Handling NATS clusters in multiple namespaces

[lwierzbi-work]

Hi, I have k8s cluster with two NATS clusters in two different namespaces (dev and stg environment), installed NACK in dev with helm chart, on install in stg I have bumped into resource ownership confilict:

helm install nack-jsc nats/nack --set jetstream.nats.url=nats://nats:4222 -n nats-io-${TIER} 

Error: INSTALLATION FAILED: rendered manifests contain a resource that already exists. Unable to continue with install: ClusterRole "jetstream-controller-cluster-role" in namespace "" exists and cannot be imported into the current release: invalid ownership metadata; annotation validation error: key "meta.helm.sh/release-namespace" must equal "nats-io-stg": current value is "nats-io-dev"

Since NATS URL is one of NACK install parameters, I assume that one instance can controll only one NATS cluster. Is there a way to have several deployments of NACK, each controlling configuration of different NATS cluster within single k8s cluster?

[TarasLykhenko]

I am also hitting this issue. Is it possible to implement a similar solution flux that uses labels to partition the CRs over multiple instances of the same controller?

https://fluxcd.io/flux/installation/configuration/sharding/#assign-resources-to-shards

[arguile-]

I tried using Accounts -- which specify the server -- in order to do this but it seems like its ignored and uses the underlying NACK config.

Eg.

---
apiVersion: jetstream.nats.io/v1beta2
kind: Account
metadata:
  name: a
spec:
  name: a
  servers:
  - nats://nats.namespace_a.svc.cluster.local:4222

---
apiVersion: jetstream.nats.io/v1beta2
kind: Stream
metadata:
  name: foo
spec:
  name: foo
  subjects: ["foo", "foo.>"]
  storage: file
  replicas: 1
  account: a # <-- Create stream using account A information

I would have expected the account information to be respected and used in creating the stream.

[arguile-]

For anyone struggling we this, our end work-around was to change to namespaced: true and then centrally manage the stream CRDs for each cluster within the controller namespace.

This doesn't allow adding the CRDs between various project namespaces throughout the cluster though so may not work for everyone.

[nico151999]

There is a flag called crdConnect. If the nats URL is omitted the chart currently just enables this flag instead of using the one nats URL globally. If I'm not mistaken the intention of the controller developers is to either have one global config via values.yaml or n configs via accounts. I didn't find it documented anywhere though. That's just what I found by having a quick look at the code. Maybe this behaviour will change. There are multiple things that seem not to be documented (or I couldn't find it) which I wonder what they are used for, e.g. the jetstream.enabled value in the nack chart.