Fail over/Recovery roadblock

[Scott Lovenberg]

On 2/10/2012 4:38 PM, Pavel Herrmann wrote:
> Hi,
>
> I am currently using a single samba4 server for everything (migrated from a
> signel samba3 DC, so everything includes fileserving) in my domain, but i am a
> bit concerned about reliability, as more and more services integrate
> themselves into AD environment.
>
> my plan is to set up two samba4 DCs and two samba3 fileservers as virtual
> machines on two separate hosts, with the aim that hardware failure of one of
> the hosts would not affect the availability (it would certainly affect
> performance, but everything should still work).
I'm thinking you'll take a performance hit on file locking, but you can 
probably configure around this.
> the samba4 DCs should be synchronized by DRS (hopefully DNS partition
> replication will get fixed soon, or is fixed already), GPOs would be
> synchronized by lsyncd
>
> the problem is with fileserving, as I dont have a SAN (thay cost way too much
> for this setup), i would use drbd (with clvm and GFS2) to synchronize the
> fileservers, and CTDB to keep samba state in sync. with this i would like to
> use MSDFS to automagically balance load between the nodes (and transparently
> switch to single-node in case of a failure)
>
Sounds reasonable.  Have you thought about using iSCSI for the 
back-end?  You could use a single box to start and then scale that out.  
That's supposedly (according to Red Hat) the next step up from just 
using CLVM.  IIRC, I got GFS working with iSCSI on the back end.  YMMV, etc.