Our use case about metallb HOT 2 CLOSED

Thanks for the experience report! This is useful information.

When you say you run a Bird instance on every host: is that part of the Calico network addon, or a custom pod for your infrastructure?

Per-node peer configurations are an interesting idea, I had not considered it for my simple use case. I'll create an issue for that, although I can't promise we'll implement it... But we should keep it in mind as a potential enhancement.

Thanks for the experience report! If you run into any trouble, don't hesitate to file issues/feature requests!

from metallb.

We run a bird daemon on each host in our infrastructure. In the particular case of k8s, we run it on CoreOS and we create a system unit that starts a container running bird, so everything happens before and outside k8s.

As a long term vision we would like to integrate our custom bird container functionality inside k8s using something like calico or kube-router but we need to solve the chicken-egg problem first. When we boot and install a CoreOS machine it needs to be able to download the resources necessary to run k8s installation and so on and to do that it needs a bgp daemon running (the machines can only talk to their rack neighbors without bgp). So we need to set up bgp even before we run k8s.

Maybe annotations annotations could be used somehow for the per-node configurations?

from metallb.