[Samba] SAMBA Clustering

[Wieprecht, Karen M.]

I have an SGI cluster (two servers share same physical disk space).   I made
the username map live in clustered space and made the
~samba/locks/winbind_idmap.tdb also live in the cluster space via a symbolic
link on both servers.  Bother server run samba using the same smb.conf
settings (different hostname of course).  

 The PC users run a login script that tests is one server exists,  if so,
it maps the share to a network drive from that server,  if not, it checks to
see if the other server exists,  and (hopefully that machine is up) maps the
drive from the alternate server.  If something bad happens to the primary
server or we have to shut it down for maintenance,  the user can simply log
out and back in  and get his samba services from the alternate machine.  

If a domain user without a corresponding UNIX account creates a file on a
public share,  he gets assigned a UNIX UID on the fly from the winbind
range, and that is tracked in the winbind_idmap.tdb.  If the other server
ends up having to serve the files,  we want it to get the same UID mapping
info. Putting that out on the cluster was our solution,  and the cluster
file locking mechanism will guarantee that only one samba session write to
that file at a time,   plus,  the pcs should only be connected through our
primary server under normal circumstances,  the other samba is running "just
in case".   Keep your fingers crossed,  but so far so good. 

	karen wieprecht