breaking change in gateway CRDs between latest tagged and master

[lizbound]

While I was troubleshooting #373, I noticed another error repeating in the gateway CRD manager logs:

Feb 20 10:22:25 jinu cloud-provider-kind[517221]: E0220 10:22:25.587181  517221 reflector.go:204] "Failed to watch" err="failed to list *v1.ReferenceGrant: the server could not find the requested resource (get referencegrants.gateway.networking.k8s.io)" logger="UnhandledError" reflector="pkg/mod/k8s.io/client-go@v0.35.1/tools/cache/reflector.go:289" type="*v1.ReferenceGrant"

06427d4 updated the versions of the gateway CRDs, and that introduced a v1 ReferenceGrant, where previously there was only v1alpha1. The issue is that the mechanism in crd_manager is not sophisticated enough to "upgrade" this CRD:

Feb 19 16:31:13 jinu cloud-provider-kind[515405]: I0219 16:31:13.883316  515405 crd_manager.go:122] CRD "referencegrants.gateway.networking.k8s.io" already exists, skipping creation.

Thus, the only way to use this new version of cloud-provider-kind is to delete all the existing CRDs:

kubectl --context kind-kind get crd -oname | grep -E "gateway.networking.(x-)?k8s.io" | xargs kubectl --context kind-kind delete

Then they'll be recreated.

Kind clusters are often short-lived, so this might be a WONTFIX, but it certainly needs to go in the changelog and may necessitate a version bump.

[aojea]

I will happy to accept a fix for that, as you see, this implementation for CRD management is just "enough to get things going", and it needs some love