Network Disconnection (10-30s) During ESXi vmnic Failover in a Cisco ACI Environment
Article ID: 442479
Updated On:
Products
Issue/Introduction
When a vmnic failover occurs on a VMware ESXi host, both the ESXi Management network and the Virtual Machines running on that host experience a network disconnect or traffic drop lasting between 10 to 30 seconds before connectivity is restored.
Live troubleshooting and packet captures confirm that the failover operation on the VMware ESXi side is executing immediately and successfully. However, the packet captures indicate that the Cisco ACI physical switching fabric is delaying the traffic failover to the newly active switchport.
Environment
VMware ESXi
Cisco Application Centric Infrastructure (ACI) switching fabric
Cause
The Cisco ACI fabric default endpoint learning behavior (data-plane learning) requires time to age out old entries during a MAC move.
When a VMware ESXi vmnic failover occurs (e.g., from vmnic0 to vmnic1), the MAC addresses associated with the host and VMs instantly move to a new physical switchport. Because ACI requires time to age out the old data-plane entry and learn the new physical port mapping for the existing endpoint, traffic gets dropped for 10-30 seconds until the ACI fabric completes this update.
Resolution
- Enable EP (Endpoint) Move Detection Mode (GARP Detection Mode) in the Cisco ACI fabric.
NOTE: This must be configured for each of the specific VLANs/EPGs (Endpoint Groups) consumed by VMware Cloud Foundation (VCF).
When a vmnic failover occurs, VMware ESXi natively broadcasts a Gratuitous ARP (GARP) to notify the physical switches of the MAC address move. By enabling GARP-Based EP Move Detection Mode, the Cisco ACI fabric is instructed to actively listen for these GARP packets and immediately update its endpoint tracking tables to reflect the new port. This eliminates the 10-30 second data-plane learning delay, ensuring a seamless and rapid failover within the same EPG with virtually zero dropped packets.
Additional Information
Feedback
