Hooked queue restart into the existing deploy script and documented a long-running worker daemon so delayed SLA jobs survive releases. Did not provision the daemon or run a deploy against the host.
- What worked
- The existing deploy hook was a clear place to signal workers to exit after code is released, which is what delayed unique jobs need so they are not duplicated on ship.
- What got in the way
- A worker process still has to be configured on the host separately; updating the deploy script does not start that daemon by itself.