My editor of choice is neovim and when editing yaml files, I make use of language servers, specifically yaml-language-server. When working with kubernetes manifests specifically, I have the following config, making use of the versioned kuberentes json schemas available in yannh/kubernetes-json-schema.
1local lspconfig = require("lspconfig")
2lspconfig.yamlls.setup({
3 settings = {
4 yaml = {
5 schemas = {
6 ["file:///home/user/third_party/kubernetes-json-schema/default/v1.27.3-standalone/all.json"] = {"*.k8s.yaml"},
7 kubernetes = "",
8 },
9 },
10 },
11})
But now there's an issue:
that all.json only references the builtin kubernetes types,
meaning it doesn't include any CRDs.
Apparently, VS Code can connect to a running K8s cluster and grab CRDs from there,
mentioned in an issue for content based schema matching
but the rest of us don't have that.
Kubernetes serves openapi v2.0 over on /openapi/v2,
retrievable with kubectl get --raw /openapi/v2 or kubectl proxy --port 8001 & ; curl localhost:8001/openapi/v2.
There's also openapi v3.0 on /openapi/v3,
but v3 is a bunch of linked documents rather than one big one like v2.
I thought I could use the same tooling as kubernetes-json-schema,
specifically openapi2jsonschema
and generate a new all.json refeencing all the types i have in my cluster,
but I think that might be too many types,
and yaml-language-server just ends up confused:
it kept thinking my Gateway should be something ValidatingWebhookConfiguration.
For now, I think just using modelines might be the most reliable way for schema validation.