[Pkg-utopia-maintainers] Bug#480552: fails to stop hald on upgrade...

[Bruce Sass]

On Sat May 10 2008 03:54:33 pm Loïc Minier wrote:
> On Sat, May 10, 2008, Bruce Sass wrote:
> > Most likely we are being bitten by a start-stop-daemon bug and you
> > may want to see how the exim Maintainer worked around the problem.
>
>  Yes; there are many bugs in ssd in particular conditions.  Could you
>  write more on the conditions under which you encounter this bug?

There is not much to write, it was a regular upgrade and hal was left 
unconfigured; it failed to configure during the next couple upgrades 
(expected sans a new hal package since there was no PID file), then 
there was a reboot which seemed to set the sys right (hald started and 
had a PID file in /var/run) but the package would still not configure 
with symptoms of the old process still running and no PID file (at 
which point I did the "killall", "--configure ---pending", and filed a 
report). 

The only strangeness with this setup is that /usr is an NFS mount (from 
an even less capable box, over a 100Mbps connection). I usually do 
daily upgrades, sometimes twice a day if a transition is in progress 
(e.g., like when a new PERL wants to remove 100+ packages :). All three 
boxes have been running Unstable for many years, this is the only one 
which has had problems with daemons failing to restart during upgrades. 
Dselect and its APT method are the only tools I ever use for updating 
and upgrading.

>  In my experience, the --exec flag to ssd on stop is particularly
>  fragile; it was already fixed for some use cases, but I'm pretty
> sure it will fail under e.g. unionfs.  Could you try without the
> --exec flag?

Sorry, no, I don't really have time to experiment with it in the short 
term right now and anytime I have done so the only significant result 
has been that "apt-get --reinstall install ..." rarely triggers the 
bug---afaict, I need a new package for each try and it should be part 
of a regular upgrade cycle. The exim Maintainer called it a "heisenbug", 
see #396944, he also filed #440657 shortly after but I don't know if it 
is related to the problem I was having or just something he ran across 
as he experimented. :(

However, if there are some specific bits to flip or some code I could 
wedge into the upgrade process to get better data/finer resolution re 
the bug I would be happy to to so for future upgrade cycles.


- Bruce