Re: [core] Chair's review of draft-ietf-core-hop-limit-02.txt

[<[e-mail]>]

Hi Klaus, 

Agree. The reasoning is straightforward for error codes. 

For the few cacheable success codes, we maintained the same wording because a change of the hop-limit is seen as a hint of a topology change and thus as a trigger to check the forwarding path. The conditions are as follows: 
* The forwarding path is changed between the proxy and the server (e.g., adding new proxies to the path). Such path may not be loop-free. 
* The client is informed out of band to increase the initial hop limit to accommodate the increase of hops.

Cheers,
Med

> -----Message d'origine-----
> De : Klaus Hartke [mailto:[e-mail]]
> Envoyé : mardi 26 février 2019 12:19
> À : BOUCADAIR Mohamed TGI/OLN
> Cc : Carsten Bormann; core; [e-mail]
> Objet : Re: [core] Chair's review of draft-ietf-core-hop-limit-02.txt
> 
> Mohamed Boucadair wrote:
> > Carsten Bormann wrote:
> > > 2. I don't understand the rule about using cached copies.  As a
> > >    general observation, rules like this should not simply be stated
> > >    without a rationale why they are the way they are.
> 
> I assume you're referring to this rule:
> 
>    A CoAP proxy which understands the Hop-Limit option MUST NOT use a
>    stored response unless the value of the Hop-Limit option in the
>    presented request is less than or equal to the value of the Hop-Limit
>    option in the request used to obtain the stored response.
> 
> (This matches the language in [1].)
> 
> Let's say proxy receives a request and has a matching Hop Limit
> Reached error response in its cache. If the hop limit in the request
> is equal to the hop limit in the request for the stored response, then
> the stored error response can be returned. If the hop limit is less,
> then the stored error response can also be returned. (If 6 hops are
> not enough to reach the server, then 3 hops are also not enough.) If
> the hop limit is greater, then the stored error response cannot be
> returned. (The server might not be reachable in 6 hops, but it might
> be reachable in 7 hops.) (This assumes that the number of hops to
> reach a server doesn't change. If it changes, then a stored error
> response might be served incorrectly until its max-age expires.)
> 
> I have to admit that I never checked if this also makes sense for
> stored responses that are not a Hop Limit Reached error response...
> 
> Klaus
> 
> [1] https://tools.ietf.org/html/rfc7252#section-5.6