# I built a Kubernetes operator so you stop copy-pasting imagePullSecrets across namespaces (MIT, GitOps-safe) — retained record

> Retained reddit record from reddit:kubernetes.

Canonical: https://highsignal.app/data/records/d32cf1c0f7f81da4

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:kubernetes)  published 2026-10-21 10:51 ingested 2026-10-22 02:41

[original source ↗](https://www.reddit.com/r/kubernetes/comments/1xbjwur/i_built_a_kubernetes_operator_so_you_stop/)

## retained text



```
Built this out of a recurring annoyance from client work: imagePullSecrets are  
namespaced, so private-registry creds get copy-pasted into every namespace, and  
rotating them means chasing down N copies.

I know there are already tools that replicate secrets across namespaces  
(reflector, kubernetes-replicator, Kyverno generate). My problem with  
"replicate everywhere" is that it pushes credentials into namespaces that never  
pull private images — more blast radius than I want. So this works the other way  
around:

You define the creds once as a cluster-scoped ClusterPullSecret, and a workload  
opts in with an annotation. The operator then creates the pull secret only in  
that workload's namespace, and removes it again (owner refs) when the annotation  
goes away. On-demand instead of gießkanne.

The two parts I spent the most time on:

\- The optional webhook mutates the Pod at admission, not the Deployment/CR. That  
  was deliberate — I didn't want Argo CD/Flux to see a diff and start reconcile-  
  looping. failurePolicy is Ignore, so the operator being down never blocks pod  
  scheduling.  
\- It reads the source Secret uncached and can actively log into the registry to  
  catch expired creds (surfaced via status + events + Prometheus). The uncached  
  read also means it sits nicely behind External Secrets Operator: ESO syncs from  
  Vault/cloud → Secret → ClusterPullSecret, so nothing sensitive lives in Git.

Go + controller-runtime, Helm chart, unit/integration/e2e tests. MIT.

https://github.com/itdpk-de/pullsecret-operator

Would appreciate a sanity check — especially if I've missed a reason the  
reflector/replicator approach is actually better, or edge cases around the webhook.
```



document fetched 2026-10-22 00:40 · canonical https://reddit.com/r/kubernetes/comments/1xbjwur/i_built_a_kubernetes_operator_so_you_stop
