Time-critical problem at Sun: exploding smbd memory usage

Michael E Osborne mosborne at jacads.com
Tue Aug 21 00:25:11 GMT 2001


I don't know if this is related, but we once experienced a massive (10x)
increase in smbd (2.0.7) size which we eventually tracked down to a problem in
the smb.conf (a bad include statement) where we were causing the smb.conf to be
parsed over and over at smbd startup. This was under AIX 4.3.3. Eliminating the
re-parsing fixed the ballooning.





jeremy at valinux.com on 08/20/2001 12:17:57 PM

To:   farrar at parc.xerox.com
cc:   David.Collier-Brown at sun.com, Gerald Carter <gcarter at valinux.com>, Kris
      Desjardins <kris_desjardins at hotmail.com>, tonys at aus.sun.com,
      craig at aus.sun.com, allenw at sun.com, samba-technical at samba.org (bcc: Michael
      E Osborne/JACADS/REC)
Subject:  Re: Time-critical problem at Sun: exploding smbd memory usage



Keith Farrar wrote:
>
> It's not just the number of printers, it's the total number of shares. We
> have no printers defined, but lots of disk shares (roughly 900 on one box
> and 1500 on a second host). The servers are Sun E450s, but the same type
> of growth pattern occurs on Linux (redhat 7.1 x86).
>
> The smbd processes start up at 6 MB each, then grow until killed by
> process limits (currently 20 MB). Max observed growth is 115 MB... within
> an hour.
>
> The growth was much slower under 2.0.7, but happens quickly under 2.0.10+
> and 2.2 .
>
> It's not too hard to test: create an smb.conf file with 2000 static shares
> (squirt it out with a script & reuse the same directory path). Then watch
> the memory growth. Someone with familiarity with the code, access to a
> memory leak finder, and a good debugging environment should take a look at
> this (i.e. not me :).

Can you trigger growth by touching the smb.conf and
then hitting an smbd with a SIGHUP ?
If so, then it's smb.conf parse related.....

Jeremy.









More information about the samba-technical mailing list