James Raftery <james.raftery@ucd.ie> wrote:
2782 says if you want to contact host foo for service X over protocol Y, then lookup "_X._Y.foo IN SRV". This draft says if you want to send a query ABOUT foo to a whois server, lookup "_whois._tcp.foo IN SRV". The semantic meaning of ``foo'' has changed and that change is not explicitly stated. I would like it to be.
RFC 2782 says in "Overview and rationale", 3rd paragraph: Clients ask for a specific service/protocol for a specific domain (the word domain is used here in the strict RFC 1034 sense), and get back the names of any available servers. The wording in the introductory example also supports this. In the case of "whois.denic.de" you actually do not want the whois server offering info about "denic.de" but that one covering "DE" (although they happen to be the same, but you have to *know* that in advance). SRV RRs are not only meant to (re-)direct host specific services, but also domain (or organisation) based services. This is what comes to use in this case. The "whois" or maybe "finger" or even "www" cases are special since there are common DNS aliases (RFC 2219) for them. SRV avoids cluttering the "visible" part of your name space by moving services into the "_" area.
Secondly, I feel a reminder that the target of an SRV must have an A RR is in order. Taking my isp.net example, above, again. If I controlled customer.com and wanted to use the mechanism in this draft to publish the fact that customer.com is in isp.net's whois server I cannot use:
_whois._tcp.customer.com. IN SRV 10 0 43 whois.isp.net.
as ``whois.isp.net'' does not have an A RR. I must ``undo'' the name indirection desired by isp.net and instead publish the following RRs (and keep them up to date):
_whois._tcp.customer.com. IN SRV 10 0 43 dbserver.isp.net. IN SRV 20 0 43 dbbackup.isp.net.
I wonder whether this will happen very often (i.e. we'll have to discuss, what the actual problem/task is we're going to solve). This might occur in cases where the registrar (instead of the registry) keeps the whois data and each and every domain will have to point to the appropriate one. However, unless the customer is running an additional whois server on their own, you might get away with a CNAME RR instead of "unrolling" the SRV.
Sounds interesting. I'll play wit it later. -Peter