In article <f4f41481-c2a7-4fdf-af76-65c36ab23...@nohats.ca> you write: >Not convinced the situation should be this black and white - eg perhaps >partial glue would be enough >not to require TC=1 or behaviour for resolvers could be a little more advanced >to try with partial >before going to TCP. > >If my request seem stupid, the draft needs clarification for stupid people >like me :)
The draft, which I hope we adopt, could use clarification if that seems like a good idea. Imagine you have foo.example with two nameservers, ns1.foo.example and ns2.foo.example. Client looks up something.foo.example, server returns a referral with two NS records but only has room for one A record for ns1. The Internet is having a bad day and ns1 is unreachable while ns2 is fine. Since there's no TC=1 the client has no idea that requerying would return the A record for ns2, so it wrongly assumes ns2 has no A record and the domain is kaput. It can't separately requery for ns2, since who would it ask? It's fine to return a partial result with TC=1, that's always been the case. R's, John _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop