I had always thought of Keepalived mainly as host-level failover with VRRP. While working on containerized DNS HA, I realized Keepalived track scripts can make failover decisions based on the application itself. This is pretty cool functionality. For instance, with DNS you can check to see whether port 53 or a DNS query is actually responding. Check out the full config here, including rise, fall, interval, and the difference between automatic failback and nopreempt.