I conduct actual experiment with my domain, how to say, « testing the failover ». Three DNS providers: KnownHost, RackNerd, Leaseweb. All with different IP, different ASN, the GeoDistribution.
Setup: primary NS at KnownHost, secondary RackNerd, tertiary Leaseweb. I am cut them one by one with the FirewallRules.
Test 1: kill KnownHost. RackNerd take over in 2min. Yes, working.
Test 2: kill RackNerd too. Leaseweb alone, still resolve. Good.
Test 3: kill Leaseweb. Dead. Total Dead. Why?
The Glue Records at the Registrar point only to KnownHost IPs. When all three NS names resolve to dead IPs, no path to find the remaining hidden IP. Single Point of Failure was the Registry itself.
I am very surprised. Three providers, zero redundancy in practice.
Cost of experiment: one broken domain, 4 hours of debugging, my Ego.