Can't kill stranges smbd processes

[Jeremy Allison]

On Sun, Dec 09, 2001 at 04:47:48PM +0100, Jerome Le Tanou wrote:
> Hi,
> 
> We've got a file/print server based on Linux RH 7.1 with samba (kernel
> 2.4.3 and samba 2.0.10 recompiled) on IBM Netfinity 5100 (2 PIII733 /
> 1GB /  260GB Raid 5) for about 650 users (MS Windows 9x).
> We have a problem with unused smbd processes which have difficulty in
> disappearing, and which take all the cpu, while there are no more user
> opposite (the station is shutdown).
> 
> Sometime (often), the user touched by this phenomenon can not connect
> any more but a new smbd process is again launched to every attempt, and
> at the end we've got a server saturated by unused smbd process smbd.
> Furthermore when we try to kill theses processus with kill -9 sometime
> nothing is done and we must restart the server to clear up the mess
> (Once a week for about one month).
> 
> We tried to deactivate oplocks for 2 weeks but the situation did not
> change (kernel oplocks was not deactivated).
> 
> The problem, is that it is difficult to us to do more investigation seen
> the number of served customers. With such a load, the activation of logs
> is painful.
> 
> If anyone can offer some suggestions on how to troubleshoot this problem
> and suggest solutions... We would appreciate it.

This looks like a kernel problem (the failure to respond to kill -9
is a giveaway). 2.0.x doesn't do linux kernel oplocks so that's not
the problem.

I'd start by upgrading the kernel to the latest 2.4 stable, 2.4.3
is pretty old and has vm issues under load.

Once you've got a newer kernel on the box, monitor the situation
to see how it goes.

Jeremy.