Bryan Boreham

> does ‘usage groups' feature solve this problem? Maybe, but it gets more expensive to compute the more splits you have, to the point where I wouldn’t want to use...

> So, it means it has to take more than 15s to propagate the changes, right? There is some jitter on how each node reacts to the times; this defaults...

Nit: I said to spread them 2-3 minutes _early or late_, so total spread 6 minutes. There is some spreading in current behaviour since we do max 10 concurrently; I...

Could you explain what "OCI" means here? Note you referenced a different repo too.

@asridharan thanks for the question; we discussed it at the maintainers' meeting today. There was general agreement that we do want to allow CNI to be used with hypervisor-based container...

Agreed, we need to clean up the language so it isn't directly contradictory. > give explicit examples of what they mean by these `hypervisor/VM-based plugins`? I can do that: [hyper_](https://hypercontainer.io/),...

We discussed this at the maintainers' meeting, and felt we don't have enough concrete information to make a spec change. So, if one or more of the interested parties would...

At the top level of libcni we would do something like: ``` span, ctx := opentracing.StartSpanFromContext(ctx, "ADD") ``` (so all the entry points need to add a context parameter) Then...

Note that there is no tracing in Kubernetes at the moment (https://github.com/kubernetes/kubernetes/issues/26507). Still, it might be useful if CNI itself is a performance concern, or for other orchestrators. (The upshot...

> possible to ask about the overall state of the network I meant the runtime could make a call at startup saying "is the overall network ready?", and give better...