Planning to standardize a dual-DC core into leaf–spine with EVPN/VXLAN to reduce east‑west latency about 20% and cleanly integrate segment routing and gNMI telemetry; anyone running this across mixed vendors with MP‑BGP RRs at the border and only two VLANs stretched for legacy? Looking for real-world scaling and convergence gotchas (host routes, IRB, ZTP) that might derail a seamless cutover.
Biggest saver for us was enabling RT-Constrain on EVPN and turning off ‘advertise-all-vni’ on the border RRs so only the RTs you expect are imported — otherwise mixed vendors will happily leak every type‑2 host route into your L3VNI during ZTP churn. Are you going symmetric IRB end‑to‑end?
Match the DF algorithm (HRW/mod) for ESI multihoming across vendors and lock the anycast‑gw MAC/IP per VRF; otherwise you’ll hit MAC musical chairs and brief IRB blackholes. Add BFD on the underlay and conservative MAC‑mobility dampening so vMotion/ZTP churn doesn’t spray type‑2 updates everywhere. @meganf21, are you running symmetric IRB end‑to‑end or keeping type‑5 only at the border?