feat(metrics): add registry_monitor_registries metric

Author: plaffitt

api/kuik/v1alpha1/registrymonitor_types.go

@@ -1,6 +1,8 @@
 package v1alpha1
 
 import (
+	"strings"
+
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -34,6 +36,10 @@ type RegistryMonitorStatus struct {
 	// RegistryStatus is the status of the registry being monitored
 	// +kubebuilder:validation:Enum=Up;Down
 	RegistryStatus RegistryStatus `json:"registryStatus"`
+	// LastMonitor is the last time the registry health was checked
+	LastMonitor metav1.Time `json:"lastMonitor,omitempty"`
+	// LastError is the last error encountered while trying to health check the registry
+	LastError string `json:"lastError,omitempty"`
 }
 
 // +kubebuilder:object:root=true
@@ -66,3 +72,11 @@ type RegistryMonitorList struct {
 func init() {
 	SchemeBuilder.Register(&RegistryMonitor{}, &RegistryMonitorList{})
 }
+
+func (r RegistryStatus) ToString() string {
+	value := string(r)
+	if value == "" {
+		value = "unknown"
+	}
+	return strings.ToLower(value)
+}

api/kuik/v1alpha1/zz_generated.deepcopy.go

@@ -87,6 +87,15 @@ spec:
           status:
             description: RegistryMonitorStatus defines the observed state of RegistryMonitor.
             properties:
+              lastError:
+                description: LastError is the last error encountered while trying
+                  to health check the registry
+                type: string
+              lastMonitor:
+                description: LastMonitor is the last time the registry health was
+                  checked
+                format: date-time
+                type: string
               registryStatus:
                 description: RegistryStatus is the status of the registry being monitored
                 enum:

config/crd/bases/kuik.enix.io_registrymonitors.yaml

@@ -86,6 +86,15 @@ spec:
           status:
             description: RegistryMonitorStatus defines the observed state of RegistryMonitor.
             properties:
+              lastError:
+                description: LastError is the last error encountered while trying
+                  to health check the registry
+                type: string
+              lastMonitor:
+                description: LastMonitor is the last time the registry health was
+                  checked
+                format: date-time
+                type: string
               registryStatus:
                 description: RegistryStatus is the status of the registry being monitored
                 enum:

helm/kube-image-keeper/crds/kuik.enix.io_registrymonitors.yaml

@@ -101,6 +101,31 @@ func (m *kuikMetrics) Register(elected <-chan struct{}, client client.Client) {
 			}
 		}))
 
+	m.addCollector(NewGenericCollector(
+		prometheus.Opts{
+			Namespace: info.MetricsNamespace,
+			Subsystem: subsystemRegistryMonitor,
+			Name:      "registries",
+			Help:      "Number of registries monitored up and running, labeled by registry.",
+		},
+		prometheus.GaugeValue,
+		[]string{"registry"},
+		func(collect func(value float64, labels ...string)) {
+			registryMonitorList := &kuikv1alpha1.RegistryMonitorList{}
+			if err := client.List(context.Background(), registryMonitorList); err != nil {
+				logf.Log.Error(err, "failed to list registry monitors for metrics")
+				return
+			}
+
+			for _, registry := range registryMonitorList.Items {
+				if registry.Status.RegistryStatus == kuikv1alpha1.RegistryStatusUp {
+					collect(1, registry.Name)
+				} else {
+					collect(0, registry.Name)
+				}
+			}
+		}))
+
 	metrics.Registry.MustRegister(m.collectors...)
 }
 

internal/controller/collector.go

@@ -89,10 +89,13 @@ func (r *RegistryMonitorReconciler) Reconcile(ctx context.Context, req ctrl.Requ
 	log.Info("found images matching registry", "count", len(images.Items), "monitoredDuringInterval", monitoredDuringInterval)
 
 	patch := client.MergeFrom(registryMonitor.DeepCopy())
+	registryMonitor.Status.LastMonitor = metav1.Now()
 	registryMonitor.Status.RegistryStatus = kuikv1alpha1.RegistryStatusUp
+	registryMonitor.Status.LastError = ""
 	err := registry.HealthCheck(registryMonitor.Spec.Registry, nil, nil)
 	if err != nil {
 		registryMonitor.Status.RegistryStatus = kuikv1alpha1.RegistryStatusDown
+		registryMonitor.Status.LastError = err.Error()
 	}
 
 	if errStatus := r.Status().Patch(ctx, &registryMonitor, patch); errStatus != nil {