Container Runtime (beta): production deployment scales down after 30 seconds, docs say 5 minutes

The Container Runtime docs say functions with no traffic for 5 minutes in production, and 30 seconds in preview, will scale down. Our production deployment behaves like the preview figure.

Setup: Pro plan, Node.js (NestJS on Fastify) container, region pdx1, deployed with services (runtime: "container") in vercel.json, plus a separate worker daemon in the same project. Cold Start Prevention is on. The deployment's target is production, and we tested through its custom production domain.

Test: one client sending sequential GET /healthz requests at a fixed idle gap, no other test traffic.

Idle gap	Result
25s	Warm, 20 of 20
29s	Warm, 7 of 7
30s	Cold, 9 of 10 (about 5.8s)

Questions:

Is 5 minutes the current production window for containers using services?
If so, why would a production deployment get 30 seconds? Is there a setting or known beta issue?
If 30 seconds is correct, could the docs be updated?

We can share deployment IDs privately. Vercel's support assistant told us beta Container Runtime cases can't be opened and pointed us here.

Hi , I'm sorry you are facing this issue. Can you confirm if you are getting a SIGTERM signal?

When a function instance is scaled down, the container receives a SIGTERM signal with a 30 seconds grace period to clean up before being forcefully terminated. Try checking for the SIGTERM to see if the https://vercel.com/docs/functions/functions-api-reference#sigterm-signal

Additionally, the 5-minute threshold represents the maximum idle time before shutdown rather than a guaranteed minimum warm lifespan, meaning instances can occasionally be recycled earlier by the underlying orchestrator. This can also be a beta only behavior that I can confirm with the team once we have the SIGTERM information.