Using TCP for checking

Kevin Darcy kcd at chrysler.com
Tue Apr 7 20:28:40 UTC 2009


RFC 1123 (Section 6.1.3.2) says "DNS servers MUST be able to service UDP 
queries and SHOULD be able to service TCP queries.". You, as a registry, 
may opt to enforce a "must service TCP queries" policy, but if you're 
checking minimal standards-conformance, UDP is mandatory, TCP is 
optional, you could presumably skip the TCP check altogether.

"TCP only" is permitted "by private agreement", according to that 
section of RFC 1123, but, again, you're the registry, you can set 
whatever rules you want, including, if you wish, the rule that your 
registry validation script has no "private agreements" with registrants 
in this regard, and thus UDP must be supported for registry<->registrant 
transactions, even if they have "private agreements" with everyone else 
they talk to.

- Kevin

Mark Elkins wrote:
> I'm involved in the CO.ZA Registry. In the process of registering a
> domain name in the co.za zone - we do a bunch of DNS checks using
> 'dig'. 
>
> for each nameserver, 
>   a) check that the zone exists (fetch the SOA), 
>   b) fetch the NS RRSet count and compare entries.
>   c) if Nameserver inside the domain being registered (glue needed)
>     i) check the reverse glue (can be multiple v4 + v6 addresses)
>     ii) check each reverse has a forward
>
>
> Currently - many of these (dig-9.4.1) checks include the flags +time=9
> +retry=5..
>
> ..the assumption being that for any 'dig' action - try, timeout 9
> seconds - repeat another 5 times... - so a totally failed lookup would
> take 54 seconds... however - an ethernet trace/dump seems to indicate
> queries go out one after the other - with little inter-query delay..
>
> If we do a lookup with UDP - a low but significant number of 'digs' fail
> - which results in our checks failing - and the registration checking
> process delaying that particular registration for a few hours.
>
> If we switch to using TCP for 'dig' lookups  - the failure rate
> basically disappears to Zero. This would result in happier customers
> (less registration delays).
>
> I've always been taught (and teach others) to use UDP and not TCP for
> DNS queries - but in the case of a registry checking for info like we do
> - would it not be politically correct to instead do TCP checks?
>
> What does the net-dns wisdom say?
>
> My current thought is to do a UDP check (don't change timeout/retry from
> default) and only if that fails - retry immediately with a TCP Check.
> Others in my group are for using TCP immediately.
>
>   




More information about the bind-users mailing list