Re: [DNSOP] More work for DNSOP :-)

[Andreas Gustafsson <[e-mail]>]

Olafur Gudmundsson wrote:
> There is a new version in the works, expect it late tomorrow (monday) 
> [...]
> I tries to define that resolver treat NOTIMP as long term signal
> that resolver should keep track of and not retry.

That's a bad idea, IMO.  When the resolver gets a NOTIMP response, it
has no way of knowing if it means "meta-type not implemented", "class
not implemented", "opcode not implemented", or any number of other
"not implemented" conditions.

There have been many operational problems caused by resolvers applying
cached SERVFAILs or cached indications of non-responsiveness to entire
servers, entire names, or excessive periods of time in violation of
RFC2308 sections 7.1 and 7.2.  We are now at risk of repeating all
those mistakes with the caching of NOTIMPs.

If we really do want NOTIMP caching in spite of those risks, it should
be specified in a separate draft updating RFC2308.  Discussing it in
the context of ANY will only make things worse by biasing the reader
towards assuming that NOTIMP responses are ANY related.

Also, ANY queries are relatively rare, and with this draft, should
because even rarer, and cheaper for the authoritative server to
respond to.  Keeping state in the resolver is expensive, and the
memory would probably be better spend on other things such as a
larger cache.
-- 
Andreas Gustafsson, [e-mail]