tdb_lock failed (was Re: ... domain authentication?)

[Jeremy Allison]

On Mon, Oct 27, 2003 at 03:29:15PM -0500, Brandon Craig Rhodes wrote:
> Following your advice, Andrew, I was able to convince our admins that
> we should simply throw all of our servers into the same domain and not
> trying to keep the domains separate.  Thanks!
> 
> Now we are having extensive problems with performance in the labs we
> are attempting to move to domain authentication under Samba-3, because
> of contention over the secrets.tdb file from which each thread must
> now fetch the SID for our domain controller; errors look like:
> 
>  tdb_chainlock_with_timeout_internal: alarm (10) timed out for key <PDC> in tdb /usr/local/samba-3.0.0/private/secrets.tdb
> 
> where <PDC> is the name of the primary domain controller.
> 
> This is happening in two different labs under both Solaris 2.7 and 2.8
> and renders samba-3 essentially unusable.  In one lab we have compiled
> Samba with tdb spin locks instead, which solves the problem, but (if I
> understand the relevant warnings correctly) makes us vulnerable to an
> unusable secrets file if a thread dies while having the file locked.
> 
> The /usr/local hierarchy is locally mounted on both systems.
> 
> Has anyone else ever seen this?  Meanwhile I suppose I shall try
> having to determine who is holding the problem locks.  The lsof
> command shows every samba child process (but not the parent) having a
> read lock on the secrets file, but does not give details as to where
> in the file they are.

Ok, do  you have the Sun patch for the "thundering herd" problem with
fcntl locks ? This is common on Solaris systems, they have a problem
with wakeup of many threads waiting on an fcntl lock.

There are references to the patch ID in the Samba mail archives (on
Samba-technical I think).

Jeremy.