Switching a site between FPM and FrankenPHP left its worker units pointing at the previous container. Going back to FPM removed the per-site FrankenPHP container while the queue and schedule units still had BindsTo and exec'd into it, so the workers could not start and healing could not recover them, since the healer only resets and restarts a unit and never rewrites one. The runtime switch now stops the site's running workers and recreates them against the new container in both directions, so they always exec into the container that actually exists.

A second issue compounded it. WorkerStartForSite rewrote a changed unit but never asked systemd to re-read it, so enable and start operated on the stale cached unit and a re-pointed worker kept failing on the old container even though the file on disk was already correct. It now runs a daemon-reload whenever the unit content changed, before enabling and starting, which also makes any on-demand worker start robust to a content change.