Kube-rs: Drop implicit kind->resource conversion

[teozkr]

Motivated by #14, but the current behaviour mostly works by accident anyway.

• [ ] Add RESOURCE_NAME const to k8s_openapi::Resource (and add it to their generated object types)
• [ ] Replace kube::Resource::kind with resource_name
• [ ] Generate RESOURCE_NAME from the CustomResource deriver
• [ ] Drop inflector runtime dependency

[clux]

Totally! If it's possible to grab the actual values without relying on pluralization then that would be great. I just assumed the information was not available in the openapi spec by looking at the underlying types; e.g. Pod and introduced the dep in a bit of a hurry.

Wonder how k8s-openapi does this.

[teozkr]

You can get the mapping at runtime from the APIResourceList endpoints, which seems to be what kubectl is doing.

Of course, that isn't great for our use-case (wouldn't want to need a running K8s cluster to be able to compile k8s_openapi.. :P). Haven't looked through the OpenAPI stuff yet to see whether anything similar is available there.

[teozkr]

Looks like the OpenAPI does not include the resource name. On the other hand, I don't think it would be too bad to use a heurestic like "find the last path item of the list endpoint`.

[clux]

Found another case of our implicit naming not working today when doing #305. CustomResource assumes PascalCase + singular kinds, but even kubernetes fails to adhere to this: APIService, CSINode, NodeMetrics, Endpoints (as well as numerous CRDs that do not have enforcement (details).

Ended up removing case/plural pedantry on the dynamic api to a mere debug check. Which is fine, dynamically created resources work if we aren't being strict. But THIS issue means that some of the objects noted in that PR might not work with Api. Thankfully, only found Endpoints (badly pluralised) and APIResource/APIService (bad pascal) as NATIVE failure cases when listing all available resource types (though all of istio's crds were also in violation..) with the new dynamic_api example.

Not even sure excess pluralisation in Endpoints is an actual failure case either. to_plural should be idempotent.

Can at least confirm that the required data is there on the api for all cases (albeit awkwardly split between two api calls). Hopefully there's a way we can get the data in the trait in k8s-openapi without using any api though. It's not like it's generated from a live cluster, and Arnav is creating the urls with plural somehow, the data must be there.

[clux]

There's no easy way of getting the plural as it stands. It happens to be in the urls of the schemas as noted in https://github.com/Arnavion/k8s-openapi/issues/73#issuecomment-671570768. If the k8s-openapi issue resolves, we can progress here, otherwise marking as awaiting upstream work.

[clux]

This is less of an issue now that we maintain a pluralisation fn that handles the few non-conformant cases.

All other entry ways no longer need to do this implicitly.

• api-discovery will automatically deduce the plural
• kube-derive optionally allow #[kube(plural = "fooz")]

[MikailBag]

I've sent https://github.com/Arnavion/k8s-openapi/pull/88 which adds resource name to the k8s-openapi::Resource trait