search

microsoft / retina

eBPF distributed networking observability tool for Kubernetes
https://retina.sh
MIT License
2.52k stars 172 forks source link

Operator deployment in different namespace than kube-system. #492

Open rbtr opened 2 weeks ago

rbtr commented 2 weeks ago

Discussed in https://github.com/microsoft/retina/discussions/487

Originally posted by **illarios** June 19, 2024 As retina is able to be deployed in other namespaces and I can see that the retina-operator-config gets its namespace from the values file: https://github.com/microsoft/retina/blob/316ad80b4829a464ed60daa50459df4ec27aab92/deploy/manifests/controller/helm/retina/templates/operator.yaml#L258C1-L263C37 I would suggest the same to happen with the operator itself, as the case now is that if you choose another namespace for deployment of retina the operator cannot start without its configuration which apparently exists in the different namespace specified in the values file. You can see here that the operator's deployment namespace is not configurable: https://github.com/microsoft/retina/blob/316ad80b4829a464ed60daa50459df4ec27aab92/deploy/manifests/controller/helm/retina/templates/operator.yaml#L6

IMO this is actually a bug - some of the resources in the Helm chart respect the namespace set in the Values, but some have it hardcoded to kube-system, breaking the deployment.