[Samba] LVS and SAMBA

[Jim McDonough]

On Thu, Nov 18, 2010 at 3:56 PM, Michael Adam <obnox at samba.org> wrote:
> Ciro Iriarte wrote:
>> 2010/11/3 Volker Lendecke <Volker.Lendecke at sernet.de>:
>> > On Wed, Nov 03, 2010 at 11:03:20AM -0300, Ciro Iriarte wrote:
>> >> Hi, would it be possible to  run two nodes with SAMBA+LDAP and a OCFS2
>> >> filesystem, with LVS load balancing WITHOUT CTDB?. This would be
>> >> relying only on OCFS2 file locking. The idea is to provide
>> >> authentication, HA file service and load balancing.
>> >
>> > That would cause data corruption. As far as I know (please
>> > correct me if I'm wrong) OCFS2 does not support the full
>> > semantics required for share modes, oplocks and all the
>> > other fancy cifs features that Samba provides. That's the
>> > point of the ctdb and clustered samba combo.
>> >
>> > Volker
>> >
>>
>> Well, the new "cool" feature on OCFS2 is fcntl, apparently is not
>> enough. With CTDB I see that nodes use the same netbios name. How
>> would that affect domain controller feature?. I won't like to build
>> other server pair just for PDC/BDC....
>
> When you think a little about it you will agree that
> when serving the same folder from a cluster file system
> like ocfs2, as mounted on two different nodes, as
> samba shares from these nodes, then you will have to
> configure the two sambas to use the same netbios name.
> Not only that, but you will also have to make sure that
> the windows SID --> unix ID mappings are identical.
> (Unless, of course, you like data corruption. ;-)
> And so on.
>
> The typical use case for a clustered samba with ctdb
> is a file server that is member in a domain, but you
> can in principle also run samba as a clustered Domain
> controller on the cluster. THis will just be one DC
> then (since the samba's on the nodes will have to
> appear as one server together), but this way you could
> replace the PDC/BDC replication, failover and load
> balancing mechanism of the classical PDC/BDC scheme
> buy using e.g. one clustered PDC. This could even use
> tdbsam then (instead of ldapsm), since replication
> is done by CTDB then.
>
> I personally have not done such a setup. But it should
> not be a big problem. Be sure to use the "cluster addresses"
> variable in smb.conf here to make nmbd happy.
> Maybe Jim McDonough can share some of his experiences here? :-)

I think you've basically covered it, Michael.

This setup is supported on SLES11+HAE as described here.  Ok, we're
reworking the ctdb resource agent, but if you run ocfs2 under the HA
stack, and ctdb outside this, it works just fine.  A new resource
agent is coming soon, or if you've got support on SLES11SP1+HAE,
please contact Novell support.

-- 
Jim McDonough
Samba Team
SUSE labs
jmcd at samba dot org
jmcd at themcdonoughs dot org