[Samba] frequent tdb corruption

[Volker Lendecke]

On Thu, May 16, 2013 at 11:15:51AM +0100, Adam Thorn wrote:
> Hi Andrew,
> 
> > Can you please clarify:
> > 
> >  - Is the filesystem on this disk in any way shared?
> >  - Is the block device involved in any way shared?
> 
> I'm not 100% sure what you mean by "shared" in this context, but the
> filesystem where the tdbs are stored is: pair of local disks => mirrored
> together via zfs => zfs filesystem.
> 
> In case it's relevant: the samba config files are stored on a different
> filesystem. There are a set of multiple disks which are mirrored over
> the network with FreeBSD's HAST (pretty much equivalent to DRBD), and
> the HAST devices are then combined into a RAIDZ2 pool. I have recently
> had some disk problems with the filesystem where the samba config files
> are stored (but no problems with the filesystem which stores the tdbs) -
> is there any mechanism whereby e.g. a timeout in smbd trying to read
> it's config file could cause problems with the tdbs?
> 
> >  - Has the server ever had a unexpected poweroff?
> 
> No; I installed a different version of FreeBSD (downgraded from 9.1 to
> 9.0) a week ago to see if it made any difference. The OS partitions were
> totally wiped and reinstalled, and since then the server has been
> powered-up and stable, and still exhibiting corrupt tdbs.
> 
> >  - Do Samba processes ever crash?
> 
> No.
> 
> > If the answer is no to all these, then I would strongly suspect a
> > hardware or OS/kernel issue.  
> 
> I'd have said the same, but I've seen this problem on two (nominally)
> identical pieces of hardware (whilst that doesn't rule out hardware, I
> think it reduces the likelihood). I do plan to run memtest etc when I
> can, though.  Also, I'm running the same OS/kernel on another server
> which is *almost* identical hardware, and that's been completely
> trouble-free for over a year.
> 
> > Could you put your TDB files on a different file system, to rule in our
> > out ZFS (or the glue between FreeBSD and ZFS)?
> 
> I can certainly give that a go!

"use mmap = no" might provide another data point.

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de