Dave Anderson

Those pods/nodes won't get any traffic. When using the `Local` traffic policy, each node is responsible for attracting traffic to itself, if it contains pods for the relevant service. In...

You're correct about the additional behavior of the `Cluster` policy. Courtesy of kube-proxy, it will do "perfect" even distribution across all service pods, at the cost of rewriting the source...

Yup, ECMP group size is a concern. The 100% fix for that is WCMP (weighted cost multipath), which the silicon typically supports but the upper layer protocols don't really. Thanks...

So far, I've only tested with OSS BGP implementations, mainly [BIRD](http://bird.network.cz). I used to have some Juniper and Cisco hardware that could speak BGP, but I got rid of them...

@xnaveira MetalLB supports 32-bit AS numbers! It's the one BGP extension that's implemented in v0.1.0. v0.2.0 will also support MP-BGP, so if you use Quagga or Cisco devices, you'll want...

\o/ What patching did you need to interface with JunOS? I'm going to spin up a virtual JunOS later today and do interop testing, but it sounds like you already...

In general, MetalLB already checks existing services with an assigned IP, and verifies that the IP makes sense given the configuration (i.e. that this service is "allowed" to have this...

That's what this bug is about: the behavior I described doesn't happen, because MetalLB refuses to apply the new configuration. This bug is about removing that.

Okay, I've talked this over with a coworker (hi @maisem!), and I think I know how this needs to work. It deserves a proper design doc with graphs and stuff,...

That's something to take up with upstream kubernetes, it's not something I control. Offhand I can't think of a reason why it couldn't become a LoadBalancer (other than circular dependency...