On Thu, 2020-04-02 at 13:00 +0100, Daniel P. Berrangé wrote: > On Wed, Apr 01, 2020 at 08:53:39PM +0200, Andrea Bolognani wrote: > > There is nothing really systemd-specific about passing extra > > arguments to daemons so it's reasonable, although not currently the > > case, that startup scripts written for other init systems might want > > to source these sysconf files; for those init systems, which likely > > do not support socket activation, making the daemon quit after a > > timeout has expired is probably not a good idea. > > > > More generally, the sysconf files should not reflect the default > > behavior, but only contain overrides explicitly put in place by the > > admin; now that we have a mechanism to disable timeouts regardless > > of the default set in the service file, that argument for having the > > default timeout in the sysconf file is moot as well. > > The effect on this though is that --timeout arg now has to be > specified twice so we'll get a running process of > > "libvirtd --timeout 120 --timeout 0" > > which I find quite unappealing, so I'm not really in favour of > this revert, especially as we don't actually use the sysconf > files from other init systems I don't think it's a big deal, especially considering that most people will not end up actually changing the default, but I'm okay with flipping this around and moving --timeout from the service file to the ARGS variable in the corresponding sysconf file for all daemons instead, especially since Jano pointed out that a lot of sysconf files already look like that on a Fedora installation. Would that work for you? -- Andrea Bolognani / Red Hat / Virtualization
