Design Probes That Mean Different Things
Design startup, readiness, and liveness probes that protect traffic and avoid premature restarts.
stepper comparison common-trap-callout The recommendations API needs 75 seconds to load a model, but a liveness probe starts killing it after 20 seconds. Use startup to protect initialization, readiness to gate traffic, and liveness to restart only a truly stuck process. The common trap is reusing one /healthz endpoint for every probe. That makes Kubernetes treat slow startup, temporary dependency trouble, and process deadlock as if they were the same failure. Step 1 Create a startup probe for /started with enough threshold to cover normal model load plus headroom. Startup shields slow initialization from premature liveness restarts. Step 2 Create…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in