Bug#752259: systemd: User interaction regarding Ordering cycle found: D-Bus System Message Bus Socket

[Peter Gervai]

On Sun, Jun 22, 2014 at 9:20 PM, Michael Biebl <biebl at debian.org> wrote:
> Am 21.06.2014 17:41, schrieb Peter Gervai:

>> The loop was "broken" by using a rescue disk and _EMPTYING_ /etc/init.d/
>> completely. I'm not yet finished about figuring out which components were
>> causing the loop, but it wasn't "chkconfig" (tried that first) and were not
>> unessentials (I have tried to move them first without luck).
>
> Do you have a copy/backup around of /etc/init.d (or rather /etc/rc?.d/).
> This could give us a clue which service caused the dep cycle.

Yes, I'll try to sort it out in a few days.


> In summary, there are three issues:
>
> 1/ a (faulty) service causing a dep cycle
> 2/ the dep cycle resolver being broken under certain circumstances (this
> doesn't happen always).
> 3/ the different fstab behaviour.
>
> 2/ needs to be fixed in the actual package shipping that SysV or systemd
> service file. For that we need to identify that service.
> 1/ is still under investigation.

Should I try the one in experimental or it's better to play with the sid one?

> 3/ we intend to handle via the preinst check (even though it can't
> detect all cases of broken fstab lines)

Life is probably not that simple: I'm not sure whether bad modules in
/etc/modules (old clutter doesn't existing anymore) resulted the same
kind of emergency root shell or not. There seem to be really lot of
points where systemd considers the system broken: sysvinit simply
skipped these errors but systemd detects the failure and results fail
for the whole job which possibly fail its dependant jobs and result
the system unusable.

There should be
1) a strong warning about this at upgrading, briefly mentioning that
systemd will not skip a few kinds of failures at system startup, and
2) in case of a broken system where to look at first in that nice root shell.

People usually won't read upgrade notes anyway but a small chance is
better than nothing, and it could be included in README.Debian as
well.

-- 
 byte-byte,
    grin