Fastweb has an application, called gateway-v4, that acts as an access point, allowing the user to access this pod and then from there access the other application pods. 
This gateway appliation contacts the other pods using the fqdn of the sharded routes.
From what I understand by analyzing the must-gather, the gateway pod asks the default-dns for the resolution of the sharded FQDN which in turn forwards the call to the coredns (openshift-vsphere-infra) which asks one of the two external DNSs for resolution.
During the upgrade, specifically whenever a master node restarts, the gateway pod receives an NXDOMAIN generating disservice.

This happened during all the upgrades    

    1.
    2.
    3.