I've read draft-jabley-dnsop-ordered-answer-section-02.

Overall, the proposal makes sense to me. Given that:

- there are deployed clients that assume a certain ordering of RRsets
  in a CNAME chain and would be difficult to upgrade; and
- there has been at least one interoperability issue caused by a
  server implementation that did not follow that ordering,

mandating the ordering seems likely to help avoid similar
interoperability problems in the future.

I have a couple of comments and questions on some minor points:

1. Should the requirement also apply to chains involving DNAME or
   SVCB/HTTPS (for the latter, potentially including RRsets in the
   Additional section)?

   It may not matter much in practice because the clients that rely on
   ordering are likely to be older implementations and therefore less
   likely to support DNAME or SVCB/HTTPS. Also, in the DNAME case,
   responses include synthesized CNAME records, so specifying the
   ordering of the CNAME chain may already be
   sufficient. Nevertheless, it may be worth considering whether
   similar guidance should apply to other forms of RRset chaining for
   completeness, or explicitly stating why they are out of scope.

2. Section 5 states:

      Client implementations SHOULD accept responses where CNAME RRSets
      appear in any order

  I'm not sure that "SHOULD" is the most appropriate requirement
  here. One of the motivations for this document appears to be the
  existence of deployed client implementations that do not tolerate
  arbitrary ordering. If this document is published with the above
  text, those implementations would immediately become non-conformant,
  even though they are unlikely to be modified in practice.

  At the same time, assuming most server implementations already
  follow the proposed ordering, and future implementations are
  expected to do so as well, this receiver-side requirement does not
  seem likely to materially improve interoperability. I do not have a
  strong opinion on the exact requirement level, but "SHOULD" feels
  somewhat stronger than necessary. Would "MAY", or perhaps not using
  an RFC2119 keyword at all, be more appropriate?

--
jinmei
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to