When a change comes from dependabot, it's likely to change the go.mod / go.sum files.
We should not run the checkgenerated CI when prs are coming from it, as we do in metallb.

Similar to metallb/metallb#2409

github.com/pkg/errors should be updated to use errors as the package is no longer maintained

When merging two configurations with different prefixes, one config has allowmode: all for outgoing, only the prefixes local to the allowAll are advertised.

I know that we said a config must be self contained, but this is different from the incoming prefixes and I think not intuitive.
All means all.

Hi, there seems to be an internal SrcAddr setting to populate update-source for neighbors that is not exposed in the K8S API. Is that intentional?

I think my use case would benefit from it. I'm trying to expose some routes to IPv6 load balancers from multiple nodes in my cluster, and to do that I'm peering with an IPv6 router. Now, the interface that is connected to that router has many IPv6 addresses (on all nodes involved), so I need to make sure that the one selected by the kernel matches the one that is used to establish the BGP session from the peer end. This leads me to the following problem:

• In MetalLB in "regular" FRR mode, I can't seem to define two BGPPeer resources with the same neighbor address (even with mutually exclusive node selectors) because the API forbids that, so I can only align the source address for at most one node.
• With frrk8s.enabled=true, I thought I could get away with a single BGPPeer for all nodes and override the source address on a per-node basis by introducing extra FRRConfiguration resources scoped to specific nodes. But since this setting is apparently not exposed in the API, the sourceAddress feature doesn't work at all in this setup.

Is that something that simply hasn't been implemented yet, or is there something fundamental I'm not understanding here?

Since I control all the devices involved, I can probably work around this problem by giving extra IPs to the interface on the peer router's end, allowing me to create one BGPPeer per node in my cluster, but that feels like a hack. Thoughts?

Thank you!

Is there a way to inject service-ip/external-ip routes into the outbound advertisements?

Awesome project by the way! Love the ability to inject raw FRR configs.

I run a mixed-architecture kubernetes cluster and some of my services that require load balancer addresses run inside those arm64 nodes. It would be wonderful if there were native arm64 images available for me.