Symptoms:
In this scenario where there is a single VDS in use by the unreachable VMs and the Edge node VMs, the unreachable VMs will have ARP entries for the Standby Edge node and no ARP entries for the Active Edge node.
VMware NSX-T Data Center
VMware NSX 4.x
The root cause of this behavior is not currently known.
Separating Edge nodes across ESXi hosts will protect against a host failure scenario.
Migrate one Edge node to a different ESXi Host than the active node.
Create an affinity rule (VM-Host) to keep Edge nodes on separate ESXi Hosts.
Similar behavior has been seen in older ESXi/NSX-T versions.
If this issue is seen, please collect and provide the following data through a Support Request: