Add `kubectl config merge`

[segevfiner]

What would you like to be added

Add a subcommand to merge in a kubeconfig file into your default kubeconfig file.

Why is this needed

Many tools already do this anyhow, gcloud container clusters get-credentials, minikube, k3d and so on. It's basically like using multiple config files and outputting a merged version using kubectl config view --flatten, so the logic is essentially already there. So all that remains is exposing it as a conveinent subcommand.

[kubernetes-prow]

This issue is currently awaiting triage.

SIG CLI takes a lead on issue triage for this repo, but any Kubernetes member can accept issues by applying the triage/accepted label.

The triage/accepted label can be added by org members by writing /triage accepted in a comment.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[ardaguclu]

I think, this would be a good plugin use case, instead of directly adding it in kubectl.
@oscrx mentioned that there is already a plugin for it.

[oscrx]

I've used this one in the past
https://github.com/corneliusweig/konfig
(Installable via krew)

[segevfiner]

Which then duplicates this logic from kubectl, and requires a plugin for a rather basic task IMHO, that other tools just end up each implementing again on their own... Possibly with differing results... Tradeoffs tradeoffs...

[oscrx]

Spent some time actually prototyping this to see how bad it'd be: kubernetes/kubernetes@master...oscrx:kubernetes:kubectl-config-merge

Not bad at all honestly, one file, maybe 150 lines. Mostly just calling GetStartingConfig / LoadFromFile / FlattenConfig / ModifyConfig, same stuff view, use-context and set-context already do under the hood. Merge is basically "load both configs, overwrite on name collisions, write it back with the existing modify-config machinery." So @segevfiner fair point, this isn't really new plumbing, kubectl already has all the pieces, nobody's built the one command that chains them together.

But while looking around for existing solutions I ran into konfuse and that kind of talked me out of it. It renames stuff on import and fixes up the internal references for you (so the context still points at the right cluster/user after you rename them), takes a timestamped backup before it touches your config, cleans up orphaned clusters/users when you delete a context, and has a proper --json mode so you can actually script against it. None of that's in my prototype, my version just does the dumb merge and nothing else. And that's honestly the part that matters if you're trusting a tool with your actual kubeconfig — not whether it can merge two files, but whether it does something sane when names collide or a write fails halfway through.

So yeah, cheap to build != should build. Building the bare merge is easy, but the moment you try to make it actually safe and pleasant to use day to day you're basically rebuilding konfuse inside kubectl, badly, and then stuck maintaining it forever. konfuse (or konfig, mentioned earlier in this thread) covers this better than kubectl core probably ever would, and can keep iterating on it way faster than a KEP cycle allows. Not my call either way, just wanted to share what I found while digging into this.