On Thu, 31 Aug 2000, John Horne wrote:
> 1) At present the Uni has one central mailhub with a disaster recovery plan
We've had two "mail hubs" for some time now --- neither actually holds
mail (it gets delivered down to NetWare systems plus 1 awkward user who
insists on his mail being on a Unix system). The "failover" stuff is
done with MX records, and works sufficiently well that we will take the
main mail server out of service during working hours for maintenance.
Gotcha's:
* Mailing lists (which you've spotted). I'm still thinking
about how to have redundant mailing lists.
* An idle backup machine. Our backup machine is quite a capable box,
and it may cause problems if people don't understand what it is there
for (seeing it as wasted resources).
* O/s differences. Our mailservers are sufficiently different that we
rebuild the data on each box. If you're planning to build the data
once, and copy the databases to the other hub, check that this will
work.
Advantages:
* You can re-build the data on the backup mail server to avoid
re-building the data on the main box, and causing problems if the
source data is bad (we fetch most of the aliases from the Netware
servers which sometimes miss out chunks of the data).
> 2) We used to get ask quite a bit, although not so much now for some reason,
> the name and/or IP address of the mailhub so that users could configure
We have a name (smtpserver.port.ac.uk) that is usually two A records
pointing to both the main mailserver and the backup. When taking the
main mailserver out, we run a script on the DNS server that changes the
name to have just one A record. Seems to work quite well.
- --
mike.meredith@??? http://www.iso.port.ac.uk/~mike
Senior Informatics Officer (Postmaster, Hostmaster, and security)
