The long term way to fix this is for DNS servers to *always* fill in the additional section for select RR types (e.g. SRV) including chasing down missing additional records and setting TC=1 if those additional records will not fit for recursive queries. TC=1 is already required when glue records do not fit.
This lets the applications get the entire chain without having to come back and ask again. This service would be requested via a EDNS option which will appear in the response and its presence in the response indicates that the client does not have to query for missing RRsets. The recursive server is guarenteeing that the response is complete. Now if SRV is inappropriate for the application, e.g. because it doesn't work well with wild cards, then a application specific RR needs to be defined. This gets you a single lookup with no followup queries required once the recursive server supports this. If the client is still talking to a legacy server it would still need to do followup queries for missing records. Clients would ask for A, AAAA, and SRV in parallel until this is well supported or we have a flag day after which A and AAAA records should not be requested. CNAME does 50% job. Lets do something that does a 100% job. Yes it requires both DNS recursive servers and applications update their behaviours. Mark -- Mark Andrews, ISC 1 Seymour St., Dundas Valley, NSW 2117, Australia PHONE: +61 2 9871 4742 INTERNET: ma...@isc.org _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
