High Availability, Resiliency, and Troubleshooting
Domain 7 tests whether you can make a dual-homed host survive a leaf failure and then prove, hop by hop, why a flow is broken.
Multihoming models
ESI-LAG: the same ESI on two leaves, LACP to the server, and multi-homing-mode all-active under protocols evpn interface ae0.0, so both leaves forward and the server gets an active-active bundle. EVPN gives aliasing (Type 1) and DF election (Type 4) so BUM from the server is delivered once. MC-LAG with ICCP is the older alternative and requires a peer link and inter-chassis control protocol.
Control-plane resilience
BFD on underlay sessions, BGP graceful restart, and nonstop routing (set routing-options nonstop-routing) so the data plane keeps forwarding while the routing process restarts. Anycast gateway removes the need for VRRP on IRBs.
Troubleshooting order
Underlay VTEP loopback reachable, then Type 3 route present for the VNI, then show evpn database for the destination MAC, then show ethernet-switching table for the local VLAN, then MTU, then filters.
Worked example
Host A on Leaf1 cannot reach Host B on Leaf2 in the same subnet. Ping 10.0.0.2 from Leaf1's loopback: succeeds, so the underlay is fine. show evpn database shows no entry for B's MAC; show evpn instance extensive shows the Type 3 route for VNI 100 arriving from the wrong VTEP address. The remote leaf has a stale vtep-source-interface pointing to a management address - fixed, and B's MAC is learned through the control plane within seconds.
Traps
ESI mismatch between the two leaves, which silently produces split-brain forwarding; omitting multi-homing-mode all-active so only one link carries traffic; troubleshooting the overlay before proving underlay loopback reachability.