Hi, almost as expected by some on the bug here this now hits libvirt as well. Recent builds will shutdown guests (via libvirt-guest.service) and fence guests (via virtlogd.service) now. This is due to their postinst having replaced the former safe `start` with `restart` for `--no-stop-on-upgrade` units due to this bug here.
For now I had to follow the example of dbus and try to handle the system services that shall not be restarted ourselves (proposed in [1]). The libvirt case even seems slightly worse, as the bug described here also affects `.socket` files which - even if the service isn't restarted - will implicitly re-cycle the related `.service` and thereby cause the same effect. Maybe I'm not deep enough in the reasoning that led to adding this new behavior. But from a bit away it seems wrong that using the `--no-stop-on-upgrade` option no more allows you to "not restart" services through debhelper anymore. And furthermore if we acknowledge that there are valid cases that do not want to be stopped/restarted ever but keep the current `--no-stop-on-upgrade` handling as-is, what would a new option look like '--really-no-stop-on-upgrade'? P.S. Thanks to Mbiebl for helping me to spot this known bug when I was confused about my dying guests and service behavior when testing our new libvirt package builds. [1]: https://salsa.debian.org/libvirt-team/libvirt/-/merge_requests/132 -- Christian Ehrhardt Staff Engineer, Ubuntu Server Canonical Ltd
