Back to glossary

Health Check

An endpoint or probe that reports whether a service instance is functioning correctly, used by load balancers, orchestrators, and monitoring systems to detect and route around unhealthy instances.

Health checks are the foundation of automated reliability. Load balancers use them to remove unhealthy instances from the traffic pool. Kubernetes uses liveness probes to restart crashed containers and readiness probes to control when instances receive traffic. Monitoring systems use health checks to trigger alerts and incident response.

A basic health check returns a 200 status code if the service is running. More sophisticated health checks verify that critical dependencies are accessible (database connections, external APIs, model files), that the service can perform its core function (execute a test query), and that resource utilization is within acceptable bounds (memory, disk, GPU).

For AI serving systems, health checks should verify that models are loaded and inference is functional. A model serving health check might execute a test prediction with known input and verify the output matches expectations. This catches issues like corrupted model files, GPU memory errors, or dependency failures that would cause the instance to return errors despite being technically running.

Related Terms