* Stephane Bortzmeyer: > I'm still uneased about the notion of "silent truncation" (section > 2.3). To me, what is described here (not sending part of the glue) is > not truncation at all (a word which has a very specific meaning in the > DNS). What is the opinion of the WG? Am I the only one complaining > here?
I'm not concerned so much by the choice of name, but I'm not sure if those delicate rules actually buy us anything. I'm not sure if this applies to name servers and their use of glue records, but I've helped to track down one bug in an MTA which assumed that once there were A or AAAA records in the additional section, the additional section contained complete RRSETs. As a result, it did not issue another round of queries for A or AAAA RRSETs at those names. This turned out to be a mistake because AAAA and A RRSETs can have different TTLs, and expire at a different time. The bug was that the MTA saw the AAAA records, didn't issue an A query, couldn't deliver mail to the IPv6 addresses, and bounced messages as a result. The lesson I took from this is that the shape of the additional section doesn't tell you much about what records and record sets are actually available. Perhaps the situation with glue records and referrals is different. But I also expect that caching resolvers will gradually shift to agressive validation of glue records against more authoritative data, to get more consistent behavior (independent of what queries have previously been processed by a particular resolver). This would make efforts to reduce query load by carefully populating referral responses rather pointless. _______________________________________________ DNSOP mailing list [email protected] https://www1.ietf.org/mailman/listinfo/dnsop
