CD

Cloudflare Developers•3mo ago•

4 replies

ceelsoin

Hello everyone,from workerd runtime

Hello everyone,from workerd runtime

I’m currently experimenting with self-hosted workerd at the company I work as part of a proof of concept to evolve our internal serverless platform (https://github.com/globocom/functions).

While studying Cloudflare Workers’ production architecture, I saw in some diagrams that there is a supervisor layer responsible for managing a pool of workerd instances, restarting immutable instances and reloading the capnp configuration to update workers.

I’m building a similar supervisor myself, and my current idea is to rely on

--socket-fd

--socket-fd

to run multiple workerd processes in parallel listening the same socket when an update is required. The goal would be to start a new workerd instance with an updated config, shift traffic to it, and then gracefully shut down the old instance — achieving zero downtime.

However, I have a few questions about this approach, and I’m not sure if I’m heading in the right direction.