Re: [exim] round robin -> load balancing query

Tony Finch wrote:
>
> We use round robin DNS and it works fine, so long as nothing breaks :-)
> Most MUAs will not try more than one IP address, so if the IP address they
> have fixed on stops working they will break too. The way we deal with this
> is to just shift the IP addresses of the broken machine onto a hot spare.
> (We haven't automated it because our machines are too reliable to make it
> worth adding complexity.) The other problem is that client computers often
> don't respect the TTL on DNS records, so they may not respect changes
> promptly. (This seems to be less of a problem now than a few years ago.)
> The upshot is that you need to change the DNS several days before taking a
> machine out of service.

Round-robin DNS isn't quite as reliable today as it used to be, thanks
to RFC 3484. Take a look at the last couple months of the Debian tech
committee archives for the gory details.

- Marc