It seems to me up to 8 records (ie. 8 IPs) per endpoint (like "api.mysuperwebsite.com" will return up to 8 IPs that can serve the request for that endpoint). That would mean that you can have hundreds of endpoints, but each of them would return up to 8 records. I wonder if there are more than 8 corresponding IPs, how the returned ones are selected (besides the health check), but that's a different question, I guess.

Considering this is DNS there has been a historical limit of 512 bytes. Despite this not usually being a limitation now you are really pushing the ideal packet size with hundreds of answers each of which are multiple bytes. High chance of packets being dropped.

> Considering this is DNS there has been a historical limit of 512 bytes.

Only with UDP transport, longer responses are told to requery via TCP.

These days EDNS0 allows bigger UDP responses in many cases, which may mean some fragment re-assembly. Unfortunately there are a staggering number of networks and firewalls that don't open TCP 53, and also ones that don't permit UDP fragments. So if you want DNS to work reliably /everywhere/, sadly it's wise to stay below the 512 limit.

We're talking about service discovery here. This is internal DNS traffic in AWS, where these issues to which you refer are nonexistent.

That means 8 IPs will be returned in a DNS query.