Tutorial
Ask your Laravel app why it cannot sleep
php artisan brewless:sleep asks every public page as a first visitor does, and names the rule each one breaks, so you know what wakes your container.
In this post
A serverless container sleeps only after a quiet spell: fifteen minutes at Scaleway, and every request that reaches it starts those fifteen minutes again. So the thing that decides whether it ever sleeps is not how busy your site is, but how much of it the edge answers by itself. One request a minute is enough to keep a container running day and night.
Knowing that does not tell you which of your pages is the problem. Why your Laravel container never scales to zero explains the rules; this post is about the command that applies them to your own application, page by page, and tells you what to change.
How do I find out what keeps my Laravel container awake?
Run php artisan brewless:sleep. The command comes with the brewless/laravel package, asks every public page the way a first visitor asks it, and names the rule each page breaks. Pages are grouped by reason, because on a real site hundreds of them share one.
It asks your application, it does not read it
A linter that greps for @csrf would miss most of this. Whether the edge can keep a page depends on the answer your application gives, after every middleware, view composer and service provider has had its say.
So the command builds a request for each address on your own APP_URL, sends it through the application's HTTP kernel without a single cookie, and holds the answer against the same rules the package's edge middleware applies: the answer must be a 200, 301, 308, 404 or 410, set no cookie, start no session, set no Cache-Control of its own, vary on nothing but Accept-Encoding, and not have looked at who is signed in.
Nothing leaves the machine, and no container is woken. It is the judgement the edge makes, made locally.
It asks only what a visitor can reach: GET routes, without a required parameter in the address, not behind an auth middleware, and not the addresses of tooling such as /_ops, Telescope, Horizon or Livewire.
What it prints
php artisan brewless:sleep
INFO What keeps this container awake.
Edge keeps a page for ................................ 3600 seconds
Guests without a session ............................. on
Sessions in .......................................... database
Queue on ............................................. sqs
Asked in ............................................. production, on https://example.com
Kept at the edge ..................................... 58 of 74 pages
14 pages — it keeps a session, so the answer is this visitor's. A page that shows a CSRF
token, or puts anything in the session, keeps one even when a guest has nothing in it
/
/pricing
/contact
and 11 more
2 pages — the answer is a 302; the edge keeps 200, 301, 308, 404 and 410
/docs
/careers
Every 60 seconds — Check pending payments, and every tick wakes the worker
3 files your pages name have no ?v= — the edge fetches those from your container, not from
the storage zone
/favicon.ico
/apple-touch-icon.png
/img/logo.svg
The block at the top is the handful of settings that decide most of it, as the container really has them: how long the edge may keep a page, whether a guest without a cookie gets a session, where sessions and the queue live, and which environment answered.
The reasons, and what to do about each
| What it says | What it usually is | What to do |
|---|---|---|
| it keeps a session | A CSRF token on a page with no form of its own, often from a newsletter field or a modal in the layout | Send that form with fetch() and leave the token out, or move it to a page of its own. Wrap routes that need no session in Route::withoutSession() |
| it sets a cookie (…) | Analytics, a consent choice, a remembered locale | Set it from the browser, or on the one request that needs it, not on every page |
| it starts a session | Something writes to the session on a page that shows nothing from it | Find the write; a flash message or a "last visited" counter is enough |
| the answer is a 302 | A trailing slash, an old address, a locale redirect | Link to the address it ends on, or make a permanent redirect a 301 |
it sets Cache-Control itself |
A middleware or controller that chose its own header, which silently overrules edge.cache |
Take it out where the page is the same for everybody; keep it where it is not |
| it varies on … | Accept-Language, or a header a middleware reads |
The edge keeps one answer per address and understands no Vary but Accept-Encoding, so give each variant an address |
| it looked at who is signed in | @auth or auth()->user() in a layout, often for one menu item |
Render that part from the browser, or accept that the page wakes the container |
Below the pages come two more things the command notices. Files your pages name without a ?v= are fetched from your container instead of from the storage zone, so give every <link>, <script> and <img> an address with a version made from the file's content rather than from filemtime(). And scheduled work that repeats within the hour wakes your worker that often; daily or hourly is almost always enough.
