use cluster dns to reach apiserver for rustls - for #587

[clux]

untested based on suggestion in #587 to fix #153

[nightkr]

FWIW, you still need to use the environment variable when implementing a custom DNS server...

[clux]

Do we? I thought that was the whole point to rely on DNS, rather than use the evars. Bumped PR.

[clux]

This passed integration tests on k3d now.

[nightkr]

Do we? I thought that was the whole point to rely on DNS, rather than use the evars. Bumped PR.

Yeah, but the DNS server itself needs to have some way to reach the apiserver as well (chicken-and-egg problem :p). Not contesting that this is the correct solution for 99% of applications though.

[clux]

Ohh, when implementing a DNS server yeah. I get you. Yeah, I guess we need to expose a (probably non-default) way to decide what method to use when using incluster config.

[MikailBag]

I think kube-rs should just expose a way to completely override apiserver address (after all, it is perfectly valid for a Kubernetes pod to access apiserver from another cluster). And then user can read environment variables, some configs, etc.

[clux]

I think kube-rs should just expose a way to completely override apiserver address (after all, it is perfectly valid for a Kubernetes pod to access apiserver from another cluster). And then user can read environment variables, some configs, etc.

This is actually supported out of the box already:

let mut cfg = Config::from_cluster_env(); // or Config::infer
cfg.cluster_url = some_uri;

[clux]

Going to merge this, tested this locally with controller-rs against a cluster running rustls for sanity so going to speculatively close the related issues. If it turns out to be positive, then we can discuss whether to swap out the default for non-rustls.