Why a 200 Health Check Does Not Mean Your Microservice Is Actually Healthy

A microservice returning a healthy status code does not guarantee that users are receiving correct responses, as standard health checks only observe one narrow dimension of service behavior at a single moment. Like Schrödinger's cat existing in two states until measured, a service can simultaneously appear healthy to monitoring tools while failing real users across unobserved dimensions such as database pool exhaustion, stale cache data, or broken code paths. Different probe types — process pings, DB connection checks, synthetic transactions — each reveal specific failure modes while missing others, meaning no single check captures the full picture. The observer effect adds further complexity: health checks that open database connections consume resources and can compete with real traffic under load, meaning the system being monitored may differ from what users actually experience. Engineers are advised to treat service health as a probability distribution across many dimensions rather than a single binary value, and to design observability strategies that sample multiple failure modes in parallel.