Currently the endpoint is a CNAME to control1003, where a single haproxy then forwards to the nodes, but if contol1003 goes down we have to manually change the CNAME to another host to do the failover.

The goal is to review the current setup and achieve at least automatic failover (active-passive), and in the best case scenario load balancing (active-active).

Maybe check if it's possible to do DNS load balancing on the client (if the clients support getting more than one ip from DNS and then use that as 'servers' and fail over to the next).