Re: Cluster Sendmail?

[John Haxby <jch scalix com>]

Ed Wilts wrote:

> My mail systems (Red Hat Linux 6.x) don't crash :-).  Seriously, they've
> never, ever crashed.  We had a data center power outage a couple of
> times but the server came back and processed its mail queue.
>  
Mine doesn't either.   On the other hand, I've had hardware go bad in 
all sorts of interesting ways.   If mail is stuck on a defunct server 
for a couple of days and the European Sales manager sent out an 
important message just before he left ... well, you aren't going to be 
terribly popular when the message eventually makes it out of the 
system.  I mean, how were you to know that there was an important 
message sat in that queue waiting for a network connection?

The OpenMail (remember OpenMail, the mail system that refuses to lie 
down and play dead?) HA solution used HP's ServiceGuard.   We put each 
OpenMail server's associated sendmail queue into a file system along 
with the necessary config files and if something went wrong the whole 
lot switched over to a back-up machine.   The configuration file(s) for 
sendmail set the host name for the instance (package, whatever) as well 
as things like the listening IP address, location of the spool and 
whatnot.  It worked well (and some people are still using it, I think).

In a lot of cases, the need for HA is over-rated.  Fix the common 
problem -- disk failure -- with a decent array and you've got a system 
that's going to be reliable enough for most people.   On the other hand, 
if you want to get from 99.99% to 99.999% then you're probably after an 
HA solution.

jch