I may be running into a situation that is completely normal. But I want to talk it out anyway. In my home lab, I have a single worker node Rancher-controlled k3s cluster. I also have a FRR VM acting as the BGP peer to MetalLB within the cluster, since a UDM Pro cannot run BGP natively. I spun up a simple nginx 1-pod deployment and backing service with LoadBalancer IP. Everything did its jobs, and the IP is accessible.

• Client desktop: 192.168.0.121
• UDM Router: 192.168.0.1 / 192.168.100.1
  • static route for 192.168.110.0/24 nexthop 192.168.100.2
• FRR BGP Router VM: 192.168.100.2
• k3s worker node: 192.168.100.11
• MetalLB BGP-advertised service subnet: 192.168.110.0/24
• nginx service LoadBalancer IP: 192.168.110.1

The FRR router VM has a single vNIC, no tunnels or subinterfaces, etc. Accessing the nginx service LoadBalancer IP by HTTP is perfectly fine, so I know routing is fine. But from a ping and traceroute perspective, it looks like I have a routing loop.

Client traceroute:

PS C:\Users\sbalm> tracert -d 192.168.110.1

Tracing route to 192.168.110.1 over a maximum of 30 hops

  1    <1 ms    <1 ms    <1 ms  192.168.0.1
  2    <1 ms    <1 ms    <1 ms  192.168.100.2
  3     1 ms    <1 ms    <1 ms  192.168.100.11
  4    <1 ms    <1 ms    <1 ms  192.168.0.1
  5    <1 ms    <1 ms    <1 ms  192.168.100.2
  6     1 ms    <1 ms    <1 ms  192.168.100.11
  7    <1 ms    <1 ms    <1 ms  192.168.0.1
  8     1 ms    <1 ms    <1 ms  192.168.100.2
  9     1 ms    <1 ms    <1 ms  192.168.100.11
...

Something doesn't feel "normal" here. Ideas?