i32 for port number

[sombralibre]

i32 for port number

[Arnavion]

Is there something that I missed?

Yes, you missed that this code is auto-generated from the OpenAPI spec. It's not like I made the conscious decision to represent port numbers as i32. :)

So, to that end, the OpenAPI spec says:

    "io.k8s.api.networking.v1.ServiceBackendPort": {
      "description": "ServiceBackendPort is the service port being referenced.",
      "properties": {
        "name": {
          "description": "name is the name of the port on the Service. This is a mutually exclusive setting with \"Number\".",
          "type": "string"
        },
        "number": {
          "description": "number is the numerical port number (e.g. 80) on the Service. This is a mutually exclusive setting with \"Name\".",
          "format": "int32",
          "type": "integer"
        }
      },
      "type": "object"
    }

"format": "int32"

At this point you might say it's a spec bug and would be resolved by specifying a better format. But even the API server code says:

https://github.com/kubernetes/kubernetes/blob/9b09d0600a69a2eb36b0d136465ccc3c179dacdb/pkg/apis/networking/types.go#L624-L636

Number int32

🤷 Complain to upstream.

[sombralibre]

Yes, you're right, just after create this issue I went to the OpenApi specs and found the same.

I will scale the question to the openapi people.

Thanks a lot man.

[Arnavion]

Well, OpenAPI supports specifying integer ranges using minimum and maximum, but the Kubernetes OpenAPI spec is downstream of their golang code so they'd have to invent new annotations to express that. Besides, ServiceBackendPort::number is far from the only field that has this problem; searching for : Option<i32>, in this repo reveals all sorts of time durations and replica counts and percentages, and upstream would have to fix all of them for completeness.

[Arnavion]

Upstream issue: kubernetes/kubernetes#105533

[sombralibre]

Upstream issue: kubernetes/kubernetes#105533

👍🏻