[Herefordshire] High Availability Linux

[Malcolm Herbert]

Chris, 

the RHEL3 Cluster Suite product only provides Active-Passive clustering,
requires shared storage (its uses a Quorum partition). Assume you are
trying to provide HA SMTP and POP3/IMAP solution which is Active-Active
(?). 

You'll probably need to use some form cluster file system (for maildir)
or preferably a database. GFS, GPFS are choices for the filesystem,
whilst MySQL (with replication) maybe an option. 

Perhaps can organise a quick chat and look at the options.  I can
potentially organise a good speaker on the subject from within Red Hat. 

Malcolm


On Fri, 2005-01-28 at 10:54, Chris Bond wrote:
> Hi John,
> 
> > I confess to having no experience with clusters but, 
> > conincidentally, yesterday I read an article in next months 
> > (Feb 05) PCPlus about using heartbeat for failover. I'll can 
> > email it to you if you like. It sounds as though you are on 
> > the right track with heartbeat, drdb, rsync etc.
> 
> If you could send the article over that would be great.  Im thinking for
> disk sharing to use our SAN instead of drdb as it should be a lot better
> than an external scsi array.
>  
> > For mail, a single DNS entry should suffice as your failover 
> > will have the same IP. Secondary MX is a little redundant as 
> > most mail servers will keep resending for a few days at least 
> > and your HA cluster will allways be, well, available :) 
> > Spammers sometimes exploit secondary MX in the hope that 
> > there will be less spam checking etc on backup hosts, so 
> > maybe use it only if you can keep the mailserver configs in sync.
> 
> What I might do is setup an IN A with a low ttl of say 60 then add both ips
> to say mailcluster then set the MX preference to that.  I'd rather have two
> boxes fed the mail and be able to deal with it rather than waiting to an
> "emergency" to test it out.
>  
> > This all sounds great fun! It would be good to hear of your 
> > experiences when the cluster is up and running. It will be 
> > really handy to be able to switch off a box for maintenance 
> > without interrupting services.
> 
> I'll let you know how it goes, should get to play with it in a few weeks.
> 
> Kind Regards,
> Chris Bond
> 
> 
> _______________________________________________
> Herefordshire mailing list
> Herefordshire at mailman.lug.org.uk
> http://mailman.lug.org.uk/mailman/listinfo/herefordshire
-- 
-----------------------------------------------------------
Dr Malcolm Herbert
Manager, Client Development             GPS, Red Hat Europe
m: +44 7720 079845   w: people.redhat.com/mherbert
GPG ID:  C051F446
GPG FPR: 68AD CBDD 8EEB 63ED 0802  4B8E A5B3 EB72 C051 F446
-----------------------------------------------------------