Comment 15 for bug 1670291

@xnox Scheduling with /sbin/shutdown normally does work ... however, what we're dealing with here is the anomalous situation of an instance which starts off as trusty and is upgraded to xenial. During the upgrade process, the /sbin/shutdown implementation is changed from the upstart one to the systemd one. At the end of the process, /sbin/shutdown is (unsurprisingly) a link to systemctl ... but it doesn't work because it's expecting to be able to talk to systemd which isn't fully operational until a reboot. As Landscape is expecting to be able to reboot via /sbin/shutdown the user relying on Landscape in such a scenario finds themselves in a bit of a bind (having to have some other means of dealing with the server, like SSH).

However, /sbin/reboot and /sbin/poweroff *do* work in this (admittedly rare) scenario, hence the patches committed for this issue leave the system using /sbin/shutdown by default but if/when it fails, they fall back to trying /sbin/reboot or /sbin/poweroff (as appropriate) instead.

As to why -h is used instead of -P, in systemd's implementation of /sbin/shutdown they're exactly the same thing. In the upstart implementation it's more vague; quoting from the man-page: "Requests that the system be either halted or powered off after it has been brought down, with the choice as to which left up to the system". I'm not sure what that means in practice though!