Hi Joe!
On 05/11/2024 12.47, Joe Abley wrote:
On 5 Nov 2024, at 13:13, Shane Kerr <sh...@time-travellers.org> wrote:
I wrote a quick draft to specify that answers returned should be returned in a
random order:
https://datatracker.ietf.org/doc/draft-kerr-everybodys-shuffling/
I think that you might need to nail down what "random" means. I presume you
don't mean random in the xkcd-221 sense, for example. You probably don't mean lava-lamp
or double-pendulum random either, at least not as a MUST. Would it be ok for an rrset to
be shuffled once per minute rather than every time a response is generated?
In the security section I do mention that you don't need
cryptographically-secure random numbers. I could expand that a bit, if
it is useful.
I did consider the idea of periodic shuffling. That makes sense to me,
since I think we can reasonably assume that servers will not be
shuffling at exactly the same time and should have different results. It
would mean slightly more state on the server (tracking when the last
shuffle was) in exchange for saving the work of re-ordering results on
every response. If you think it makes sense I'm happy to add text about
this approach.
More generally from your OARC talk I think your experience is that
(paraphrasing) not shuffling is fine for most people, but sometimes causes
problems. Does it make sense to translate that into a normative MUST for the
protocol?
My own feeling is that probably it impacts many zones but people are not
aware (which could be interpreted as "is fine", but also could be
interpreted as "basically works but doesn't actually do what the user
thinks it does"). There are for sure some zones that experience problems.
I learned offline that at least one resolver operator recent changed
their system to randomize results recently independent of my research
and presentation. Plus most open source resolvers randomize results by
default.
The fact that people keep independently discovering the same thing is
what prompted me to write this draft!
I think good practice on the consumer end of this is to understand that
addresses obtained from a dns response be treated as an unordered set and not
an ordered list. Maybe it makes sense to talk about recommendations there as
well as recommendations for DNS responders?
The draft does recommend that both stub resolvers and applications treat
the RR in a random order.
I'm happy to expand on that although I am doubtful it will help much.
Maybe it could be convincing enough that OS or library authors will
change how they present answers to applications though... 🤔
Thanks for your feedback!
--
Shane
_______________________________________________
DNSOP mailing list -- dnsop@ietf.org
To unsubscribe send an email to dnsop-le...@ietf.org
