Mac OSX 10.7

[Tim Chown]

On 21 Jul 2011, at 12:30, Jeroen Massar wrote:

> According to that there is first an "Addr" query, I assume a "A" record
> query, then an IPv6/AAAA one, they probably just nearly race eachother
> and as the A was asked first it is of course returned first.
> 
> Keeping an eye on the log though, I sometimes see the AAAA answered
> before the A, which is well, weird, seeing that the recursive DNS cache
> for those labels is sitting only a few ms away, it would mean that
> unbound handles requests async or so ;)

Right, which was exactly the behaviour I saw in 10.6.7 during W6D, but it affected the address selection too. You could further highlight it by ping6-ing a target to 'force' IPv6. 

As you point out, with the local resolver nearby, you'd probably expect the query order to determine the response order, but it often doesn't.  We filed the 'picking IP version by DNS response order' issue with Apple as bug ID 9580547, though it isn't a bug per se. Similar results are reported in http://openradar.appspot.com/7333104.

Anyway, the behaviour is now different in Lion.  It may have changed in 10.6.8, which iirc was released a couple of weeks after W6D.

>> I guess whatever their mechanism is is no longer 'secret' within the
>> beta, so we ought to be able to get some public confirmation from
>> Apple.
> 
> Which 'beta' would that be?

/jediwave This is not the beta you are looking for

> The real solution to this all is 'simple':
> - Working RFC3484 support (an RFC from 2003...)
> - Clarity / Documentation on how resolving really works in OSX
>   (seems everything goes through mDNSresolver with then lots of magic)
> - A toggle to tweak this thing as defined in RFC3484
>   eg ala /etc/gai.conf would be awesome

I would vote for that :)

Tim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cluenet.de/pipermail/ipv6-ops/attachments/20110721/d579d75e/attachment.htm>