[dhcwg] Re: Question about draft-ietf-dhc-rfc8415bis

[Tomek Mrugalski <[e-mail]>]

On 12/12/25 14:37, Adrian Farrel wrote:
> Resend to fix up the (non-standard) WG email address
> 
> -----Original Message-----
> From: Adrian Farrel <[e-mail]> 
> Sent: 12 December 2025 13:20
> To: '[e-mail]'
> <[e-mail]>
> Cc: '[e-mail]' <[e-mail]>
> Subject: Question about draft-ietf-dhc-rfc8415bis
> 
> Hi,
> 
> I'm currently doing a Directorate review of
> https://datatracker.ietf.org/doc/draft-ietf-spring-dhc-distribute-srv6-locat
> or-dhcp/
> 
> They make reference to your draft saying:
> 
>    This document assumes that a single transaction for all of the IA
>    options required on an interface is used, as recommended in
>    Section 18.1 of [I-D.ietf-dhc-rfc8415bis].
> 
> Looking at your section 18.1 I find:
> 
>    This document assumes that a client SHOULD use a single transaction
>    for all of the IA options required on an interface; this simplifies
>    the client implementation and reduces the potential number of
>    transactions required.
> 
> The use of "SHOULD" puzzles me. It appears a bit of a double conditional:
> you are assuming client implementation behavior, and the clients should act
> in a certain way.
> What happens if a client decides to use more than one transaction, and why
> might it do that?
Hi,

As Bernie and Michael pointed out, it's easier for both client and
server if everything is done in one transaction. But I want to elaborate
a bit on the "why it might do that" part. There might be different
reasons why certain parts of the configuration are requested, not being
initiated at the same time and following different life cycle. Couple
examples: having separate address for a container or service that's just
went up or is shutting down, requesting PD because new interface was
enabled on a router etc. Depending on how the client is implemented, it
might be cleaner in some cases to have per IA state model. I'm not
saying it's a good idea. IMHO it isn't. But you asked for why a client
might decide to do such a thing.

>From the server's perspective, it doesn't matter, except a slightly
increased traffic. Nothing bad would happen if few client would do that,
but we want to avoid all clients doing that. Hence the SHOULD (as in you
don't have to if you have good reasons).

Cheers,
Tomek

p.s.
BTW it might be useful to update the 8415bis reference to RFC9915 soon.