That's a very good point. Unfortunately is not as simple as having discrete states: 'healthy', 'degraded'. For example a tier might be deployed to an auto-scali

That's a very good point. Unfortunately is not as simple as having discrete states: 'healthy', 'degraded'. For example a tier might be deployed to an auto-scaling-group (ASG) running 100 instances. One of them is failing the health-check. Common sense dictates you can take action (kill the offending instance). What's the overall state of the ASG after this? Are you allowed to kill another instance that's having issues? How many after that? In general you want to put some rules into place. (Percentage of the ASG that's safe to kill, latency boundaries for the service, time to wait before taking the next action, etc.)