Memory Leak

[John Bollard]

OK, first of all , thanks for all the responded.

Here is what we have.  

Previous Environment:  
Sun E3500
4 CPU's 
4 Gig Ram
A5200 Array
Solaris 2.6
samba 2.0.0 compiled with gcc 2.8.1

Using top to monitor performance, we continually saw
2 smbd processes continually consume memory and processor
time.  The 'biggest' one grew to over 2 Gig (size) and 1.7 Gig (Res)
and was using about 8-10 % of processor time.  The process did not
show up in the output of an smbstatus.  The second one was always
smaller but still growing, and did actually show up in smbstatus.
The biggest one would eventually die, the second one would then
begin to grow faster and would no longer show up in smbstatus.  
And a new one would start to show up in top and begin to grow as 
well.  And the cycle continued.....

Environment Changes:
upgraded gcc to version 2.95.2
upgraded samba to version 2.0.7 using gcc 2.95.2 to compile

We have now run a full day of production, and life is good.
No smbd processes in top (except for the occassional normal
one or two with a Size of 6 Meg and Resident Size of about 4 Meg.)

I will be on-site again tomorrow, and if anything changes, I will
let you know.

BTW:  They do have over 100 printers setup on the system, but
they are  not using printcap.

Thanks again!

John

On Tue, 12 Dec 2000, Tim Potter wrote:

> Gerald Carter writes:
> 
> > > John,
> > > We have had the same problem here with a build with gcc 
> > > 2.8.1, and, I'm sorry to say, building it instead with 
> > > gcc 2.95.2 did not appear to have any effect. We are 
> > > running samba 2.0.7, and have about 1000 printers
> > > shared out. It does certainly appear to be related 
> > > to the print subsystem. We restart our smbd processes every 
> > > night to contain the leak (as well as restarting out nmbd 
> > > process to ensure registration with our WINS servers, but 
> > > that's a separate problem). The restart only takes a
> > > second or two and is at 2 AM, so it doesn't impact service.
> > 
> > There were some patches for this (printing memory leaks).  
> > Check the samba-technical mailinglist archives and the 
> > patch listing possibly.
> 
> The URL I have for this patch is:
> 
> http://lists.samba.org/pipermail/samba-technical/2000-September/008961.html
> 
> 
> Tim.
> 
>