Pires

Those changes are related to the VK API authn/authz and this is just about TLS bootstrap, ie what happens before the VK API is actually served and any auth happens.

@myechuri I'm assuming you didn't get a chance to finish what you were working on, were you? Need help?

Two years and changes later, I still think this make sense. I can see providers exposing devices that are unique to their scenarios and leveraging that during scheduling.

Maybe @henryprecheur knows more?

@rbitia @cpuguy83 I've eliminated the `vk 1.0` label since we have now a milestone for it. Do you think this is reasonable to have as part of that milestone? I...

Would the operator be able to deploy different providers or just one particular provider? Is there a design of the CRD we can take a look at?

Please, go ahead and pick it up if it's meaningful to you. IMO this is not a VK thing per se, but an accessory for operators deploying VK instances, so...

IIUC this is not as trivial as an interface addition but actually requires the VK to support SPDY (which is part of moby/spdystream) so the kube-apiserver can do async bidirectional...

Unless I'm missing something, this is not something you can handle w/ HTTP/1 like we do for `kubectl logs` (eg https://github.com/virtual-kubelet/systemk/pull/69 and https://github.com/virtual-kubelet/systemk/pull/70), since `port-forwarding and exec are bidireccional, while...

#321 is also related and we hit a dead end, as the community was not willing to add an escape hatch for scenarios like the one described above. Also, for...