Multi-pod control-plane walk — one Type-2 route crossing pods

Host 192.168.1.10 · Type-2 MAC/IP · L2 VNI 10100 · originated behind Pod-1 leaf VTEP 10.0.1.11 → installed on a Pod-2 leaf. Click a hop, or step through.

changed at this hop deliberately untouched — the property the design must preserve

Same route, same next hop, two delivery machines. In both models the Pod-2 leaf ends with next hop 10.0.1.11 — the tunnel stays leaf-to-leaf and no spine ever decapsulates VXLAN. Everything else about how the route got there differs.

Failure modes differ in kind: a broken RR in Model A hides the route — visible as absence in show bgp l2vpn evpn. A forgotten next-hop-unchanged route-map in Model B blackholes traffic — BGP looks healthy, the spine attracts VXLAN it cannot decapsulate, and only the data plane fails.

Model B assumes the rewrite-evpn-rt-asn design from section 7.4; the alternatives (fabric-wide static RTs, FRR wildcard import) change the RT rows but not the next-hop or AS-path behavior.