Workers
Keep queue workers, Horizon or any long-running command alive for a Git application, with restart-on-deploy and restart-on-crash.
A web request ends. Some work should not: a queue consumer, a scheduler, a Node process serving a server-side-rendered app. Workers is where you hand those commands to the server's process supervisor so they are started at boot, restarted when they die, and restarted after a deploy.
Until the first one exists the screen offers two routes to the same dialog: Add worker, or starting from the Custom command template.
Add a worker
| Field | What it means |
|---|---|
| Name | Your label for this worker, e.g. background-jobs. |
| Processes | How many copies of the command to run side by side. Defaults to 1. |
| Type | Queue worker, Horizon, or Custom for your own command. |
| Command | The command to keep running. Use template fills in a known one. |
| Restart after deploy | On by default — a deploy that changes the code restarts the worker so it picks it up. |
| Restart automatically if it stops | On by default. The supervisor brings the process back if it exits. |
| Enabled | Off leaves the worker configured but not running. |
Advanced settings holds the rest: Working directory, Log file and Log level, how long to wait for the process to stop before killing it, Start with the server, and a box for Extra supervisord settings when you need a directive the form does not expose.
Run the real process, not a package manager
The same rule applies here as to a start command: npm start or
composer run forks the real process, so the supervisor watches the wrapper
and shutdown signals never reach your code. Give the command that is the
process — node server.js, php artisan queue:work.
After it exists
Each worker gets a row with its state and controls to start, stop, restart, edit and delete it. Restart after deploy is the one worth leaving on: without it, a deploy updates the code on disk and your worker keeps running the version it loaded at boot — which is the kind of bug that takes a day to find.
Workers belong to one application and run as that application's system user, so a worker cannot reach another application's files.