samba - Sep 2009 - random samba failures

Hi,

I don't know if I'm posting to the right list, so ... please, let me  
know.

We have a samba 3.4.0 running on Linux SLES 11 which is using winbind  
to authenticate to a Windows server 2008.
Everything was working flawlessly (July) until samba started to hang  
randomly (september). When this happens, already connected client are  
still accessing their network shares but any new client gets rejected  
and a defunct smbd process appears in the process list on the server.
At that point we have to stop/start samba server and kill zombi  
processes or shutdown/restart the server.

Clients OS are Windows 2000, XP and Vista and also Gentoo Linux  
(2.6.24 kernel) or Fedora 11 Linux (2.6.29 kernel).

failure isn't occuring after a regular period of time, and doesn't  
seem related to a number of users. Also neither cpu nor memory charge  
happens to be critical at the time of failure.

We also observed that winbindd process is still alive and ready when  
defunct smbd process are growing.

We suspect a kind of client OS to be responsible but it's just a  
thought.

we didn't find any useful information in the log files (smbd,  
winbindd, client) even if the debug level was raised to 3.

The problem is that we know when the first client gets rejected, but  
not when the originating error occurs.

if you have any advice to narrow what's causing this issue, please help!

I can provide any config or log file along with every details for any  
tests you may think of.


thanks for your help

Alban

--
Alban Rodriguez
Centre de Ressources Informatiques
Universit? de La Rochelle
http://cri.univ-lr.fr

> we didn't find any useful information in the log files (smbd, winbindd,
> client) even if the debug level was raised to 3.
> 
> The problem is that we know when the first client gets rejected, but not
> when the originating error occurs.
> 
> if you have any advice to narrow what's causing this issue, please
help!
> 
> I can provide any config or log file along with every details for any
> tests you may think of.
The only thing I can suggest is to attach to one of the hung smbd
servers with GDB, and see if you can find out where precisely it has
hung.  That might at least give you a clue as to what it's waiting for.

Cheers,
Adam.