Bug: Operator constantly reconciling

[michizubi-SRF]

Describe the bug

The operator is constantly reconciling, around 4-5 times per second, even though there are no changes.
The following is logged:

{"level":"info","ts":"2026-05-07T13:42:34Z","msg":"Start reconciling","controller":"rabbitmqcluster","controllerGroup":"rabbitmq.com","controllerKind":"RabbitmqCluster","RabbitmqCluster":{"name":"redacted","namespace":"redacted"},"namespace":"redacted","name":"redacted","reconcileID":"051fe833-3b4d-4562-964a-b9b9e4c1e603"}
{"level":"info","ts":"2026-05-07T13:42:34Z","msg":"updated resource redacted-nodes of Type *v1.Service","controller":"rabbitmqcluster","controllerGroup":"rabbitmq.com","controllerKind":"RabbitmqCluster","RabbitmqCluster":{"name":"redacted","namespace":"redacted"},"namespace":"redacted","name":"redacted","reconcileID":"051fe833-3b4d-4562-964a-b9b9e4c1e603"}
{"level":"info","ts":"2026-05-07T13:42:34Z","msg":"updated resource redacted of Type *v1.Service","controller":"rabbitmqcluster","controllerGroup":"rabbitmq.com","controllerKind":"RabbitmqCluster","RabbitmqCluster":{"name":"redacted","namespace":"redacted"},"namespace":"redacted","name":"redacted","reconcileID":"051fe833-3b4d-4562-964a-b9b9e4c1e603"}
{"level":"info","ts":"2026-05-07T13:42:34Z","logger":"statefulset.RabbitmqCluster","msg":"Warning: Memory request and limit are not equal for \"redacted-server\". It is recommended that they be set to the same value"}
{"level":"info","ts":"2026-05-07T13:42:34Z","logger":"statefulset.RabbitmqCluster","msg":"Warning: Memory request and limit are not equal for \"redacted-server\". It is recommended that they be set to the same value"}
{"level":"info","ts":"2026-05-07T13:42:34Z","msg":"updated resource redacted-server of Type *v1.StatefulSet","controller":"rabbitmqcluster","controllerGroup":"rabbitmq.com","controllerKind":"RabbitmqCluster","RabbitmqCluster":{"name":"redacted","namespace":"redacted"},"namespace":"redacted","name":"redacted","reconcileID":"051fe833-3b4d-4562-964a-b9b9e4c1e603"}
{"level":"info","ts":"2026-05-07T13:42:34Z","msg":"Finished reconciling","controller":"rabbitmqcluster","controllerGroup":"rabbitmq.com","controllerKind":"RabbitmqCluster","RabbitmqCluster":{"name":"redacted","namespace":"redacted"},"namespace":"redacted","name":"redacted","reconcileID":"051fe833-3b4d-4562-964a-b9b9e4c1e603"}

To Reproduce

Steps to reproduce the behavior:

1. Deploy the latest rabbitmq operator on OpenShift ROSA with ArgoCD and Helm
2. Create a cluster with the following manifest

{{- if .Values.rabbitmq.enabled }}
apiVersion: rabbitmq.com/v1beta1
kind: RabbitmqCluster
metadata:
  name: {{ include "redacted.fullname" . }}-rabbitmq
  namespace: {{ .Release.Namespace }}
spec:
  replicas: {{ .Values.rabbitmq.replicas }}
  {{- if .Values.rabbitmq.resources }}
  resources:
    {{- toYaml .Values.rabbitmq.resources | nindent 4 }}
  {{- end }}
  {{- if .Values.rabbitmq.persistence }}
  persistence:
    storageClassName: {{ .Values.rabbitmq.persistence.storageClass }}
    storage: {{ .Values.rabbitmq.persistence.size | default "10Gi" }}
  {{- end }}
  override:
    statefulSet:
      spec:
        template:
          spec:
            {{- if .Values.global.imagePullSecrets }}
            imagePullSecrets:
              {{- range .Values.global.imagePullSecrets }}
              - name: {{ . }}
              {{- end }}
            {{- end }}
            securityContext: {}
            containers:
            - name: rabbitmq
              securityContext: {}
            initContainers:
            - name: setup-container
              securityContext: {}

3. See logs of operator pod

Expected behavior
Reconciliation happens a lot less or only when resources actually change.

Version and environment information

• RabbitMQ: 4.1.3
• RabbitMQ Cluster Operator: v2.20.1
• Kubernetes: OpenShift 4.20 - K8s v1.33.9
• Cloud provider or hardware configuration: OpenShift on AWS (ROSA)

[Zerpet]

I can't reproduce this locally without ArgoCD or Helm. The most likely explanation I can provide is that either ArgoCD or Helm (most likely ArgoCD) is updating some resources created by Operator, and that's causing a reconcile, because the Operator watches and owns all resources it creates to support rabbit i.e. StatefulSet, Service, Secret, etc.

Alternatively, if ArgoCD is deploying the Helm chart, which deploys a RabbitmqCluster, it could happen that Cluster Operator updates some fields in the Spec, to set some defaults. This is not great, and we have #2121 open to address that. In addition, there's #2098 open, from which we have concluded that the Operator should not be using Update() on the Spec.

[Zerpet]

I'm also curious about using imagePullSecrets in the override. Have you tried to set the image pull secrets in the Spec directly?

https://www.rabbitmq.com/kubernetes/operator/using-operator#image-pull-secrets

[michizubi-SRF]

Thanks for the hints. Setting imagePullSecrets within spec didn't help, but I found an unrelated(?) setting of a specific label by ArgoCD created this reconciliation loop. It stopped as soon as I set this label in the RabbitmqCluster manifest. It appears that the operator tries to remove this label while ArgoCD wants to apply it.