Problem with BIND-8.2.3-REL loading large zones?

[Mark_Andrews at isc.org]

> We subscribe to the RBL+ zone transfer service from mail-abuse.org, and we
> have noticed some problems with BIND since we started using it. When the
> slave server reloads the zone (which happens approximately every 2 hours),
> I see things like this in the syslog file:
> 
> Oct 20 13:22:15 ensemada named[21575]: reloading nameserver
> Oct 20 13:22:33 ensemada named[21575]: master zone "rbl-plus.mail-abuse.org" 
> (IN) loaded (serial 1003612469)
> Oct 20 13:22:35 ensemada named[21575]: master zone "lave.net" (IN) loaded (se
> rial 2001101701)
> [...]
> Oct 20 13:22:35 ensemada named[21575]: Ready to answer queries.
> Oct 20 13:22:35 ensemada named[21575]: drained 59 queries (delay 20 sec)
> 
> It appears that while named is reloading the rbl-plus zone, it just drops
> incoming DNS queries in the bitbucket. The rbl-plus zone is the largest zone
> that we have, weighing in at around 15 MB. Is this a known problem with
> BIND 8.2.3-REL that is fixed in some other version? Or is BIND simply not
> able to both reload a large zone while handling queries?

	Upgrade to BIND 9.2 (currently 9.2.0rc7).  It has incremental
	loading of master files.  BIND 9 also does a much better job
	of IXFR than BIND 8 (we learn't a lot about how not to do
	IXFR with BIND 8).

	Zone loads in BIND 8 are atomic operations and it takes
	time to process 15 M of master file.

	Note: most clients will have retried well before 20 seconds
	has elapsed.  Rather than waste time on processing queries
	and sending back answers that will be bit bucketed we just
	bit bucket the queries if we have been "off the air" for
	long enough to be noticed.  Redundancy is built into the
	protocol.  It shouldn't matter if one server is not answering
	for a short period of time.

	Mark
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: Mark.Andrews at isc.org