And get some replication in there. Even if there's not a single point of failure, if a DB instance ever goes tits-up, you'd better have a standby.

Source: I've cleaned up others' messes where they didn't.