fix: Improve change detection based on generation

internal/pkg/kubernetes/client.go

@@ -61,6 +61,9 @@ func (c *Client) WatchDeploymentEvents(ctx context.Context, namespace string, no
 				if !ok {
 					continue
 				}
+				if deployment.Generation == deployment.Status.ObservedGeneration {
+					continue
+				}
 				if ev := processObject(e, deployment.ObjectMeta, deployment.Spec.Template); ev != nil {
 					notifyFunc(ev)
 				}
@@ -87,11 +90,14 @@ func (c *Client) WatchStatefulsetEvents(ctx context.Context, namespace string, n
 				if e.Object == nil {
 					return
 				}
-				deployment, ok := e.Object.(*v1apps.StatefulSet)
+				statefulset, ok := e.Object.(*v1apps.StatefulSet)
 				if !ok {
 					continue
 				}
-				if ev := processObject(e, deployment.ObjectMeta, deployment.Spec.Template); ev != nil {
+				if statefulset.Generation == statefulset.Status.ObservedGeneration {
+					continue
+				}
+				if ev := processObject(e, statefulset.ObjectMeta, statefulset.Spec.Template); ev != nil {
 					notifyFunc(ev)
 				}
 			case <-ctx.Done():
@@ -104,20 +110,7 @@ func (c *Client) WatchStatefulsetEvents(ctx context.Context, namespace string, n
 	return nil
 }
 
-var cache = map[string]time.Time{}
-
 func processObject(e watch.Event, objectMeta metav1.ObjectMeta, podTemplate v1.PodTemplateSpec) []Event {
-	val, ok := cache[string(objectMeta.UID)]
-	if ok && val.After(time.Now().Add(-10*time.Second)) {
-		// Skip processing if the object was processed in the last 10 seconds.
-		logrus.WithFields(logrus.Fields{
-			"namespace": objectMeta.Namespace,
-			"name":      objectMeta.Name,
-		}).Trace("Skipping, recently processed")
-		return nil
-	}
-	cache[string(objectMeta.UID)] = time.Now()
-
 	switch e.Type {
 	case watch.Added:
 		if objectMeta.CreationTimestamp.Time.Before(time.Now().Add(-1 * time.Minute)) {