readiness probe is flapping because /readiness does deep dependency checks (Postgres/Redis/Box) and sometimes takes ~3–4s+, occasionally more via proxy. Kubelet times out and marks the pod Unready, which slows rollouts and reduces capacity.

ports:
            - name: http
              containerPort: XXXX
          livenessProbe:
            httpGet:
              path: /health
              port: http
            periodSeconds: 20
            initialDelaySeconds: 30
            timeoutSeconds: 15
            failureThreshold: 3
            successThreshold: 1
          readinessProbe:
            httpGet:
              path: /readiness
              port: http
              httpHeaders:
                - name: Accept
                  value: application/json
            initialDelaySeconds: 25
            periodSeconds: 10
            timeoutSeconds: 15
            failureThreshold: 3
            successThreshold: 1