Your zero-downtime deploy is probably fine. Check your p99 before you believe it.
dev.to·
Achieving zero-downtime deployments is a standard goal for production backend systems, but surface-level health checks can mask serious operational flaws. This practical analysis explores what happens during a rolling restart of Node.js and Express replicas behind an Nginx proxy. When a replica shuts down mid-request, Nginx can automatically re-route in-flight HTTP requests to another node if headers haven't sent yet. While this prevents client-facing errors, it quietly duplicates execution, causing p99 latency spikes and dangerous side effects for non-idempotent operations like payment authorizations or outgoing emails. As you design resilient distributed systems, auditing proxy retry behavior and ensuring strict endpoint idempotency are vital steps to ensure your deploys are truly seamless under heavy traffic.