From: Bernhard Schmidt (bernibirkenwald.de)
Date: Tue Oct 09 2007 - 17:50:31 CDT

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Mon, Oct 08, 2007 at 06:45:27PM +0200, Mark Martinec wrote:

Hi,

> We initially had a single MX pointing to a host
> with both the A and AAAA records.
>
> Seems like some (rare, unproven) hosts *think* they have
> an IPv6 connectivity whereas in reality they may not,
> or a link may be poor, and it appears they were reluctant
> to try the IPv4.
>
> For this reason we now have two MX records, the first
> one pointing to a hostname with both the A and AAAA records,
> and the second, lower priority, MX points to a hostname
> with only an A record (actually the IP address in both
> A records is the same). Seems to have done the trick:

This is something I've been thinking about as well. While >99% of the
IPv6 deliveries work just fine (700 just yesterday), there are sometimes
persistent errors in the TCP session (mostly caused by pMTU issues). At
the moment we grep those hosts from the logfile ("timeout after DATA" or
"lost connection after STARTTLS") and nullroute them, which forces the
sender to fallback to IPv4. This has proven to work so far. I haven't
heard of any software that attempts IPv6 and never fallbacks to IPv4.

Did you experience any issues with IPv6-enabled setup so far, where the
mail bounced due to broken sending side software?

Regards,
Bernhard

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]