Added metrics to count the number of rollout restarts

Fixes #1129

Author: mnacharov

helpers/rollout_restart.go

@@ -19,6 +19,7 @@ import (
 
 	"github.com/hashicorp/vault-secrets-operator/api/v1beta1"
 	"github.com/hashicorp/vault-secrets-operator/consts"
+	"github.com/hashicorp/vault-secrets-operator/internal/metrics"
 )
 
 // AnnotationRestartedAt is updated to trigger a rollout-restart
@@ -62,9 +63,11 @@ func HandleRolloutRestarts(ctx context.Context, client ctrlclient.Client, obj ct
 			errs = errors.Join(err)
 			recorder.Eventf(obj, corev1.EventTypeWarning, consts.ReasonRolloutRestartFailed,
 				"Rollout restart failed for target %#v: err=%s", target, err)
+			metrics.RolloutRestartsErrors.WithLabelValues(obj.GetNamespace(), obj.GetName(), target.Name, target.Kind).Inc()
 		} else {
 			recorder.Eventf(obj, corev1.EventTypeNormal, consts.ReasonRolloutRestartTriggered,
 				"Rollout restart triggered for %v", target)
+			metrics.RolloutRestartsTotal.WithLabelValues(obj.GetNamespace(), obj.GetName(), target.Name, target.Kind).Inc()
 		}
 	}
 

internal/metrics/metrics.go

@@ -39,6 +39,7 @@ const (
 	NameRequestsTotal         = "requests_total"
 	NameRequestsErrorsTotal   = "requests_errors_total"
 	NameTaintedClients        = "tainted_clients"
+	subsystemRolloutRestarts  = "rollout_restarts"
 )
 
 var ResourceStatus = prometheus.NewGaugeVec(prometheus.GaugeOpts{
@@ -50,6 +51,18 @@ var ResourceStatus = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 	"namespace",
 })
 
+var RolloutRestartsTotal = prometheus.NewCounterVec(prometheus.CounterOpts{
+	Name: prometheus.BuildFQName(
+		Namespace, subsystemRolloutRestarts, "total"),
+	Help: "Number of total rollout restarts.",
+}, []string{"namespace", "name", "target_name", "target_kind"})
+
+var RolloutRestartsErrors = prometheus.NewCounterVec(prometheus.CounterOpts{
+	Name: prometheus.BuildFQName(
+		Namespace, subsystemRolloutRestarts, "errors_total"),
+	Help: "Number of total rollout restarts.",
+}, []string{"namespace", "name", "target_name", "target_kind"})
+
 func MustRegisterResourceStatus() {
 	metrics.Registry.MustRegister(
 		ResourceStatus,