This article describes the impact of connecting a single ESXi host using active/active uplinks to two independent EVPN/VXLAN fabrics located in different datacenters without EVPN multihoming (ESI-LAG).
In the reported deployment, the ESXi host was connected simultaneously to a leaf switch in DC1 and a leaf switch in DC2. Each datacenter operated as an independent EVPN/VXLAN fabric with separate leaf and spine infrastructure. The customer experienced intermittent gateway reachability issues and VMware vMotion failures. Investigation determined that the design could result in MAC learning inconsistencies across fabrics, leading to duplicate MAC detection and traffic forwarding anomalies.
The following symptoms may be observed:
The ESXi host was connected active/active to two independent EVPN/VXLAN fabrics using standalone interfaces rather than an EVPN multihomed ESI-LAG.
Because both fabrics learn and advertise the same host MAC address independently, MAC ownership ambiguity can occur. This condition may lead to:
Although this is not necessarily a traditional Layer 2 loop, the forwarding behavior can create symptoms similar to a loop or MAC flapping condition
Connecting an ESXi host active/active to two independent EVPN/VXLAN fabrics can result in duplicate MAC detection, MAC ownership ambiguity, asymmetric forwarding, and intermittent connectivity issues.
The recommended solution is to use Active/Standby host connectivity or EVPN multihoming (ESI-LAG) and clear any existing duplicate MAC suppression entries after correcting the underlying design.