Re: [dnsext] WG opinion on draft : Improvements to DNS Resolvers, for Resiliency, Robustness, and Responsiveness

[Brian Dickson <[e-mail]>]

On Wed, Feb 23, 2011 at 12:19 PM, Florian Weimer <[e-mail]> wrote:
> * Florian Weimer:
>
>>> nlnetlabs nsd and verisign atlas both understood the spec in this
>>> regard also.
>>
>> Actually, ATLAS sent NXDOMAIN instead of NODATA for existing,
>> non-delegated names with a missing RRset (that is, empty terminals) at
>> one point in the not too-distant past.  I don't remember if they
>> implemented the old BIND behavior, too, but it would be odd to send
>> NXDOMAIN for empty terminals, and NODATA for empty non-terminals.
>
> Off-list, I was told that my terminology is off, which is true.
> Please read "empty RRsets" for "empty terminals".

So, this is an interesting data point, arguably in support of Paul,
because of the following:
- it was clearly broken behavior (really, really broken)
- you won't find a shorter tail than that - the vendor *is* the *only*
user of the software (!!)
- it helps illustrate the difference between NODATA and NXDOMAIN (and
when the latter is the wrong answer)
- wow - see point 1 :-)
- it shows that regardless of deployment scope, there should be
pressure to fix broken deployments, rather than to ankylose (rigidly
anchor, forever) that broken-ness into the specifications (IMHO)

Brian