On Sun, 16 Jan 2000, Mike wrote:
> If I set this up in DNS, what would happen if an outgoing server in the
> cluster went down? Also how effective would the load balancing be?
>
> > Add an entry on the RedHat server to smtproutes, that points to an
> > internal DNS record.
> >
> > Then, initialize your DNS server and put 10 A records for this hostname.
> > Each individual message will then be sent to a randomly chosen IP address,
> > one out of 10, balancing the load.
Most deliveries will be unaffected. Simple arithmetic shows that 10% of
the time the message will have the unavailable IP address picked. When the
connection attempt fails, the relay will simply switch to another server.
If the unavailable server is still up, but not running SMTP, the
switchover will be almost instantaneous, because it will refuse the
connection immediately. If the server is completely dead, there will be a
delay of about a minute for the connection to time out.
Note that most TCP/IP stacks will allow you to bind multiple IP addresses
to the same interface. If the server is going to be offline, simply bind
its IP address to one of the other available machines, until you fix the
server.
Of course, that machine will now get twice its regular load, so this
should be simply used as a stopgap measure, and if the server is going to
be down for a prolonged period of time, its IP address should simply be
removed from the DNS record until it's fixed.
