Bug 781657

Oh, when booted to level 1 then both shutdown and reboot work without any issues. Only when starting a machine in a way one could want it possibly use when this goes wrong.

Try adding these two lines to /lib/systemd/system/syslog.socket:

Conflicts=shutdown.target
Before=shutdown.target

(as in http://cgit.freedesktop.org/systemd/systemd/commit/?id=ead51eb4ed55981f290e40a871ffbca6480c4cd3)

... and don't forget to do 'systemctl daemon-reload' before you reboot after editing.

(In reply to comment #2)
> Try adding these two lines to /lib/systemd/system/syslog.socket:
> 
> Conflicts=shutdown.target
> Before=shutdown.target

Yes, this makes a substantial difference.  Thanks. After "Shutting down interface ..." line I see now

Sending SIGTERM to remaining process

and the rest of shutdown or reboot follows as expected.

I still think that it is too easy to get something wrong here and the whole design is too brittle.  I had cases (not understood, not repeatable) when systemd died for unfathomable reasons and a machine was sort of running but becoming impossible to reboot other than by pulling a plug.

syslog-related problems are often nasty. Try sending a SIGSTOP to the syslog daemon sometime ;-)

When running Rawhide, I definitely recommend having SysRq enabled for the Alt+SysRq + {S,U,B} emergency combination.

(In reply to comment #5)
> When running Rawhide, I definitely recommend having SysRq enabled for the
> Alt+SysRq + {S,U,B} emergency combination.

In theory this should be somewhat different from "pulling a plug"; in practice not necessary. :-)

(In reply to comment #5)
> syslog-related problems are often nasty. Try sending a SIGSTOP to the syslog
> daemon sometime ;-)
> 
> When running Rawhide, I definitely recommend having SysRq enabled for the
> Alt+SysRq + {S,U,B} emergency combination.

This is new to me
A pointer?

(In reply to comment #2)
> Try adding these two lines to /lib/systemd/system/syslog.socket:
> 
> Conflicts=shutdown.target
> Before=shutdown.target
> 
> (as in
> http://cgit.freedesktop.org/systemd/systemd/commit/?id=ead51eb4ed55981f290e40a871ffbca6480c4cd3)

I did this, then ran "systemctl daemon-reload" before rebooting, and it worked the first time, but not any more. Now it hangs during shutdown as before.

(In reply to comment #7)

> > When running Rawhide, I definitely recommend having SysRq enabled for the
> > Alt+SysRq + {S,U,B} emergency combination.
> 
> This is new to me
> A pointer?

Ahem!
https://www.kernel.org/doc/Documentation/sysrq.txt
https://fedoraproject.org/wiki/QA/Sysrq
https://en.wikipedia.org/wiki/Magic_SysRq_key

If Wikipedia has an article about it then it cannot be that new. :-)

Trust me, we had no pc's at school, many still don't.
I'm almost middle-aged and on catch-up. :)

(In reply to comment #8)

> I did this, then ran "systemctl daemon-reload" before rebooting, and it worked
> the first time, but not any more. Now it hangs during shutdown as before.

I reverted the edit, then hard powered off, then did the edit and systemctl daemon-reload command again and rebooted, and it seems to work persistently this time. Only difference is that I rebooted rather than powering off after originally making the changes (normally I poweroff each day). I don't see anything in the systemctl man page indicating that should make a difference.

I'm sure the original edit was correct, since as I said it worked the first time, and the edit itself was persistent (though not its effect).

Sorry, meant to say that I powered off after making the original changes (which only worked once). This time, I rebooted instead.

Same problem with systemd-38-4.fc17, same fix works (comment 2 and comment 3).

Fixed in systemd-38-6.git9fa2f41.fc17.