samba - Aug 2016 - Samba with CTDB able to serve non-clustered data?

Hi,

I'd like to set up a HA Samba server that serves out data from NFS mounts​.
I was thinking that CTDB was the thing to do use but it is only mentioned
with serving clustered file systems like Gluster, Lustre, etc. Is there a
way to set up a HA Samba server without a clustered file system?

​Thanks,

Steve​

Hi Stephen,

On Tue, 9 Aug 2016 18:08:15 -0400, Stephen Cousins via samba
<samba at lists.samba.org> wrote:
> I'd like to set up a HA Samba server that serves out data from NFS
mounts​.
> I was thinking that CTDB was the thing to do use but it is only mentioned
> with serving clustered file systems like Gluster, Lustre, etc. Is there a
> way to set up a HA Samba server without a clustered file system?
You don't need a cluster filesystem to use CTDB.

Cluster filesystems are mentioned in relation to CTDB for 2 reasons:

* Coherency of the file data you serve with Samba

  You need to make sure that updates to the data that you are serving
  are seen coherently across the cluster.

* Split brain protection

  CTDB recommends using a recovery lock file in a cluster filesystem.
  You don't have to use this but without it you aren't protected
  against your cluster being partitioned (a.k.a. split brain).

  From Samba >= 4.5, CTDB will support a configurable helper to handle
  the recovery lock.

If you can satisfy these requirements then you're good to go...  the
important thing is to test... :-)

The ping_pong tool can test a bunch of the necessary features:

  https://wiki.samba.org/index.php/Ping_pong

Good luck...

peace & happiness,
martin

Hi Martin,

Thanks very much for the information.

I guess the part:

* Coherency of the file data you serve with Samba

  You need to make sure that updates to the data that you are serving
  are seen coherently across the cluster.

Is the tough one. Is there a way to do this that you know of when using NFS?

I see version 4.5 RC2 was just released. Do you know when it will be fully
released?

Thanks again,

Steve

On Wed, Aug 10, 2016 at 7:07 PM, Martin Schwenke <martin at meltin.net>
wrote:
> Hi Stephen,
>
> On Tue, 9 Aug 2016 18:08:15 -0400, Stephen Cousins via samba
> <samba at lists.samba.org> wrote:
>
> > I'd like to set up a HA Samba server that serves out data from NFS
> mounts​.
> > I was thinking that CTDB was the thing to do use but it is only
mentioned
> > with serving clustered file systems like Gluster, Lustre, etc. Is
there a
> > way to set up a HA Samba server without a clustered file system?
>
> You don't need a cluster filesystem to use CTDB.
>
> Cluster filesystems are mentioned in relation to CTDB for 2 reasons:
>
> * Coherency of the file data you serve with Samba
>
>   You need to make sure that updates to the data that you are serving
>   are seen coherently across the cluster.
>
> * Split brain protection
>
>   CTDB recommends using a recovery lock file in a cluster filesystem.
>   You don't have to use this but without it you aren't protected
>   against your cluster being partitioned (a.k.a. split brain).
>
>   From Samba >= 4.5, CTDB will support a configurable helper to handle
>   the recovery lock.
>
> If you can satisfy these requirements then you're good to go...  the
> important thing is to test... :-)
>
> The ping_pong tool can test a bunch of the necessary features:
>
>   https://wiki.samba.org/index.php/Ping_pong
>
> Good luck...
>
> peace & happiness,
> martin
>


-- 
________________________________________________________________
 Steve Cousins             Supercomputer Engineer/Administrator
 Advanced Computing Group            University of Maine System
 244 Neville Hall (UMS Data Center)              (207) 561-3574
 Orono ME 04469                      steve.cousins at maine.edu