[Samba] Hanging SMBD processes - Samba CRASHING

[Brian_Pickering at selinc.com]

I never was able to strace or delve too much into the debug logs when we 
were having problems, since about 100 developers were yelling at me any 
time it crashed.   The symptoms though were identical:  no new connections 
could be made, tons of smbd processes stacking up which wouldn't die 
except by kill -9, and lots of stale locks listed with smbstatus.  About 
the only thing I had time to do while I was compiling 2.2.12 was to stop 
the master smbd process, kill -9 the rest, remove the tdb files, and 
restart.  Hopefully I'll get a chance to do some more thorough testing 
with RHEL4 before it goes production. 

I'm running the IBM/Rational recommended samba settings, which also shuts 
off oplocks.

--------------------------------------------
Brian Pickering
System Administrator - Information Services
Schweitzer Engineering Laboratories, Inc.
Email - Brian_Pickering at selinc.com
Telephone - 509-332-1890 x1212




Matt Johnson <mwj at doc.ic.ac.uk> 
12/08/2005 11:58 AM

To
Brian_Pickering at selinc.com
cc
samba at lists.samba.org
Subject
Re: [Samba] Hanging SMBD processes - Samba CRASHING






On Thu, 8 Dec 2005, Brian_Pickering at selinc.com wrote:

> We've had similar troubles with Samba 3.x on our ClearCase VOB server
> running RHEL3.  Our fix was to go back to the old 2.2.12, and we haven't
> had a problem since.  Unfortunately I was never able to devote enough 
time
> to tracking down the problem fully.  I had hoped that upgrading to RHEL4
> using a 2.6 kernel would help, but your experience doesn't bode well for
> that.

We had similar problems. stracing hung child smbd processes pointed the 
finger at *local* locking issues: processes were hanging in 
fcntl(...,F_SETLKW64,...) for one byte locks on fd 14 = locking.tdb; the 
culprit function is tdb_brlock.

Turning off oplocks seems to have helped, but we'd also like to get to 
the bottom of this... is this fcntl/locking hang behavior what you are 
seeing too?

Matt

-- 
======================================================================
Matt Johnson <mwj at doc.ic.ac.uk>               (020) 7594 8440 / x48440
Systems Programmer, Computing Support Group         Office: Huxley 225
Department of Computing, Imperial College London
======================================================================


----------------------------------
This e-mail may contain SEL confidential or 
legally privileged information.  The opinions 
expressed are not necessarily those of SEL.  
Any unauthorized disclosure, distribution or 
other use is prohibited.  If you received this 
e-mail in error, please notify the sender, 
permanently delete it, and destroy any 
printed copies.