On Tue, 21 Sep 2021 at 09:27:04 +0200, Fabian Grünbichler wrote: > also started to run into this, this breaks two not that uncommon > use-cases rather badly: > > - services that support hot-reloading (via ExecReload) without service > interruption, as opposed to restarting which does > - services that don't support hot-reloading, but where restarting has > dangerous side-effects and should be done manually (e.g., hypervisor > services where restarting means stopping all running guests, services > that have long-running tasks that would get interrupted and start over > or get lost when restarted, ...)
The D-Bus well-known system bus in the dbus package is a particularly prominent example of the second use-case. As far as I can tell, next time dbus/unstable or dbus/experimental gets binNMU'd, they will break desktop systems (terminate NetworkManager, abruptly log out from GNOME, etc.) when upgraded. This seems like an unexpected time-bomb. I'm going to try to work around this in dbus by using --no-start, and starting the service from the handwritten parts of the maintainer scripts instead. smcv