Re: [DNSOP] [Last-Call] [Ext] Genart last call review of draft-ietf-dnsop-rfc8499bis-09

[Shumon Huque <[e-mail]>]

On Mon, Sep 18, 2023 at 9:34 AM Paul Wouters <[e-mail]> wrote:

> On Sun, 17 Sep 2023, Salz, Rich wrote:
>
> [ speaking as individual only ]
>
> >>> On the other hand, spending a week or two repeating a cycle to get an
> important term in the current document seems like a better solution.
> >
> >> If the WG agrees that this is an important term, sure.
> >
> > Well, if the IETF has consensus :)  I'm raising the issue during this
> last call that "round-robin" should be in the list of defined terms.
>
> I would say that if the WG didn't think it was important at the time by
> forgetting it, it probably is not an "important term", and I can see
> this not being fixed in an IETF LC anymore as an acceptable outcome.
>
> Especially as the DNS Terminology document seems to be getting refreshed
> pretty regularly to begin with.
>
> But also, I didn't find a good definition of round robin in existing
> RFCs either. There is a mention in rfc1794 but it doesn't really define
> the term there. So I am not sure what the DNS Terminology document would
> reference?
>

I agree.

My personal experience is that it isn't a widely used term and it's been
more
than a decade since I last heard it. The fact that no-one even brought it up
at any point during the development of this doc or its predecessors seems
to
confirm that.

Back then, my recollection was there were 2 distinct meanings of round robin
DNS: the first was simply a Resource Record Set with multiple Resource
Records in it, where the DNS server returned the constituent records in the
RRset in a rotated or shuffled manner. That is in fact the typical behavior
of
many current DNS implementations, so I think that is already covered by the
inclusion of the term RRset in the document.

The second meaning was from implementations that selectively returned
one record from the RRset in the answer, and rotated the selection of that
record in response to successive queries. By spec, since RRsets are "atomic"
units where all the records have to be returned in the response, this is an
example of a non-standard DNS trick. There are many other examples of
such non-standardized response generation mechanisms (GLSB, weighted
answers, etc), and none of those are documented in the DNS terminology
BCP. It would probably not be appropriate to do so. If necessary, those
could
be documented separately in an informational document.

Shumon.