make undeploy deleted EVERY POD IN THE NAMESPACE

[utdrmac]

Holy hell! Thank goodness this was a test K8S, but still, why in <insert holy figure>'s name would you have a command that deletes all the pods!? And no warnings?!

Steps:

• Change the namespace using kustomize to mine (this is a shared K8S, and we each have our own namespace).

cd config/default
kustomize edit set namespace myuserns

• Run make deploy IMG=my.private.reg/valkey-operator:tag. Seemed to work. Let's remove it and try again.
• Run make undeploy (per the documentation) and suddenly all 33 of my pods are in terminating state! one-by-one I see all my pods die off.

The documentation literally says "UnDeploy the controller from the cluster:" It does not say "Remove all pods in namespace". This command should only remove the controller, as stated.

[sandeepkunusoth]

i think this is issue in kubebuilder generated scaffolded projects. kubernetes-sigs/kubebuilder#5355

best way to fix this would be to fix it in upstream kubebuilder project and upgrade kubebuilder version in this project https://book.kubebuilder.io/plugins/available/autoupdate-v1-alpha

[jdheyburn]

Thanks for raising @utdrmac. The specified command removed all the pods in the namespace, not just the ones managed by Valkey?

[utdrmac]

Yes, it's because this file, https://github.com/valkey-io/valkey-operator/blob/main/config/manager/manager.yaml, contains a namespace CR thus when you pass this through kubectl delete (which is what make undeploy does), then K8S cascade-deletes the entire namespace. And since I had used kustomize to change the namespace to my own, everything was removed.

In my local branch, I have removed the Namespace stanza from manager.yaml, and the issue no longer happens.

[jdheyburn]

Okay got it - we should then have a mechanism to remove the operator but keep the existing resources in place.

It appears then make undeploy should only be used to wipe everything that was previously deployed from scratch - we should improve that documentation too.