Health Checks and Service Discovery

A backend service rarely runs at one permanent IP address. Instances start, move, scale, restart, and disappear. Even when an address exists, the process behind it may still be initializing, overloaded, or unable to serve useful work.

Callers therefore need answers to two separate questions:

Discovery provides candidates. Health checking filters those candidates according to an observer's recent evidence.

A discovered endpoint is not necessarily healthy. A healthy process is not necessarily reachable from every caller. A health check is also not proof that the next real request will succeed.

These mechanisms manage uncertainty; they do not eliminate it.