Implement cordon/uncordon for Node

[ChinYing-Li]

Signed-off-by: ChinYing-Li chinying.li@mail.utoronto.ca

Motivation

This PR aims to solve #706.

Solution

Currently only cordon/uncordon is implemented, as I would like to have some feedback before moving on to drain etc.
One "test" is added as example/node_cordon.rs, but I wonder if it makes sense to have a conditional compilation feature test-integration.

Also, I suppose this change needs an entry in the CHANGELOG, but please let me know in case I am wrong.
Any suggestion is appreciated!

[clux]

initial comments. looks good, just minor points here and there.

[clux]

Currently only cordon/uncordon is implemented, as I would like to have some feedback before moving on to drain etc. One "test" is added as example/node_cordon.rs, but I wonder if it makes sense to have a conditional compilation feature test-integration.

I think having the bulk of node_cordon be an integration test inside kube-client instead makes sense here because it is so light weight.

If that works out we can maybe follow a similar approach for drain, but it might be a bit different because I don't see how we can schedule pods on a fake node to verify eviction 🤔

Also, I suppose this change needs an entry in the CHANGELOG, but please let me know in case I am wrong. Any suggestion is appreciated!

Don't worry about it! We are going to try a system with auto-generated changelogs from labels and pr title for next release.

Thanks for looking these help-wanted issues! Very much appreciated.

[ChinYing-Li]

Haven't push the "code formatting" commit yet; will do that once I wrapped up my attempt to implement drain. Thanks!

[ChinYing-Li]

The draft of node drain example is here, though I still haven't figured out how to make the created "fake" node become Ready. Testing with kubectl, the created fake node remains NotReady.

[ChinYing-Li]

I think it's best to open a separate PR for drain, which is a more complicated utility function, so this PR is ready for review! Thanks.

[clux]

Ok, will have a go over. I think it's possible that drain might be best served as an example, because it is non-trivial to startup a real node, get pods on it, then evict those, and drain it. If we can verify it as a one-off thing in an example, that might be good enough.

[clux]

minor nit on a comment, but otherwise happy with this 👍

[clux]

Looking good. I'll merge this in. I'll leave the original issue open (since we still want something for drain), but will add a comment therein.