samba - Feb 2014 - sysvol replication via clustered/distributed file system

hi there,

the samba wiki currently recommends using rsync to achieve sysvol
replication in an ad dc setup with more than one dc.

this seems a bit kludgy to me ? using some sort of a distributed file
system would probably be a more elegant solution. while researching
this, i came across a few threads on this list, but unfortunately none
of them provides a definite answer to which setup works to what extend
and what is to be avoided.

i guess that the following preconditions have to be met (please correct
me if i'm wrong):
* posix acls have to be supported on the fs level
* during writes, files must be transparently locked on the fs layer
* read- and write-operations on the shared file system should be
non-blocking, even if a node/brick is temporarily not available or goes
down during reads/writes

glusterfs, lustre and even drbd with something appropriate on top might
be feasible options ? it would be really great if we could share some
experiences here.

many thanks in advance,
thoralf.
-- 
thoralf schulze
deutsche kinemathek - museum f?r film und fernsehen
linux-administration / helpdesk
tschulze at deutsche-kinemathek.de / 030 - 300 903-531

On Tue, 2014-02-04 at 12:38 +0100, Thoralf Schulze
wrote:
> hi there,
> 
> the samba wiki currently recommends using rsync to achieve sysvol
> replication in an ad dc setup with more than one dc.
> 
> this seems a bit kludgy to me ? using some sort of a distributed file
> system would probably be a more elegant solution. while researching
> this, i came across a few threads on this list, but unfortunately none
> of them provides a definite answer to which setup works to what extend
> and what is to be avoided.
> 
> i guess that the following preconditions have to be met (please correct
> me if i'm wrong):
> * posix acls have to be supported on the fs level
> * during writes, files must be transparently locked on the fs layer
> * read- and write-operations on the shared file system should be
> non-blocking, even if a node/brick is temporarily not available or goes
> down during reads/writes
> 
> glusterfs, lustre and even drbd with something appropriate on top might
> be feasible options ? it would be really great if we could share some
> experiences here.
The issue is that using a cluster file system with Samba requires CTDB
to be able to do it safely, and CTDB and the AD DC are incompatible, so
this isn't a useful way forward.

What we need is for someone with a lot of time and talent to step up and
take on the DFS-R work.  We have some progress here, but much more needs
to be done. 

Sorry,

Andrew Bartlett

-- 
Andrew Bartlett
http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba