Hot reload downtime under load

[utay]

Describe the bug

In production a router process serves hundreds of requests per second; when it hot reloads because the supergraph has been refreshed, a few requests get 502s from the load balancer because it got connection refused on the router.

To Reproduce

I've been able to reproduce pretty consistently:

• Start the router with hot reload enabled
• Run any load testing tool such as ab -n 1000000 -c 20 http://127.0.0.1:4000/graphql
• Modify the supergraph to make the router reload
• 💥 Right around "reload complete", at least one request will fail with connection refused

Expected behavior

I'd expect 0 downtime as per the docs.

Additional context

• Router 1.45.0

[Geal]

hi, thank you for the report. This looks like an issue we have seen elsewhere, we'll investigate and get back to you