[dns-operations] HTTP SRV, was Adding CNAME for the root domain issue

[Paul Vixie]

Tony Finch wrote:
> ...
> _imap._tcp.example.com. SRV target.example.net.
>
> The client requires two round trips, first to query for example.com,
> second to query for target.example.net. It can't eliminate the second
> round trip because it doesn't know the target name until after the first
> round trip, and it can't rely on the additional data being complete.

as with mx, the additional data both need not be complete, and mostly is.

need not be, because if there are any [AAAA | A] RRsets matching the SRV 
target name in the additional section, and using them results in 
success, then there is no worry as to whether there are other [AAAA | A] 
RRsets that were not included.

mostly is, because the SRV target is often in the same zone, so barring 
truncation, the [AAAA | A] RRsets will already be in cache since they 
will have been sent with the cache miss response.

pragmatically speaking, if an SRV client sees only one RRset (AAAA 
alone, or A alone) it can query for the one it didn't receive. negative 
caching does the rest. CDN's (low AAAA/A TTL, or ECS) fold in nicely in 
this model, since the missing RRsets are either popular, or they aren't.

-- 
P Vixie