NFS automounting. Stable for mail servers?
    Brian Candler 
    B.Candler at pobox.com
       
    Tue Jun 20 14:15:37 UTC 2006
    
    
  
On Mon, Jun 19, 2006 at 05:24:01PM -0400, Francisco Reyes wrote:
> Our setup has a number of front end mail machines accesing nfs servers.
> To distribute the load we create directories like
> /mail-server1
> /mail-server2
> /mail-server3 
> 
> Each of the front end machines mounts all the servers from fstab. Problem 
> with that is that if loading from fstab if one of the nfs servers freezes 
> the front end machines may have problems while trying to access the downed 
> nfs server.
That's true, but you'd want to be a bit careful with an automounter; you
don't want your MTA to create /mail-server1/u/s/username/Maildir on the
front-end's local disk if /mail-server1 was not mounted!
What would you expect the front-end to do if a user whose mail is on
/mail-server1 connects, but /mail-server1 is down? You could add a check to
the application code to check for this case. Or, perhaps more simply, you
could just use a 'soft mount' for NFS, and let it fail. I've not tested this
myself. I've built clusters around Netapps and they are too reliable :-)
As for automount, I don't use that either, but I wouldn't expect much
problem if you were just auto-mounting /mail-server1, /mail-server2,
/mail-server3 and leaving them mounted. I expect the problems you refer to
are when user1 connects and you automount /mail-server1/u/s/user1/ (that is,
thousands of automounts and unmounts taking place all the time)
Regards,
Brian.
    
    
More information about the freebsd-isp
mailing list