[dns-operations] resolvers considered harmful

[Robert Edmonds]

Paul Vixie wrote:
> [...] we can move the stubs to an HTTPS transport if we can agree on a
> RESTful query API and either JSON or XML schema for responses, but it
> is not in our power to make UDP, especially fragmented UDP as in EDNS,
> work in the last mile. [...]

Why does moving to a new transport require re-designing the encoding
scheme?

Register a new IANA media type, e.g., "application/dnsmessage", defined
as a binary DNS message, encoded per RFC 1035 §4.1.

See RFC 4027, 2540 for a related precedent, the extant "application/dns"
media type:

    http://tools.ietf.org/html/rfc4027
    http://tools.ietf.org/html/rfc2540
    http://www.iana.org/assignments/media-types/application/dns

The HTTP client POSTs to a static entry point for this hypothetical
DNS-over-HTTP proxy, setting request headers:

    Accept: application/dnsmessage
    Content-Type: application/dnsmessage
    [...]

and attaching the RFC 1035 §4.1 encoded DNS query message as the request
body.

The HTTP server responds with status code 200 and sends the RFC 1035
§4.1 encoded DNS response message as the response body.

-- 
Robert Edmonds