On Wed, 25 Jan 2017 15:24:52 -0700
Chris Murphy <lists(a)colorremedies.com> wrote:
Thanks for the response.
Could be a labeling problem. Might be worth 'sudo restorecon -rv
/'
and then a sync and then reboot -f and see if it works with
enforcing=0 (or even without, let it enforce).
This made lots of changes, but didn't seem to affect the shutdown delay
problem. After the restorecon, the system delayed without setenforce 0
and with it, a few times. Some kind of race?
Stock Fedora 25 I have had a stop job for user 1000, that takes 90
seconds to time out. Your problem seems different. But to isolate if
this is a user environment problem, or a system problem, you could try
modifying /etc/systemd/logind.conf such that KillUserProcesses=yes
This was already in place.
(and remove the # so it's uncommented). The gotcha with this is
that
user processes are unceremoniously killed at logout time (including
restart and shutown). So if you use screen or tmux, and logout, the
tmux/screen processes get killed by virtue of tmux/screen getting
killed. There's a work around for that also which I can't articulate -
something to do with linger or gettting it to run in a separate logind
session? Anyway... just change this option temporarily for
troubleshooting purposes.
I changed some timeout options in the systemd .conf files
in /ets/systemd. It doesn't seem consistent. One time if I use
setenforce 0 there is no delay, and the next time there is. And there
aren't useful messages in the /shutdown-log.txt file that systemd is
creating. I changed journald to allow all output by setting burst to
0. The burst set to 0 didn't affect that there are lots of messages
elided by shutdown because of rate limiting. I'll keep playing with
that, maybe I can see what the problem is if I can see those.
After more than a half dozen reboots, time for a respite.