Networkmanager service is shutdown too early

[Dan Williams]

On Wed, 2008-06-04 at 17:20 +0200, Valent Turkovic wrote:
> On Wed, Jun 4, 2008 at 4:04 PM, Dan Williams <dcbw at redhat.com> wrote:
> > On Wed, 2008-06-04 at 09:42 +0200, Valent Turkovic wrote:
> >> Dan Williams wrote:
> >> > On Fri, 2008-05-30 at 19:47 +0100, Kostas Georgiou wrote:
> >> >> On Fri, May 30, 2008 at 01:46:42PM -0400, Dan Williams wrote:
> >> >>
> >> >>> On Fri, 2008-05-30 at 13:41 -0400, Bill Nottingham wrote:
> >> >>>> Dan Williams (dcbw at redhat.com) said:
> >> >>>>> On Fri, 2008-05-30 at 10:54 +0200, Valent Turkovic wrote:
> >> >>>>>> During the boot I have some samba shares mounted because I have them
> >> >>>>>> configured to mount via fstab file.
> >> >>>>>>
> >> >>>>>> When I shutdown or reboot I get a screen for 2-3 minutes that shows
> >> >>>>>> smbfs service trying to unmount samba shares but NM service has
> >> >>>>>> already shutdown and there is no working network connection :(
> >> >>>>>>
> >> >>>>>> I have seen this "bad" behaviour in F8 and have reported it on this
> >> >>>>>> mailinglist, but I hoped that the new and smarter NM would take care
> >> >>>>>> of it, but unfortunately it didn't :(
> >> >>>>> Probably need to adjust the stop priorities of NM and haldaemon to be
> >> >>>>> right after messagebus (K85) rather than where they currently are...
> >> >>>>> The problem is that NM is being stopped to early.
> >> >>>> 'After netfs' should be good enough. Although netfs stop should possibly
> >> >>>> do lazy umounts.
> >> >>> Ok, just need to bump NM a few bits later it looks like; might as well
> >> >>> be K84 to be right after messagebus.
> >> >> Why not go all the way to 90 as network to be on the safe side? With a
> >> >> quick look I can see some scripts that might not be happy if the network
> >> >> is down with a priority above 84, racoon/dund/rdisc/rpcgssd/nasd for example.
> >> >
> >> > NM depend on messagebus at least, so we should stop NM right before
> >> > messagebus.  But the same issues with startup are theoretically present
> >> > with shutdown, meaning that since messagebus depends on rsyslog, and
> >> > rsyslog depends on network.  Standard installs don't use networked
> >> > syslog, so standard installs don't actually need rsyslog to depend on
> >> > network, but because rsyslog isn't smart enough to know when it does or
> >> > does not depend on network, we can't just re-order the chain... :(
> >> >
> >> > Dan
> >> >
> >>
> >> Dan what is the conclusion about this bug? This is a looong thread but
> >> nothing is updated on bugzilla page so is there some consensus on what
> >> needs to be done?
> >
> > I will move NetworkManager shutdown to be right before dbus.
> >
> > Dan
> 
> Will this happen as an Fedora 9 update or just in rawhide (fedora 10) ?

Definitely in the next F9 update.

Dan