Readiness and liveness are different questions

A readiness probe controls traffic, while a liveness probe decides whether the process should be restarted. Returning the same deep dependency check from both endpoints couples unrelated recovery actions.

Keep liveness local

Liveness should establish that the process can make progress. Do not fail it because a database, queue or external API is unavailable; restarting every replica will not repair that dependency and can destroy useful diagnostic state.

Use readiness for service capacity

Readiness can remove a pod that cannot serve its workload, but apply timeouts and failure thresholds that tolerate brief downstream variation. Expose dependency state separately so operators can distinguish local and shared failures.

Protect slow startup

Use a startup probe for applications whose initialisation exceeds normal liveness thresholds. Test deployment, scaling and dependency-loss scenarios while watching endpoint removal, termination grace and in-flight requests.