spawning smb processes out of control

[Jeremy Allison]

On Fri, Dec 07, 2001 at 11:59:29AM -0600, Alfredo Ramos wrote:
> Please help. I have a weird problem with samba 2.2.1a.
> 
> For the most part, Samba behaves just wonderfully. I'd say, 99.9% of the
> time it chugs along serving applications, sharing files and printers like
> a champ. But at least once a week, and some weeks, three or four times,
> the samba server starts spawning smbd processes like there's no tomorrow
> until it runs out of resources. At that point, users services come to a
> complete halt. Nobody is able to login, print, or do anything else. The
> load on the server becomes extremely high, and the only way to recover is
> by killing all the smbd processes that were spawned by the server. The
> number of processes spawned goes way past the 500 mark. If you're lucky
> and the server responds, you can kill the processes and everything goes
> back to normal. If the server is beyond recovery, only a reboot will get
> the server back. 
> 
> The server is a dedicated Sun Enterprise 220R with truckloads of memory,
> running Solaris 8. The weird part is that, the out-of-control spawning can
> happen when there's only a handful (8 or 10) of users, or when the labs
> are packed, sometimes during the day, at other times at wee hours of the
> night.
> 
> The logs don't show information that I could relate the this problem. I've
> set the debug level to 3 or 4, and the only thing that I was able to spot
> was a problem with the oplocks. Something to the effect that the server
> was waiting for an oplock to be released, and then receiving a
> notification without expecting it, also about oplocks, (sorry, I do not
> have the log output anymore. I think I can get it again though).
> 
> Anyway; thinking that this was the cause of the problem, I disabled the
> oplocks parameter on the smb.conf file, but it did not make a difference.

I think this is the contention on the share modes database that I 
believe we've fixed in the SAMBA_2_2 CVS tree. If you could test
that out I'd appreciate it.

Thanks,

	Jeremy.