[mdlug] Samba and F5 for Failover?

[Adam Tauno Williams]

On Wed, 2009-10-07 at 09:19 -0400, Wojtak, Greg wrote:
> Say I have two samba servers, ServerA and ServerB.  In front of those,
> I want to set up a failover mechanism  by leveraging an F5.  The idea
> is that the data is that the data is replicated with less than a 4ms
> lag between the two servers (backend disk replication).  My question
> is, is there anything special I need to do with Samba to get this to
> work?  

You can do anything you want - it will not work.  This isn't Samba's
fault.  CIFS/SMB is stateful.  Unless you preserve the state the clients
will all error out and you will almost certainly have corrupted data.  

Purchase VMware ESX and provision Vmotion.  That is the best you can do.

Load balancers can only work [from the end-user's perspective] for
stateless protocols like DNS, FTP [and only sometimes], HTTP [and only
sometimes], etc...

Also raw disk replication will not work.  You will just end up with
trashed files.  Well, it will work for files that were not open [or at
least did not have oplocks against them] which is certainly better than
nothing.  But this kind of 'dumb' brute-force redundancy will not give
you simple straight-forward fail-over;  for that you must preserve
state.

> I want the F5 virtual name that users will hit to be, say, "smb."
> Would I then configure the samba servers' names to be "smb" or would
> it be something else, like smbA and smbB?  Any other considerations?

Yes, that it will not work.