Shedding old code with ecdysis: graceful restarts for Rust services at Cloudflare

Cloudflare has open-sourced ecdysis, a Rust library that enables graceful restarts with no dropped live connections or refused new connections. Developed for years in production and open-sourced last month, it has powered zero-downtime upgrades across Cloudflare’s Rust infrastructure. In production since 2021, it runs across 330+ data centres in 120+ countries, handling billions of requests per day and safeguarding uptime during updates.

The approach avoids the gap that a naive stop/start creates, with the parent forking a child, passing the listening socket, and only then shutting down after the child signals readiness. If the child encounters a crash during init, the parent continues serving existing connections and the upgrade can be retried. The project includes Tokio integration, systemd-notify support and systemd sockets, and the documentation and examples are available on docs[.]rs and GitHub.