Capacity
How much spare capacity should a service run with?
Whether the candidate can derive headroom from failure requirements rather than picking a comfortable number.
The situation behind the question
A service runs 12 instances across three zones at 80% CPU, autoscaling on CPU with a maximum of 16. A zone becomes unavailable during the evening peak. The remaining instances saturate, the health checks start failing under load, and instances are removed from rotation while the autoscaler is still starting replacements.