From 62d048d2158fc55413a3682ee7f9d3f04837fbe8 Mon Sep 17 00:00:00 2001
From: Mikhail Nacharov <1246847+mnacharov@users.noreply.github.com>
Date: Tue, 30 Sep 2025 18:35:14 +0200
Subject: [PATCH] Added metrics to count the number of rollout restarts

Fixes #1129
---
 helpers/rollout_restart.go  |  3 +++
 internal/metrics/metrics.go | 20 ++++++++++++++++++++
 main.go                     |  1 +
 3 files changed, 24 insertions(+)

diff --git a/helpers/rollout_restart.go b/helpers/rollout_restart.go
index 45cf4e183..f33cdc3a9 100644
--- a/helpers/rollout_restart.go
+++ b/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()
 		}
 	}
 
diff --git a/internal/metrics/metrics.go b/internal/metrics/metrics.go
index 683fadd57..e8e7b1458 100644
--- a/internal/metrics/metrics.go
+++ b/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{
@@ -56,6 +57,25 @@ func MustRegisterResourceStatus() {
 	)
 }
 
+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 MustRegisterRolloutRestarts() {
+	metrics.Registry.MustRegister(
+		RolloutRestartsTotal,
+		RolloutRestartsErrors,
+	)
+}
+
 // SetResourceStatus for the given client.Object. If valid is true, then the
 // ResourceStatus gauge will be set 1, else 0.
 func SetResourceStatus(controller string, o client.Object, valid bool) {
diff --git a/main.go b/main.go
index df1d28a64..27c0e01c6 100644
--- a/main.go
+++ b/main.go
@@ -403,6 +403,7 @@ func main() {
 		)
 		vclient.MustRegisterClientMetrics(cfc.MetricsRegistry)
 		metrics.MustRegisterResourceStatus()
+		metrics.MustRegisterRolloutRestarts()
 
 		metric := prometheus.NewGauge(
 			prometheus.GaugeOpts{