Very cool. How does failover work? I see that Spilo expects a standard load balancer in front of the cluster, but that load balancer won’t be able to track who the writer instance is on its own.
You can put HAProxy in front with a write- and a read-frontend with a backend each and all servers in the backend. To determine which server is a write instance or a standby you can provide a `external-check command` to the backends. That command can be a bash script, that connects to the server and executes `SELECT pg_is_in_recovery();`.