About 2-and-a-half years ago I wrote a blog post detailing how I was able to use the HAProxy agent-check feature to avoid DOWN Patroni hosts, instead putting them into MAINT mode on a read/write backend when not primary. This is very useful if you are, for instance, monitoring for HAProxy DOWN hosts for alerting or such, as otherwise 2/3 (or more) of your read/write backends would always report DOWN.

The script I describe there has served me well, but it's very hacky. It would be nice to have this integrated into the existing Web API (port 8008). In short, it would be an endpoint at a separate port that would return either READY on a write-accepting replica, or MAINT on a write-denying secondary. The API already works mostly like this in terms of detecting up/down hosts with the 200/503 status codes, but the key point of the agent-check is that the remote side is actively specifying the maintenance mode state in textual form regardless of status code.

The HAProxy documentation for reference: https://cbonte.github.io/haproxy-dconv/1.8/configuration.html#5.2-agent-check