skip gateway API installation in Openshift

[aleskandro]

In Openshift, the Gateway API CRDs are managed by the Ingress Operator, and a manual update as done in this repo is denied. The first commit of this PR skips the installation of the Gateway API CRD when the platform is Openshift.

Moreover, new versions of the Gateway API inference extension define the inference pool as v1 in the group inference.networking.k8s.io. The second commit introduces the WVA Helm Chart's value poolGroup in the related Jinja template, with a default to the group inference.networking.k8s.io.

These changes have been tested with the WVA scenario in Openshift 4.22.2.

[github-actions]

Unsigned commits detected! Please sign your commits.

For instructions on how to set up GPG/SSH signing and verify your commits, please see GitHub Documentation.

[aleskandro]

Hi @asm582, can you give a look at this PR?

[Vezio]

This is a good contribution however we will be getting rid of the WVA installation via helm and moving to the WVA installation via kustomize - @aavarghese is working on this - could you two collaborate on integrating this fix?

Thanks!

[aleskandro]

Hi @Vezio , thanks. I removed the commit about the Helm value and will coordinate with @aavarghese about it in a separate PR if you're ok with it.

[maugustosilva]

Hey @aleskandro thanks for your contribution. My main problem with it, as is, is that we can indeed have neweristio versions even in OpenShift (by indicating that the crds will not be managed by servicemesh). Taking this into account, I believe if context.is_openshift is far too encompassing.

I propose we solve this by reintroducing an "openshift" or "servicemesh" gatewayclass, which will then simply skip the crd installation in case of istio.

[aleskandro]

Hi @maugustosilva, thanks for the review. In my test, it was the ingress cluster operator that refused to install the Gateway API CRDs. Service Mesh or Istio was not installed at that time. Maybe I'm missing something?

[maugustosilva]

Hey @aleskandro, you might be right. I was testing in two different OpenShift clusters, a 4.18 and 4.20 and might have mixed up which components were really blocking the crd installation. Be as it may, simply saying "if context.is_openshift, then skip any crd installation" is just too generic for the needs of users/experiments which are testing/exploring multiple different gateway providers. Hence, the proposal of a new gatewayclass (e.g., "openshift")

[github-actions]

This PR is marked as stale after 21d of inactivity. After an additional 14d of inactivity (7d to become rotten, then 7d more), it will be closed. To prevent this PR from being closed, add a comment or remove the lifecycle/stale label.