How it works
Liveness
A liveness probe asks whether the process is stuck and should be restarted, not whether every dependency is currently healthy.
Readiness
A readiness probe controls whether an instance receives traffic, accounting for startup, draining, and temporary inability to serve.
Shallow vs deep checks
Deep checks catch dependency failures but can amplify an outage if every instance declares itself unhealthy because one shared service is down.
Timing matters
Intervals, timeouts, and failure thresholds balance quick removal against flapping during brief latency spikes.