BIND 8.2: 99% server load when lame delegation occurs (reprised)

[Mark_Andrews at isc.org]

	I know what is happening.  Basically when TSIG support was
	introduced it broke the clearing of RD in ns_forw().

	I'll generate a patch to TB5.

	Mark

> 
> As this problem still persits, perhaps I can take a different approach
> with my questioning:
> 
> To recap: in BIND 8.2.1, if a client queries our nameservers for domain
> information about a lame domain, each of our nameservers that is listed as
> an InterNIC nameserver for that domain suddenly throttles to 90%+ usage on
> their respective boxes for a period of seconds (usually 20-90 seconds).
> >From the dumps, they seem to be querying each other repeatedly for the
> same question (but I could be wrong about this).  After this period, they
> finally offer a response.
> 
> Others have also seen this phenomenon, and it only seems to occur on
> 8.2.x.
> 
> My question is: what is BIND supposed to do when it encounters a lame
> delegation--that is, it has been asked for zone information, though it not
> configured with that zone? What steps should it take to attempt to resolve
> the query? By default, BIND is configured for "recurse yes": how should
> this affect BIND when it encounters a lame domain?
> 
> My second question might only be answerable easily by ISC: did something
> change in the coding from 8.1.x to 8.2 that may have affected the way BIND
> handles lame delegation responses?  I've been sifting through the code
> differences, but I might do better if I had some code areas on which to
> focus.  Any suggestions?
> 
> Regards,
> 
> JD
> 
> 
> 
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: marka at isc.org