Health Check Configuration
The interval, timeout, and healthy/unhealthy thresholds that together determine how quickly a failed backend leaves rotation.
The number that matters is time to eject = interval × unhealthy threshold, plus the check timeout. An interval of 30 seconds with a threshold of 3 means 90 seconds of requests sent to a dead backend, and that figure goes directly into availability arithmetic. Teams routinely assume it is a few seconds.
Tuning is a trade: shorter intervals eject faster and add load and false positives; higher thresholds tolerate blips and slow ejection.
Two things worth getting right. The check must exercise something meaningful — a handler returning 200 unconditionally passes forever while the service cannot serve. And the healthy threshold should be higher than the unhealthy one, so a flapping instance is not returned to rotation immediately.