It looks like this behaviour is attributed to one misbehaving node in our DB cluster.
There was a hickup 2-3 weeks ago where the DB package failed to install/upgrade. In consequence, updates sent to that nameserver were not retained and propagated to the other nameservers. In other words: sending an update to
ns1 would not have shown up on
ns2 (and vice versa; although I’m not sure which servers were actually affected).
The fact that the DNS record for
dynv6.com returns the IP addresses of the nameservers in a round-robin fashion concealed the issue for most of our users. It strongly depended on the system’s/router’s/ISP’s DNS client for how long they cache the A/AAAA response for
dynv6.com for the issue to show up.
Since the DB cluster now is in a working state again, and I cannot reproduce it, I’m considering this issue resolved.