Slow shutdown in dependency-based boot system

[Kel Modderman]

On Tuesday 15 July 2008 18:46:47 Petter Reinholdtsen wrote:
> 
> [Sven Joachim]
> > Ten days ago I installed the insserv package and enabled dependency
> > based booting.  While this works quite well, the shutdown of the system
> > has slowed down noticeably, the main reason being that sendsigs now takes
> > six seconds to terminate the remaining processes while it previously
> > finished in one or two seconds.
> 
> Hm.  Can you add a call to 'ps -ef' in sendsigs, to report what
> processes are being killed?  I see nothing obviously wrong with the
> shutdown sequence.  Are you using network file systems?  Perhaps
> wpa-ifupdown need to run next to networking and ifupdown, and register
> its pid to the list of pids to survive sendsigs?  I talk about this
> part of the sequence:
> 
> > | K04wpa-ifupdown
> > | K05sendsigs
> > | K06umountnfs.sh
> > | K07nfs-common
> > | K08portmap
> > | K09networking
> > | K10ifupdown
> 
> I do not know what the wpa-ifupdown script is doing, so I have no
> better advice.

It cleanly shuts down wireless interfaces using wpa-roam feature of
wpasupplicant/ifupdown. It runs before sendsigs because it was in existance
before sendsigs ommiission interface. Etch did not have ability to omit pids
from sendsigs, so it was needed.

wpasupplicant daemons do now add pids to /lib/init/rw/sendsigs.omit.d/, but I
am reluctant to move the position of wpa-ifupdown still.

> 
> > Any idea who could be the culprit?  I see that K10ifupdown and
> > K12mountoverflowtmp are new, while K07nfs-common and K11hwclock.sh
> > had been run before sendsigs in the old system.
> 
> I doubt the K12mountoverflowtmp script is the culprit.  nfs-common
> might be related, but it seem unlikely.  ifupdown might be related,
> but it seem likely as it only clean up /etc/network/run/ifstate.
> 
> Happy hacking,


Thanks, Kel.