BGP issue with IPV6 and multipath/ecmp

[vcharlet]

On a cluster with IPV6 only connectivity, i'm trying to enable BGP with nodeToNodeMeshEnabled: true.
We have two routers in our datacenter that distribute the default gateway via router advertisements (RA) for redundancy.

By default, linux accepts both gateways and uses ecmp. The network works well, etc ...

However, there is an issue with Calico, see screenshot below :

Calico fail to parse the default route.

Steps to Reproduce (for bugs)

1. Enable BGP in an IPV6 network with two routers.

Your Environment

Kubernetes RKE2 v1.27.12 +rke2r1
IPV6 only
Calico v3.27.2
All nodes are on Debian 12

Thanks for your help !

[caseydavenport]

This error seems to be coming from the "CalicoNodeStatus" controller, here:

calico/node/pkg/status/populators/route.go

Lines 153 to 156 in e592387

	if len(previousDest) == 0 {
	log.Errorf("No destination available, failed to parse (%s)", line)
	return "", false
	}

I think this means it shouldn't be impacting actual networking, which seems to match what you're describing, and can be circumvented by deleting and CalicoNodeStatus resources (https://docs.tigera.io/calico/latest/reference/resources/caliconodestatus)

This does seem like a bug though - that status logic likely isn't expecing the ECMP routes and doesn't have logic to parse them.