I have a non HTTP server I run for supporting a popular VOIP/WebRtc protocol.
Server "A" is the old server. Server "B" is the new server. Both can run independently, but do the same thing. Server "B" is fully ready to go as the replacement for "A". Two days ago, the DNS entries were switched to point to "B" instead of "A".
The trouble is, I still see a lot of traffic going to "A" instead of "B". The TTL on the DNS entry (Route 53) was just a day (86400 seconds). So it seems like there are still clients out there either not re-querying DNS or some other issue with DNS caching.
I haven't ruled out misbehaving clients. But is there a general behavior for DNS caching downstream that I'm not accounting for that could delay the change from propagating longer than the TTL I have for the record on Route 53?