Ask production, not your laptop
A page can answer differently on a developer machine than it does live, and only the live answer decides what the edge keeps. The command says which environment answered, and tells you so when that is not production:
brewless command production "brewless:sleep"
Warning
A command has fifty seconds in your container. On a large site, ask in slices with
--path=/ --path=/pricingrather than waiting for a report that never finishes.
Keep it from coming back
One page with a form in the layout puts a session on the whole site again, and nobody notices until the bill arrives. --strict makes the command fail when a page cannot be kept, which is all a hook or a pipeline step needs:
php artisan brewless:sleep --strict
Run it against your own application, not production: it is the shape of your pages you want to catch, and the live answer is what you check after the deploy.
Hand the list to a coding agent
Every reason has a fix in your own code, which makes this a good task for Claude Code, Codex or Cursor: give it the output and let it work through the list reason by reason. The edge cache documentation has a prompt to paste, and brewless/laravel ships a brewless-sleep skill that an agent loads when a task calls for one, written into your agent's skills folder by php artisan boost:install or php artisan vendor:publish --tag=brewless-skills.
Ask for a reason for every page it leaves as it is. Some pages are supposed to wake the container.
What the command cannot see
Three things, and they are the ones that bite:
- A monitor from outside. An uptime check that polls every minute keeps your container awake on its own, and no amount of caching helps. Brewless never asks your application anything for that reason.
- Addresses with a query string. The edge keeps one copy per address, so
?utm_source=…is its own copy and its own first visit. Why?utm_sourcelinks wake your Laravel container is about that. - What visitors really ask for. The command knows your routes, not your traffic.
For the last one, the environment's page in the console has What reaches the container: the addresses of the last 24 hours that the edge did not answer itself, most first, read from the request log of its pull zone. The same list is in a terminal with brewless wakes production. Next to it, serverless usage shows how much of a period the container ran and how much it was scaled to zero, counted from Scaleway's own figures.
Judge a change there a day later, not in the hour you made it: a release empties the edge, so the hours after a deploy always look worse than a quiet day.
In short
- A container sleeps only after fifteen minutes without a request that reaches it.
php artisan brewless:sleepasks every public page as a first visitor does and names the rule it breaks.- A session on a page that shows nothing from it is the commonest reason, and usually a CSRF token in the layout.
- The answer that counts is production's:
brewless command production "brewless:sleep". - Add
--strictto a hook, then look at what reaches the container a day later.
Frequently asked questions
How do I find out what keeps my Laravel container awake?
Run php artisan brewless:sleep in the application. It asks every public page through your own middleware, as a first visitor asks it, holds the answer against the rules the edge follows, and reports the rule each page breaks, grouped by reason. Next to that it names the files your pages load without a version and the scheduled work that repeats within the hour.
Does the command send requests to my site?
No. Every page is handled inside the process the command runs in, so nothing leaves the machine and no container is woken by it. Running it on production with brewless command does use your live container for those fifty seconds, but that is one deliberate wake, not a crawl of your site from outside.
Which pages does it leave out?
Routes that are not GET, routes with a required parameter in the address, routes behind an auth middleware and the addresses of tooling such as /_ops, Telescope, Horizon and Livewire. A page only a signed-in visitor sees is never kept at the edge anyway, so asking it would say nothing.
Is it bad that some pages keep a session?
No. A cart, a wizard or a form that needs a CSRF token is meant to reach your container, and those pages should keep their session. What costs you money is the page that could have been the same for everybody and quietly is not, because then every crawler and every link preview starts the fifteen minutes again.
What if the command says edge.cache is 0?
Then the edge keeps nothing at all and every visitor reaches your container, whatever the rest of the report says. Set edge.cache in brewless.yml to how long a page may be kept, deploy, and run the command again.