Hi Stephane, > I am just not convinced with your workaround. > The two signals are from a different nature and we have them both for a > reason. Well, given that SIGPWR is the default signal sent on halt by lxc-stop, making systemd handle that signal by shutting down seems like a solution, not a workaround? Of course, SIGPWR is not technically correct, but SIGRTMIN+4 is also just arbitrarily defined by systemd (though that uses a signal without predefined meaning, instead of somewhat abusing an existing signal, so SIGRTMIN+4 might be cleaner indeed) . The advantage of SIGPWR is probably that regular init also responds to it correctly.
In any case, adding the link I suggested at least makes upgraded containers work in the same way as new containers. I can see your point about your solution being technically better. Initially, SIGRTMIN+4 looks like a hack of sorts, but reading the systemd manpage I see it just defines that as the "API" for shutdown. > I would be more expecting someone to identify where is the difference > and correct it, I guess on the upgraded lxc container itself. Yeah, that would probably be an option, since the lxc package needs to be installed in the container as well. However, inside the container, you can only fix the sigpwr symlink, not change the lxc config to use SIGRTMIN+4, so that would be a reason to use the current approach. I'll leave it up to the lxc maintainer to see what, if anything, needs to be done, but at least both approaches are documented in this bug report now. Gr. Matthijs
signature.asc
Description: Digital signature
