On Thu, May 20, 2021 at 01:36:52PM -0700, Randy Bush wrote:
> >> ever see the cat herding video?
> > https://urldefense.com/v3/__https://archive.psg.com/cat-herders.mov__;!!NEt6yMaO-gk!UwP65THoPAKtaE74N5YFw9MSNO17zkgu96E4xctycM7-Iu1KDATEZf22uke20g$
> > 
> > Does not render. Oh well, I’ve seen cat videos before, I’ll use my
> > imagination.
> 
> i suspect your corporate nannyware put some strange rubbish after the
> ".mov"
> 
> but you can also try https://archive.psg.com/catherding.mpg
> 
> >>> 3. Section 3:
> >>> 
> >>>   Any particular inetnum: object MUST have at most, one geofeed
> >>>   reference, whether a remarks: or a proper geofeed: attribute when it
> >>>   is implemented.  If there is more than one, all are ignored.
> >>> 
> >>> This makes me think the geofeed: attribute will never, ever be
> >>> adopted, because you’ve just created a flag day requirement.
> >> 
> >> no.  this is per inetnum: object, not per registry.  see beginning of
> >> para.  perhaps i should add an example of an inetnum: object.
> > 
> > I know what an inetnum: is (although it’s a fair point that maybe not
> > everybody does). In thinking about it a little more, yes “flag day”
> > was too strong, however I think there’s still potentially a problem,
> > depending on what entity has the issues.
> > 
> > If it’s only the RIR (“server”) side that lags, then presumably
> > clients will be built to consume both geofeed: and inetnum: from day
> > one. In that case cutover is easy: once the RIR implements geofeed:,
> > you cut over to it in all the inetnum:s that RIR serves, done.
> 
> it's even simpler.  the whois server would support both flavors of
> attribute in an inetnum:, remarks: and geofeed:.  (note that it MUST
> support the remark: form.)  it is the individual inetnum: objects
> registered by ISPs that cut over one by one at their leisure.  an isp
> with 42 inetnum: objects could cut them over one at a time on wednesday
> afternoons.

I was going to comment something similar to John until I noticed this
snippet:

   Until all producers of inetnum:s, i.e. the RIRs, state that they have
   migrated to supporting a geofeed: attribute, consumers looking at
   inetnum:s to find geofeed URLs MUST be able to consume both the
   remarks: and geofeed: forms.  [...]

So, AFAICT, we already do require that all clients will be able to consume
both forms, and the complications that John is worried about "just go away".

> > On the other hand, if there are clients (now, or ever) that implement
> > only inetnum:, then I do think we’re in for an eternity of supporting
> > both.
> 
> i think you mean if there are clients that do not support the geofeed:
> attribute form in inetnum:s.  yup, then they are not going to get the
> urls from inetnum:s which have moved to the new form.  such is forward
> migration.
> 
> >>> 4. Section 5:
> >>> 
> >>>   If and only if the geofeed file is not signed per Section 4, then
> >>>   multiple inetnum: objects MAY refer to the same geofeed file, and the
> >>>   consumer MUST use only geofeed lines where the prefix is covered by
> >>>   the address range of the inetnum: object they have followed.
> >>> 
> >>> Presumably this only works with unsigned geofeeds because you’re
> >>> implicitly requiring that the geofeed file be signed only once. Is
> >>> there any particular driver for this sign-only-once requirement? On a
> >>> cursory review of §4, I don’t see anything that would make multiple
> >>> signatures impracticable.
> >> 
> >> the geofeed lines would then need to be sorted, clumped, ...  seems a
> >> bit complex for not much win.  due to the administrative structure and
> >> process, inetnum: objects tend to the same granularity as RPKI objects.
> >> so having the geofeed files follow seems natural.
> > 
> > OK. My fear was that there might already be substantial investment in
> > geofeed:s that cover multiple objects, and resistance to shredding
> > them down to individual object-level feeds. But maybe there isn’t and
> > won’t be.
> 
> in parallel, i am thinking of the implications of a question in this
> space from ben.  i think the comparitive granularity of inetnum:s and
> the corresponding rpki data is important but unknown.
> 
> my head hurts.  i want to keep thinking about this space.

Makes sense (on both counts, unfortunately).

> > I took a look and it looks good. I have one new nit, in §4, because of
> > course I do:
> > 
> >    The bracketing "# RPKI Signature:" and "# End Signature:" MUST be
> >    present exactly as shown.
> > 
> > A pedant might say “exactly as shown” means verbatim "# End Signature:
> > 192.0.2.0/24”
> 
> the quote marks were insufficiently explicit?
> 
> > it may be worth being even clearer. It would be possible to write it
> > out in excruciating detail of course, but possibly replacing “exactly
> > as shown” with something like “following the model shown” might be
> > enough. (I say “ish” because I wonder if “192.0.2/24”, the way
> > right-thinking people write prefixes, would also be OK, or if it’s not
> > “exactly as shown” enough.)
> 
> i'll try it.  but the previous complaint there was that the text was
> insufficiently prescriptive.

(FWIW, this is basically the space in which my second discuss point falls.
If there's any semantic value to the stuff after "RPKI Signature"/"End
Signature", we should say what it is, even if it's just matching the
start/end lines.  If there's no semantics at all, that's probably fine too.

> > I was sad to see the tweezers left on the cutting room floor but I
> > understand why.
> 
> for you, i would throw them back.  but the strength of the current
> "brute force" phrasing appeals.

I guess I have to go back to an older version and see what I missed by only
reading the -12, then!

-Ben

_______________________________________________
OPSAWG mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/opsawg

Reply via email to