Atlas: Readiness Flapping

What this usually means

Your traffic gate is unstable. Either the workload cannot sustain load, the probe endpoint is too strict, or the probe is checking a dependency that is intermittently unhealthy.

What to inspect first

Correlate readiness transitions with load and probe failures.

• Look for timeouts vs explicit failure responses.
• If endpoints churn, validate whether upstream retries are amplifying pressure.

kubectl describe pod <pod> -n <ns>
kubectl get events -n <ns> --sort-by=.lastTimestamp | rg -n "Readiness probe failed" | tail -n 30 || true

Resolution guidance

Fix semantics first. Tune numbers second.

• Ensure readiness means “can serve” and the endpoint is fast and stable under load.
• Remove deep dependency checks from readiness unless required; otherwise you create cascading failures.
• Tune timeouts based on measured p99 under expected load; add startupProbe when warm-up is slow.
• If the service is genuinely overloaded, reduce load or add capacity—don’t tune probes to hide collapse.