Watch, reap, wake
The reaper is a plugin on the routes of your apps, and a job that drives them through their lifecycle with the Clever Cloud API. Nothing else to deploy.
Every request through a route under the reaper is an access, recorded in memory on the node that served it, and flushed every few seconds. No log pipeline, no analytics backend.
An app without traffic for its grace period, outside its must-be-up hours, is stopped through the Clever Cloud API. Its instances stop costing anything.
The next request starts it again. A browser gets a page that reloads itself once the app answers; an API call is held until the app is up, or gets a clean 503.
Every route, and where its app stands
A table of every route with the state of its app, and a page per route to tune it, act on it and read its history.
On Clever Cloud since day one
Otoroshi has been deployed on Clever Cloud since its very first day, and Clever Cloud now runs it as a managed service, Otoroshi with LLM. An Otoroshi in front of Clever Cloud apps is a common setup: it already sees every request those apps get, so it is the natural place to notice the ones nobody uses, and to wake them up when someone does.
The reaper talks to the Clever Cloud API with an API token, counts what sleeping saved with the Clever Cloud prices of each zone, and links every app to its page in the Clever Cloud console.
What the reaper gives you
Grace period, must-be-up hours, monitoring requests to ignore, what callers get while the app boots, a custom waiting page. Add the plugin to a route and you are done: no other entity to create.
The page polls its own URL and reloads the moment your app really answers it: not when a deployment says it is done, and not on a timer. Bring your own HTML if you like.
An API call on a sleeping app can simply wait, held open before the backend call, and go through once the app is up. To the caller it is a slow response, not an error.
Keep apps up during the hours people use them, started before they arrive. Tell uptime checks and health probes apart from traffic: they neither keep an app awake nor wake it up.
A table of every route with the state of its app, and a page per route with its settings, its actions (wake up, put to sleep, reset) and the history of its app.
One job per cluster drives the apps, with a lock per app so nothing is ever started twice. Workers serve from memory and report their traffic to the leaders.
An app whose last deployment failed is never stopped. An app that fails to stop or start goes into error, raises an alert, and is left alone until someone looks at it.
Start in dry-run to read what would be put to sleep. Flip the kill switch from the console to stop all reaping across the cluster, at once, without a restart.
Every transition is kept per app and sent as an Otoroshi event, to route through any data exporter. An app in error raises an alert.
Five minutes to a first sleeping app
Download the jar, give it a Clever Cloud API token, and enable the reaper on a route from the console. The quickstart walks through it with a two minute grace period, so you see it happen.
# a clever cloud api token, for the reaper
clever tokens create "otoroshi reaper"
# otoroshi and the extension
curl -L -o otoroshi.jar \
'https://github.com/MAIF/otoroshi/releases/download/v18.0.0-rc1/otoroshi.jar'
curl -L -o reaper.jar \
'.../otoroshi-clevercloud-reaper/releases/download/<version>/otoroshi-clevercloud-reaper_3-<version>.jar'
# run them together
CLEVER_CLOUD_API_TOKEN='...' java -cp "./reaper.jar:./otoroshi.jar" \
play.core.server.ProdServerStart