In article <alpine.lrh.2.20.1707190347390.10...@ns0.nohats.ca> you write: >We are adding something to DNS that's not just a new RRTYPE. It requires >code changes and has a deployment and long tail. ...
I realize that my biggest problem with this draft is not that I don't think that it's useful -- we have lots of RFCs that turned out to be useless but harmless. It's that it breaks the DNS by being egregiously not backward compatible. I would strongly prefer if we defer consideration of this draft until we figure out how to do DNS versioning, some way to say that this record type (and consequently, the zone returned to this AXFR or IXFR) requires special processing, and if you don't know how to do the processing, don't guess. This would update or perhaps even replace RFC 3597. We did this in a horrible ad-hoc way with DNSSEC, and even with DNSSEC there's the fallback that the unsigned answers you get from a server that doesn't understand RRSIG et al. are for many purposes adequate. But with BULK, if a secondary doesn't understand it, the answers will just be wrong. This might be something like an EDNS item which includes the need-to-understand rrtypes, but I'd prefer to do it in a way that will make the AXFR or IXFR result invalid to an old server that doesn't understand it. R's, John PS: h/t to Andrew Sullivan who replied to my suggestion that people do BULK in a stunt server by noting that we're here to make things interoperate. _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop