High availability mail server options

Craig White craigwhite at azapple.com
Tue Oct 4 19:55:47 UTC 2005


On Tue, 2005-10-04 at 14:21 -0500, Les Mikesell wrote:
> On Tue, 2005-10-04 at 07:03, David Hollis wrote:
> > For my company, I've setup a mail/groupware environment that uses
> > Postfix, OpenLDAP, Postgres, Apache, etc and am now looking at ways to
> > make it a more redundant arrangement.  All of the components have
> > methods to help me with the lone exception of the backend mail storage
> > for the end users.  Ultimately, I am wanting to have systems at
> > different geographic locations, not even on the same network, ideally
> > with users able to access any of them at anytime and be able to do their
> > thing.  The users mail storage is in Maildir format which seems like it
> > will help any replication type scenario.   I can't just NFS mount the
> > mail directories, because then my NFS server becomes my single point of
> > failure. 
> 
> One of the points of using the maildir format when it was invented was
> that it worked on NFS, and back then the way to make NFS reliable was
> to run it on an appliance like the NetApp filer, and if a single unit
> wasn't reliable enough you could get them with remote mirroring.  These
> days you might be able to find something cheaper, but the same concept
> should work.
----
that may be but I would definitely check with other users of the same
mail delivery agent to find out how they are handling backup and
redundancy so you get their wisdom instead of surprises.
----
> 
> > Do things like GFS work to handle this?  If so, do they
> > operate across slow links (not talking dial-up here, but general
> > Internet cable/DSL type links)
> 
> Is GFS production-ready?
----
I don't know - if you look at
http://www.redhat.com/software/rha/gfs/
you would get the impression that Red Hat thinks so
----
> 
> > If it helps, our total mail volume is not that tremendous so completely
> > instantaneous replication isn't totally necessary, but I would want
> > fairly quick convergence (say 30 minutes or less).
> 
> You could cycle through the maildir directories with rsync to a
> failover location pretty quickly - or even take everything in
> home directories if it doesn't change too much.  However you
> need something different if you expect to access the different
> copies at the same time and propagate changes both ways.
----
I guess it all depends upon one's expectations of HA and the reality of
'pretty quickly' 

Craig


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.




More information about the fedora-list mailing list