[ietf-dkim] SSP issues

[Steve Atkins]

On Jun 2, 2007, at 1:04 AM, Hector Santos wrote:

> Steve Atkins wrote:
>> On Jun 1, 2007, at 7:30 PM, Arvel Hathcock wrote:
>>>> (2) SSP record type (TXT vs. something new). Only 4 messages in  
>>>> discussion, mostly saying "if you support TXT, don't bother with  
>>>> anything else."  Again, no clear consensus.
>>>
>>> If a new RR can solve the wildcard issue and we feel that this is  
>>> a significant issue worth solving (or at least addressing) then  
>>> perhaps we should create a system that looks for a new RR first  
>>> and failing that, falls back to TXT.
>>>
>>> I don't think the "if you support TXT, don't bother with anything  
>>> else" position is correct.  If we come out with a spec that  
>>> states "SSP clients must query for new RR first, then TXT"  
>>> senders would be right to expect compliance.
>> What would "compliance" entail prior to universal, or at least  
>> widespread, support for the new RR by all stub resolvers and  
>> recursive resolvers? Or would you wait for that widespread support  
>> before releasing the spec?
>
> Steve,
>
> I am bit of a lost of what is so complex here.  It is the lack of  
> understanding of the DNS technical compatibility issues that exist  
> when it comes to handling of new RR records?

> Until most, if not all, DNS servers, especially those with cached  
> DNS servers, support RFC 3597, "Handling of Unknown DNS Resource  
> Record (RR) Types", we will need a fallback on a TXT record concept.
>
> see RFC 3597,  http://www.ietf.org/rfc/rfc3597.txt
>
> It is really isn't all that difficult.
>
> The bottom line is that there will be many systems with DNS servers  
> and domains that simply will not be able to work with a NEW RR and  
> seriously doubt DKIM is going to be the primary reason for most to  
> begin changing their setup or network in order to support the  
> "near" reliable propagation of NEW RR queries.

So if the spec states "SSP clients must query for new RR first, then  
TXT" you wouldn't expect most receivers to comply with that?

> There is really no choice here.  To choose RR only, we are  
> strategically saying that we don't want older systems and all must  
> upgrade.   That isn't a good strategy, never mind unrealistic.  We  
> are talking about a huge population of DNS systems that simply do  
> not have the capability of handling a new RRs.
>
> If we want to maximize adoption, a TXT record is required.

Cheers,
   Steve