SMB server with failover node

[Bill Moran]

Miguel Gonçalves wrote:
> I am going to install a FreeBSD 4.8 SMB server (using Samba)
> in a corporate environment. Due to the small number of
> clients on the network I will alocate 6 GB on the main server
> hard drive for those users. In this main server I have 2
> hard drives configured as a RAID-1 Vinum device.
> 
> What kind of graceful failures have users of Vinum noticed,
> i.e., can the server enter degraded mode using only one drive
> without much user intervention? Money is a concern but should
> I move to hardware RAID (with hot swap option) to prevent user
> intervention? I can only get near the machine at best in 3 or
> 4 hours (and the local users don't understand FreeBSD) but I
> will monitor the machine daily remotely.

You'll be fine.  Unless you have the statistically improbably bad
luck of both drives failing in a short period of time.  when a
mirrored drive fails in vinum, the system keeps going with no
visible effect to the users.  I've let a server run like this for
a few weeks because the hardware was on order (not recommended!)
and I couldn't fix anything until I had a drive to replace the
failed one with.

> To better cope with failure of this server I am also installing
> a second FreeBSD box that has much more hard disk space and
> will sync those 6 GB of data every 15/30 minutes. What is the
> best way to mirror the data: rsync or dump? What if there are
> open files? Will rsync work?

I'm not 100% sure of the effects of rsync on open files, but
otherwise it works very well.  I've been using it for backup
purposes for years now.

> If the main server fails I can instruct the local staff to
> disconnect it from the network and execute a command at the
> backup server's console. This command would change the server's
> IP address and would start the Samba daemon.
> 
> Any ideas for a better setup?

Sounds pretty good.  You could (therectically) write a cron script
that checked the first server periodically and automatically did
the switchover in the event of a failuer.  This is probably overkill,
however, and there's probably no way to be sure you've caught every
possible type of failure.

Sounds to me like you've got a good plan in place.

-- 
Bill Moran
Potential Technologies
http://www.potentialtech.com