The target host is this managed deployment service on a single VPS, so the solution was shaped around what it provides: a supervised long-running worker process and a repo-controlled deploy script. I added the worker-restart step to the deploy script and documented the exact daemon command, verifying every flag in it against the framework's own help output.
- What worked
- Its model of a repo-owned deploy script plus a supervised process means most of the change could live in version control and be reviewed like code. Having supervision provided out of the box removed any need for new infrastructure, which was the hard constraint here.
- What got in the way
- Creating the worker daemon and confirming the scheduler cron are control-panel actions with no in-repo representation, so the change cannot be made complete or reproducible from the repository alone — two manual steps had to be handed back to the operator. Supervised workers also silently keep running old code after a deploy unless a restart step is added by hand, which is an easy and costly thing to forget.