kubectl plugin

[knative]

kubectl plugin

[sixolet]

We talked about this a bit when we were chartering the working group. A kubectl plugin is delivered as a binary in its own right; kn can be both! We're borrowing most of the kubectl libs for output and using the same lib for arg parsing, so it should feel about right.

…

On Jan 19, 2019, at 10:51 AM, Steren ***@***.***> wrote: Hi, I have no real idea what a kubectl plugin is, but I heard of them and want to open the dialog, mostly to hear opinions and capture the rationale. I also learnt about krew, a kubectl plugin manager. Would it make sense to have a kubectl plugin for Knative? If so, should it be in addition to kn? or could kn be delivered as a kubectl plugin? or should it replace kn? Having a plugin that facilitates the usage of Knative resource could be very Kubernetes... native :) CC @sixolet — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[ahmetb]

Unless you plan on exposing kn as kubectl serverless [...] or something like that, you probably shouldn't use plugins.

That said, using https://github.com/kubernetes/cli-runtime helps you reuse the flag parsing, printing etc functionality from kubectl.

I'm happy to discuss kubectl plugins anytime. But unless the client would be advertised as a kubectl subcommand, it is probably not the best fit.

[steren]

Thanks, that makes sense.

CC @ahmetb

[steren]

I actually opened this issue to discuss this point: do we need a standalone kn client? or would a kubectl plugin be more idiomatic to our user base?

[maximilien]

I believe we need a standalone kn client. And having plugins to extend / experiment on the client is also useful.

We had some discussion on this on 04/16/2019. I have some opinions and would like to restart the discussion if others are interested.

[maximilien]

Since this issue is discussing kn as a kubectl plugin, which is a valid issue for discussion, I am opening a different one on adding plugins to kn itself, see issue #92.

I will present and demo a proposal during Tuesday 04/06/2019's call.

[rhuss]

I actually opened this issue to discuss this point: do we need a standalone kn client? or would a kubectl plugin be more idiomatic to our user base?

I think one of the main motivation of the kn client is to be completely agnostic to Kubernetes and also be able to support Knative implementations which are not based on Kubernetes.

Having plugins for kn though is a different story.

[steren]

Seems like the consensus is to not ship via a kubectl plugin but via a standalone binary. Closing.