A customer reported intermittent traffic blackholing during graceful failover events of a WatchGuard HA/VRRP firewall cluster connected to a pair of QFX5100 leaf switches operating within an EVPN fabric. During the outage, the VRRP virtual MAC address was simultaneously learned on both firewall-facing access interfaces, resulting in MAC mobility events and loss of forwarding stability.
Investigation determined that the EVPN fabric was correctly reacting to MAC ownership changes advertised by the connected firewalls. Evidence showed the VRRP virtual MAC being sourced from both firewall connections during failover, causing continuous MAC movement between local interfaces and remote EVPN VTEPs.
Analysis showed that the EVPN fabric was functioning as designed.
The VRRP virtual MAC address was learned locally on both leaf switches at approximately the same time:
Switch01:
<VIP mac address> learned on xe-0/0/43.0
Switch2:
<VIP mac address> learned on xe-0/0/46.0
As a consequence, EVPN generated MAC mobility events while ownership alternated between the local firewall-facing interface and the remote VTEP.
These indicate that:
The WatchGuard HA/VRRP implementation occasionally advertised or sourced traffic using the same VRRP virtual MAC from both firewall nodes during a failover event.
This caused:
The issue was ultimately determined to be related to the behavior of the WatchGuard HA implementation rather than a switching, EVPN, or forwarding defect on the QFX platform. The customer later confirmed that introducing a Layer 2 device between the leaf switches and the firewall cluster resolved the issue and aligned with guidance received from the firewall vendor.