Hi all. Faced a non-trivial task. The meaning of the action is to implement fault tolerance in such a way that if one of the channels or servers fails, the domain will resolve to a live IP.
I want to clarify that such "chips" as the creation of AA records, according to the scheme:
randomnet.com# host domain.com
domain.com has address хх.ххх.хх.хх
domain.com has address хх.ххх.ххх.ххх
in a zone for the domain do not work. Due to the fact that with such a scheme, the resolve goes randomly on both IPs and in the event of a loss of communication with one of the servers, the packets will still go nowhere.
If someone knows a solution to such a problem with minimal labor and technological costs (do not offer AS + BGP), I will be grateful.
If someone doesn't understand, you need to do load balancing or duplication of two machines that will be interchangeable in case of a neighbor's failure (I heard that it can be implemented on a cluster with OBSD or pFSense, but I don't know how).
In the vast majority of cases, 100% webserver load means that you will not be able to serve some user requests, that is, you will have less "effective uptime" than 100%, and very dissatisfied users.
So, I would not formulate it so gently.
The problem is that often the start of a project begins with uncertainty about what the attendance will be and how the requirements will change later. You can spend a lot of time and resources on scalability from scratch and get a dead project, because more nimble competitors have already squeezed out market share, received feedback and are sawing the architecture of their projects with the money received.
It is possible, by providing scalability in the same way, to get new requirements, which is why everything will have to be redone. generally, no matter how much vodka you take, you still have to run twice.