Sleep

What your agent writes

Your agent writes it in while.yaml, at the root of your app:

services:
  - name: web
    port: 3000
    sleep: 1h
environments:
  production:
    services:
      - name: web
        sleep: never

When to use it

For a rarely called service: it reserves nothing while sleeping, and wakes up in 2 to 4 seconds, plus the app’s startup.

How it works

A web service or a site that no longer receives requests goes to sleep and no longer reserves anything. The first request wakes it and waits for it to answer.

What to know

The setting goes per service in while.yaml. sleep: 30m puts it to sleep after 30 minutes without a request, and sleep: never keeps it always awake, with at least one instance; the delay is written in minutes or hours (30m, 2h). By default, a service stays awake in production, and goes to sleep after 30 minutes without a request elsewhere, as in a preview. A worker and a task never sleep.

A sleeping service restarts from one instance when it wakes: scale.min only counts if it does not sleep, and scale.min: 0, the old way of saying it, means sleep: 30m. while’s background monitoring does not wake a sleeping service, but the get_app_status tool wakes it, except with probe: false. status also gives the wake-ups of the last 7 days (wakeups.median_ms): a heavy startup gains by loading what it can on demand.

A preview whose services all sleep, with no visit, read or deployment for 24 hours, has its database put to sleep. The next visit gets a “waking up” page (503) that reloads itself, and the database is ready in about ten seconds. A script gets 503 and Retry-After, and a deployment wakes the database and waits for it. A production database never sleeps.