messages.tdb growing to 4GB (4294971392 bytes)

[John H Terpstra - Samba Team]

Volker Lendecke wrote:
> On Wed, Jun 03, 2009 at 11:47:24AM -0500, John H Terpstra - Samba Team wrote:
>> Has anyone else come across situations where messages.tdb grows to 4GB?
> 
> No, but a very likely reason is that an smbd holding a
> shared file is stuck in gfs in state D.

The underlying file systems are XFS and Ext3 - this is not on a cluster
- the system is an ADS domain member server.

There are a lot (up to 100) smbd's that are apparently in state D after
about 2 hours operation.  The only recovery is to shutdown smbd, delete
messages.tdb and restart smbd.  Very messy and not good for productivity.

> 
>> It seems smbd is no longer able to access this file after it reaches
>> this size.
>>
>> -rw-------  1 root root   4294971392 2009-06-03 09:39 messages.tdb
> 
> Yep, 4GB is a known limit for tdb, it uses 32 bit pointers
> internally. Tridge right now is looking at expanding that.

Understood.

> Search for stuck smbds, reboot those nodes and file a bug
> with RedHat. We've had quite a few of those supporting
> "wakku" on the #ctdb irc channel the last few days.

The OS is Fedora 9 - not RHEL.  Same problem with Fedora 10.  I have not
(yet) seen this problem with RHEL5.3 or on OpenSUSE/SLES.

Apologies for not mentioning the OS in my original posting.

> You might want to comment out the call to
> SMB_VFS_KERNEL_FLOCK() in smbd/open.c, gfs2 seems very
> unhappy with that call, leading to processes in state D
> forever.

Would this still apply for non-clustered systems?

Thanks for responding.

Cheers,
John T.