failover options

[Simon Waters]

"Nicholas Pieter Heesters Jr." wrote:
> 
> I was wondering what people use for DNS failover. In particular, if one has
> multiple A records for the same name to different IP addresses (for certain
> cluster configurations), what are some good solutions for detecting down
> nodes in the cluster and "disabling" those A records in DNS.

Good clustering solutions fail-over the IP address (and
optionally the MAC address - just make sure your switch doesn't
object to this before you need it for real). Kind of makes DNS
based failover a little clumsy.

> I know "out of the box" DNS solutions found in Solaris and HP-UX, etc, don't
> offer this functionality, but how about QIP? Or a hardware solution, such as
> an F5 or Arrowpoint?

The other DNS and BIND book [Langfeldt] mentions eddieware
www.eddieware.org (gone?!, which I assume is the same as
http://eddie.sourceforge.net). Written in Erlang - which bodes
well - even if it makes it a slightly exotic beast. I haven't
needed anything like this myself so can't comment on it.

I dare say "noddy systems" could easily be deployed with
nsupdate - where servers register themselves with TSIG signed
dynamic updates, and are unregistered if the DNS server can't
reach them, naff perhaps, but for simple purposes it might well
be enough. 

Of course the low TTL required will increase the load on the DNS
servers.

Load balancing is a totally different subject, and it is worth
saying what your load balancing as HTTP has many more and
different solutions, than say load balancing relational
databases.

Dare one say 'DNS is at best, the second best method of load
balancing for any application'