systemd improvements and better integration

[aquette]

Following the switch to systemd, a number of behavior regression have been identified and addressed (#200 and #208 particularly).

However, there are more visible improvements that can be addressed too, such as:

• ensure that systemd does well handle the UPS shutdown (upsdrvctl shutdown). This is not that obvious currently and is the main point of UPS protection systems! This probably requires a dedicated service file, with a "WantedBy: poweroff shutdown" or something alike.
  This is related to the nut-monitor.service (protection system)
  @bigon : could you please comment (i.e. have you made some tests on your side?).
• emitting SIGPWR and check sigpwr.target (http://www.freedesktop.org/software/systemd/man/systemd.special.html)
• to be completed...

[GreatEmerald]

Yes, please do!
As far as I can tell, the current way upsdrvctl shutdown is handled is... completely broken for any usbhid-ups systems. The nutshutdown script is executed right before shutdown - but nut-driver.service is stopped way sooner! It stops as soon as there is nothing left that requires it, and the only thing that requires it is nut-driver.service, which is turned off as soon as systemctl poweroff is called. That means the usbhid-ups drivers are turned off, and at shutdown upsdrvctl cannot do anything at all, since it no longer has communications to the UPS.

People on the systemd IRC channel suggested that the actually good way to implement this would be using a target (conveniently, there is sigpwr.target). upsmon should call systemctl isolate sigpwr.target; that will stop all services aside from the ones pulled in by said target. Then filesystems should be unmounted, upsdrvctl shutdown should be called (no need for any /etc/killpower files, the target would only be isolated when the UPS needs to be turned off anyway; in turn we don't need any mounted filesystems at all), and it would perhaps then proceed to final.target and poweroff.target (but not strictly necessary, as everything is off and unmounted at that point anyway, waiting for the UPS to shut down). That also prevents all power races.
Of course, that all will need quite a bit of testing to make sure it works right in all situations.

[clepple]

As far as I can tell, the current way upsdrvctl shutdown is handled is... completely broken for any usbhid-ups systems.

Are you sure of the failure mode? upsdrvctl shutdown expects the original driver to be stopped by the time it runs - it re-invokes the driver with usbhid-ups -k to skip any of the upsd connection code, and the driver simply sends the shutdown command.

Unfortunately, for a USB UPS, that means that /dev/bus/usb can't be unmounted at that point (but it shouldn't need to be unmounted, since it is a virtual filesystem).

[GreatEmerald]

I'm sure that it failed to turn off my UPS, at least. If it does expect the drivers to be turned off and turns them back on, then that's not really documented in the manual (and is quite counter-intuitive).