TDB corruption with Samba 2.2.3a

Steve Lord lord at sgi.com
Wed Mar 20 07:04:13 GMT 2002


On Wed, 2002-03-20 at 01:27, Martin Apel wrote:
> On 19 Mar 2002, Steve Lord wrote:
> 
> > On Tue, 2002-03-19 at 01:03, Martin Apel wrote:
> > > On Wed, 13 Mar 2002, Jeremy Allison wrote:
> > >
> > > > On Wed, Mar 13, 2002 at 05:25:13AM -0500, ZINKEVICIUS,MATT (HP-Loveland,ex1) wrote:
> > > > > We are also experiencing the TDB corruptions, as reported in the samba log
> > > > > files. At first there didn't seem to be an consequence, but we are now
> > > > > beginning to see user's not being able to login to the machine. As a
> > > > > workaround we have been just deleting the secrets.tdb file, restarting
> > > > > samba, and rejoining the domain.
> > > > >
> > > > > Our server if very similar to Martin's (Samba 2.2.3a + Linux 2.4.17 + XFS).
> > > > > I'll post more details soon.
> > > >
> > > > Can you try using the tdbbackup utility periodically
> > > > to determine when the corruption may be occurring ?
> > >
> > > I let the tdbbackup run for a few days now. The TDB corruption seems to
> > > happen at the time, when Amanda (a great backup tool) starts to run.
> > > I have moved the Amanda start time back and forth and the corruption
> > > starts within 10 minutes after starting Amanda.
> > > I don't think it's Amanda's fault, I assume that Amanda puts a heavy load
> > > on the filesystem layer during the first minutes, when it does its estimates.
> > > A reminder: all partitions on this system are XFS partititions, including
> > > /var, where Samba stores the TDB files.
> > > I could try to reformat the /var filesystem with ext2 to see if this has
> > > any influence. But this will probably need a server reboot, so I cannot
> > > do this before the weekend.
> > >
> > > Martin
> >
> > Can you try a kernel from the 2.4 xfs cvs tree - I just pushed some
> > changes out there which fix a corruption problem under heavy memory
> > pressure.
> 
> Hi Steve,
> 
> since this is our production server with 50 people being dependent on it
> I would rather not try the current CVS version. Is it possible to isolate
> your patch relative to XFS 1.0.2?
> For the record: I put the Samba lock directory onto an ext2 partition
> yesterday and had no problems since then.
> 


Well, the simple answer is no. Since the code hits fs/buffer.c which has
changed a lot. However, we are working on pushing current xfs code back
into a 2.4.9 kernel in rpm format.

Steve


-- 

Steve Lord                                      voice: +1-651-683-3511
Principal Engineer, Filesystem Software         email: lord at sgi.com




More information about the samba-technical mailing list