K3s setup

[rajibhossen]

No description provided.

[vsoch]

Nice job @rajibhossen! 🥳 Two quick questions and a suggested change:

• Is there a reason you want k3s to come up with the entire cluster? I think the use case we were thinking about (at least for the lab) was to have it launched with a job, meaning running in the user's namespace, and it's not running alongside flux but within it (and then multiple users could each control their own cluster). But if you are just doing some testing of how it scales, etc. this should probably be OK.
• Related to the above - do you want to test rootless mode (or not relevant yet?)
• And one suggestion / nit is that "flux-selector" should be a variable that matches the name, see here and here for an example of doing that.

[rajibhossen]

Thank you!

1. Yes, I'll integrate/work with that later. I wanted to see k3s deployment using the TF and scripts.
2. I'm not sure about that yet!
3. That's a mistake. I was using k3s_setup.sh script from inside of an instance to test. I'll remove that.

[vsoch]

Gotcha! Okay one more request - sorry just noticed the structure. I think Abhik had mentioned this, but please make a separate k3s folder within experiments and alongside the autoscale folder that looks like this:

experiments/
   autoscale/
   k3s

And don't do any modifications to the autoscale setup. The reason is because this repository is not a single hard coded research project, but a set of separate, modular experiments (each scoped to a particular purpose) and so the one you are working on (and then will have full ownership over!) is this k3s experiment! And someone can come later that just wants Flux with autoscale, and they still are able to do that without needing to untangle the k3s. Does that make sense?

[rajibhossen]

I've decoupled the examples.

[vsoch]

Perfecto! One quick question, and then we are ready to merrrrge away Merrill! ⚾