Services are generally deployed in clusters of redundant instances (for reliability and scalability, as described in Chapter 4, Scaling Applications, and Chapter 9, Anti-Fragile Systems). It is important for only clients to access health instances, to avoid service unavailability issues.

Health checks help in this regard. Generally, the pattern calls for each service instance to perform a deep health check (that is, check all subsystems) and tell the LB or the service registry about the instance health. Generally, the LB or the service registry has an agent which makes a call to each service backend instance and expects a response within a given time. To ensure good performance (low latency) for the health check API, one can make the actual deep health check asynchronous to the health URL API response.