kubelet: add certificate rotation error metric

pkg/kubelet/certificate/kubelet.go

@@ -61,6 +61,15 @@ func NewKubeletServerCertificateManager(kubeClient clientset.Interface, kubeCfg
 		},
 	)
 	legacyregistry.MustRegister(certificateExpiration)
+	var certificateRenewFailure = compbasemetrics.NewCounter(
+		&compbasemetrics.CounterOpts{
+			Subsystem:      metrics.KubeletSubsystem,
+			Name:           "server_expiration_renew_errors",
+			Help:           "Counter of certificate renewal errors.",
+			StabilityLevel: compbasemetrics.ALPHA,
+		},
+	)
+	legacyregistry.MustRegister(certificateRenewFailure)
 
 	certificateRotationAge := compbasemetrics.NewHistogram(
 		&compbasemetrics.HistogramOpts{
@@ -119,9 +128,10 @@ func NewKubeletServerCertificateManager(kubeClient clientset.Interface, kubeCfg
 			// authenticate itself to a TLS client.
 			certificates.UsageServerAuth,
 		},
-		CertificateStore:      certificateStore,
+		CertificateStore:        certificateStore,
-		CertificateExpiration: certificateExpiration,
+		CertificateExpiration:   certificateExpiration,
-		CertificateRotation:   certificateRotationAge,
+		CertificateRotation:     certificateRotationAge,
+		CertificateRenewFailure: certificateRenewFailure,
 	})
 	if err != nil {
 		return nil, fmt.Errorf("failed to initialize server certificate manager: %v", err)
@@ -199,6 +209,16 @@ func NewKubeletClientCertificateManager(
 		},
 	)
 	legacyregistry.Register(certificateExpiration)
+	var certificateRenewFailure = compbasemetrics.NewCounter(
+		&compbasemetrics.CounterOpts{
+			Namespace:      metrics.KubeletSubsystem,
+			Subsystem:      "certificate_manager",
+			Name:           "client_expiration_renew_errors",
+			Help:           "Counter of certificate renewal errors.",
+			StabilityLevel: compbasemetrics.ALPHA,
+		},
+	)
+	legacyregistry.Register(certificateRenewFailure)
 
 	m, err := certificate.NewManager(&certificate.Config{
 		ClientFn: clientFn,
@@ -231,8 +251,9 @@ func NewKubeletClientCertificateManager(
 		BootstrapCertificatePEM: bootstrapCertData,
 		BootstrapKeyPEM:         bootstrapKeyData,
 
-		CertificateStore:      certificateStore,
+		CertificateStore:        certificateStore,
-		CertificateExpiration: certificateExpiration,
+		CertificateExpiration:   certificateExpiration,
+		CertificateRenewFailure: certificateRenewFailure,
 	})
 	if err != nil {
 		return nil, fmt.Errorf("failed to initialize client certificate manager: %v", err)

staging/src/k8s.io/client-go/util/certificate/certificate_manager.go

@@ -124,6 +124,9 @@ type Config struct {
 	// allows one to setup monitoring and alerting of unexpected rotation
 	// behavior and track trends in rotation frequency.
 	CertificateRotation Histogram
+	// CertifcateRenewFailure will record a metric that keeps track of
+	// certificate renewal failures.
+	CertificateRenewFailure Counter
 }
 
 // Store is responsible for getting and updating the current certificate.
@@ -154,6 +157,11 @@ type Histogram interface {
 	Observe(float64)
 }
 
+// Counter will wrap a counter with labels
+type Counter interface {
+	Inc()
+}
+
 // NoCertKeyError indicates there is no cert/key currently available.
 type NoCertKeyError string
 
@@ -177,8 +185,9 @@ type manager struct {
 
 	certStore Store
 
-	certificateExpiration Gauge
+	certificateExpiration   Gauge
-	certificateRotation   Histogram
+	certificateRotation     Histogram
+	certificateRenewFailure Counter
 
 	// the following variables must only be accessed under certAccessLock
 	certAccessLock sync.RWMutex
@@ -213,17 +222,18 @@ func NewManager(config *Config) (Manager, error) {
 	}
 
 	m := manager{
-		stopCh:                make(chan struct{}),
+		stopCh:                  make(chan struct{}),
-		clientFn:              config.ClientFn,
+		clientFn:                config.ClientFn,
-		getTemplate:           getTemplate,
+		getTemplate:             getTemplate,
-		dynamicTemplate:       config.GetTemplate != nil,
+		dynamicTemplate:         config.GetTemplate != nil,
-		usages:                config.Usages,
+		usages:                  config.Usages,
-		certStore:             config.CertificateStore,
+		certStore:               config.CertificateStore,
-		cert:                  cert,
+		cert:                    cert,
-		forceRotation:         forceRotation,
+		forceRotation:           forceRotation,
-		certificateExpiration: config.CertificateExpiration,
+		certificateExpiration:   config.CertificateExpiration,
-		certificateRotation:   config.CertificateRotation,
+		certificateRotation:     config.CertificateRotation,
-		now:                   time.Now,
+		certificateRenewFailure: config.CertificateRenewFailure,
+		now:                     time.Now,
 	}
 
 	return &m, nil
@@ -404,6 +414,9 @@ func (m *manager) rotateCerts() (bool, error) {
 	template, csrPEM, keyPEM, privateKey, err := m.generateCSR()
 	if err != nil {
 		utilruntime.HandleError(fmt.Errorf("Unable to generate a certificate signing request: %v", err))
+		if m.certificateRenewFailure != nil {
+			m.certificateRenewFailure.Inc()
+		}
 		return false, nil
 	}
 
@@ -411,6 +424,9 @@ func (m *manager) rotateCerts() (bool, error) {
 	client, err := m.getClient()
 	if err != nil {
 		utilruntime.HandleError(fmt.Errorf("Unable to load a client to request certificates: %v", err))
+		if m.certificateRenewFailure != nil {
+			m.certificateRenewFailure.Inc()
+		}
 		return false, nil
 	}
 
@@ -419,6 +435,9 @@ func (m *manager) rotateCerts() (bool, error) {
 	req, err := csr.RequestCertificate(client, csrPEM, "", m.usages, privateKey)
 	if err != nil {
 		utilruntime.HandleError(fmt.Errorf("Failed while requesting a signed certificate from the master: %v", err))
+		if m.certificateRenewFailure != nil {
+			m.certificateRenewFailure.Inc()
+		}
 		return false, m.updateServerError(err)
 	}
 
@@ -433,12 +452,18 @@ func (m *manager) rotateCerts() (bool, error) {
 	crtPEM, err := csr.WaitForCertificate(ctx, client, req)
 	if err != nil {
 		utilruntime.HandleError(fmt.Errorf("certificate request was not signed: %v", err))
+		if m.certificateRenewFailure != nil {
+			m.certificateRenewFailure.Inc()
+		}
 		return false, nil
 	}
 
 	cert, err := m.certStore.Update(crtPEM, keyPEM)
 	if err != nil {
 		utilruntime.HandleError(fmt.Errorf("Unable to store the new cert/key pair: %v", err))
+		if m.certificateRenewFailure != nil {
+			m.certificateRenewFailure.Inc()
+		}
 		return false, nil
 	}