

proxy/internal/debug/templates/health.html

@@ -1,39 +0,0 @@
-{{define "health"}}
-<!DOCTYPE html>
-<html>
-<head>
-  <title>Health</title>
-  <style>{{template "style"}}</style>
-</head>
-<body>
-  <h1>{{.Status}}</h1>
-  <p>Uptime: {{.Uptime}}</p>
-  <p>Management Connected: {{.ManagementReady}}</p>
-  <p>All Clients Healthy: {{.AllClientsHealthy}}</p>
-  {{if .Clients}}
-  <h2>Clients</h2>
-  <table>
-    <tr>
-      <th>Account ID</th>
-      <th>Healthy</th>
-      <th>Management</th>
-      <th>Signal</th>
-      <th>Relays</th>
-      <th>Error</th>
-    </tr>
-    {{range $id, $c := .Clients}}
-    <tr>
-      <td>{{$id}}</td>
-      <td>{{$c.Healthy}}</td>
-      <td>{{$c.ManagementConnected}}</td>
-      <td>{{$c.SignalConnected}}</td>
-      <td>{{$c.RelaysConnected}}/{{$c.RelaysTotal}}</td>
-      <td>{{$c.Error}}</td>
-    </tr>
-    {{end}}
-  </table>
-  {{end}}
-  <p><a href="/debug">&larr; Back</a></p>
-</body>
-</html>
-{{end}}

proxy/internal/debug/templates/index.html

@@ -8,6 +8,25 @@
 <body>
   <h1>NetBird Proxy Debug</h1>
   <p class="info">Version: {{.Version}} | Uptime: {{.Uptime}}</p>
+  <h2>Certificates: {{.CertsReady}} ready, {{.CertsPending}} pending, {{.CertsFailed}} failed ({{.CertsTotal}} total)</h2>
+  {{if .CertsReadyDomains}}
+  <details>
+    <summary>Ready domains ({{.CertsReady}})</summary>
+    <ul>{{range .CertsReadyDomains}}<li>{{.}}</li>{{end}}</ul>
+  </details>
+  {{end}}
+  {{if .CertsPendingDomains}}
+  <details open>
+    <summary>Pending domains ({{.CertsPending}})</summary>
+    <ul>{{range .CertsPendingDomains}}<li>{{.}}</li>{{end}}</ul>
+  </details>
+  {{end}}
+  {{if .CertsFailedDomains}}
+  <details open>
+    <summary>Failed domains ({{.CertsFailed}})</summary>
+    <ul>{{range .CertsFailedDomains}}<li>{{.Domain}}: {{.Error}}</li>{{end}}</ul>
+  </details>
+  {{end}}
   <h2>Clients ({{.ClientCount}}) | Domains ({{.TotalDomains}})</h2>
   {{if .Clients}}
   <table>
@@ -32,7 +51,6 @@
   <h2>Endpoints</h2>
   <ul>
     <li><a href="/debug/clients">/debug/clients</a> - all clients detail</li>
-    <li><a href="/debug/health">/debug/health</a> - health check</li>
   </ul>
   <p class="info">Add ?format=json or /json suffix for JSON output</p>
 </body>

proxy/internal/flock/flock_unix.go

@@ -63,7 +63,11 @@ func Unlock(f *os.File) error {
 		return nil
 	}
 
-	defer f.Close()
+	defer func() {
+		if cerr := f.Close(); cerr != nil {
+			log.Debugf("close lock file: %v", cerr)
+		}
+	}()
 
 	if err := syscall.Flock(int(f.Fd()), syscall.LOCK_UN); err != nil {
 		return fmt.Errorf("release lock: %w", err)

proxy/internal/k8s/lease.go

@@ -65,7 +65,7 @@ type MicroTime struct {
 const microTimeFormat = "2006-01-02T15:04:05.000000Z"
 
 // MarshalJSON implements json.Marshaler with k8s MicroTime format.
-func (t MicroTime) MarshalJSON() ([]byte, error) {
+func (t *MicroTime) MarshalJSON() ([]byte, error) {
 	return json.Marshal(t.UTC().Format(microTimeFormat))
 }
 
@@ -148,7 +148,7 @@ func (c *LeaseClient) Get(ctx context.Context, name string) (*Lease, error) {
 	if err != nil {
 		return nil, err
 	}
-	defer resp.Body.Close()
+	defer func() { _ = resp.Body.Close() }()
 
 	if resp.StatusCode == http.StatusNotFound {
 		return nil, nil
@@ -179,7 +179,7 @@ func (c *LeaseClient) Create(ctx context.Context, lease *Lease) (*Lease, error)
 	if err != nil {
 		return nil, err
 	}
-	defer resp.Body.Close()
+	defer func() { _ = resp.Body.Close() }()
 
 	if resp.StatusCode == http.StatusConflict {
 		return nil, ErrConflict
@@ -208,7 +208,7 @@ func (c *LeaseClient) Update(ctx context.Context, lease *Lease) (*Lease, error)
 	if err != nil {
 		return nil, err
 	}
-	defer resp.Body.Close()
+	defer func() { _ = resp.Body.Close() }()
 
 	if resp.StatusCode == http.StatusConflict {
 		return nil, ErrConflict

proxy/server.go

@@ -247,6 +247,9 @@ func (s *Server) ListenAndServe(ctx context.Context, addr string) (err error) {
 	if s.DebugEndpointEnabled {
 		debugAddr := debugEndpointAddr(s.DebugEndpointAddress)
 		debugHandler := debug.NewHandler(s.netbird, s.healthChecker, s.Logger)
+		if s.acme != nil {
+			debugHandler.SetCertStatus(s.acme)
+		}
 		s.debug = &http.Server{
 			Addr:    debugAddr,
 			Handler: debugHandler,
@@ -359,7 +362,7 @@ func (s *Server) shutdownServices() {
 			defer wg.Done()
 			ctx, cancel := context.WithTimeout(context.Background(), shutdownServiceTimeout)
 			defer cancel()
-			if err := shutdown(ctx); err != nil && !errors.Is(err, http.ErrServerClosed) {
+			if err := shutdown(ctx); err != nil {
 				s.Logger.Debugf("%s shutdown: %v", name, err)
 			}
 		}()
@@ -516,7 +519,7 @@ func (s *Server) addMapping(ctx context.Context, mapping *proto.ProxyMapping) er
 		return fmt.Errorf("create peer for domain %q: %w", d, err)
 	}
 	if s.acme != nil {
-		s.acme.AddDomain(string(d), string(accountID), reverseProxyID)
+		s.acme.AddDomain(d, string(accountID), reverseProxyID)
 	}
 
 	// Pass the mapping through to the update function to avoid duplicating the
@@ -562,7 +565,7 @@ func (s *Server) removeMapping(ctx context.Context, mapping *proto.ProxyMapping)
 		}).Error("Error removing NetBird peer connection for domain, continuing additional domain cleanup but peer connection may still exist")
 	}
 	if s.acme != nil {
-		s.acme.RemoveDomain(mapping.GetDomain())
+		s.acme.RemoveDomain(d)
 	}
 	s.auth.RemoveDomain(mapping.GetDomain())
 	s.proxy.RemoveMapping(s.protoToMapping(mapping))