We have 2x /27 public IP ranges from two different ISPs at two different buildings. We have our actual servers at only 1 of the locations, which I’ll call “location A”. We don’t have any of those setup in a DMZ, instead we have NAT rules for each of those. In the event that the ISP at location A goes down, the 0.0.0.0 route will actually go across town and use the ISP at location B with a seperate /27.
What we’d like to do is check if “Location A” is down, and switch over a total of 21 A records to point to “Location B”. The reason we can’t do Round-Robin is because there aren’t 2 concurrent routes out. There would be other means of fixing this, like having a DMZ that get’s vxVlan’d across town, and obviously we’d have 2 NICs and 2 paths at all times to both ISPs. But say we wanted 21 A records (42 IPs) to be used that ultimately would go to the same 21 servers internally, but would failover if a ISP went down, would we be paying for 21 “load balancers” at $5/piece (2 origins per load-balancer)? Are there any other health checks that can be done that aren’t http(s)? (ICMP Ping?)