Now that we have dmeventd -R (for restart) upstream, it should be possible to implement a restart policy for dmeventd in the RPM's post-inst. Care needs to be taken about older versions, since dmeventd -R only works if the *running* dmevend supports the option as well. Killing old dmeventd versions unconditionally and restarting monitoring may be an option, although it's not very nice either.
...as for the rawhide, we have systemd now, so we should be fine with "ExecReload=/sbin/dmeventd -R" in dm-event.service systemd unit. If there's a new dmeventd running, everything's fine. If there's still the old dmeventd running (that does not support the -R switch), the "systemctl reload dm-event.service" fails with a warning that it was not possible to execute the reload and it keeps the original dmeventd running which is a correct behaviour. If called from within an rpm scriptlet (systemctl reload dm-event.service) on package update, we *should not try to kill* the daemon ourselves, otherwise we lose any existing registrations (we would need to call the unmonitor function and then start monitoring again directly for *all* registrations and that's not just the lvm2 case, but it could be used by other subsystems as well). Also, systemd would lose track of the daemon. A user should probably be advised to reboot his machine instead and start up a new dmeventd this way in this particular scenario. This would happen only once when the transition from old to new dmeventd (supporting the -R switch) is done so I don't think this poses a huge problem. As for the old SysV init system, the logic would be the same.
Finally included in the spec file and built - lvm2-2.02.95-2.fc18! It should have been done long time ago...