[Samba] Samba goes catatonic...?

[Ryan Beisner]

Guten Tag!

I have Samba 225 on a network of 75-100 daily users.

System has (4) SCSI drives:  (2) 18GB (RAID-1)  and   (2) 36GB (RAID-1)
on two different SCSI channels.  

The system is running Red Hat 7.3 with updated and recompiled Samba 2.25
package.  

After one day of operation, the SMBD process is listed in "ps -A" ten to
fifteen times, and some client PCs cannot connect.  There is no rhyme or
reason as to which clients cannot connect ... some are 98, 98SE, 2000,
XP, and XP SP1.

The thing that kills me is that while some cannot access it via network
neighborhood, those same clients can open a DOS shell, and run "ping
bigserver" and get 100% replies at 1 to 2ms;  whilst other clients
haven't noticed any problems via nethood at all.

OK?  So something is eating up the system.  Turns out that when theses
symptoms surface (daily), LOGROTATE is consuming around 90% CPU in
"top."  Killing all "top" processes and issuing "service smb restart"
resolves the problem of not being able to connect (from only random
workstations).

This machine is running as PDC to two NT 4 SP6a boxes.  When the Samba
server goes funky, and I reboot one of those NT servers, it gives a
message about not seeing any PDC and that it is about to use info out of
cache (as can be expected with no PDC).

I've set the DEBUG LEVEL to 1 (was at 3), and we still see these
problems.  The SAMBA log folder (/var/log/samba) would reach 22MB in one
day at level 3.

Looking through those logs doesn't tell me anything in particular:  it
lists failed authentications by some users, as expected, files that
weren't found, as expected.  Nothing alarming (I think) is listed here.


QUESTIONS:

--- Is it normal to see several SMBD processes in a ps -A command?
--- What else can help indicate what may cause this Samba server to go
catatonic?



All help and advice is greatly appreciated.  Thanks in advance!

-Ryan Beisner














*/ The source is indeed with me. */