Bind deadlock?

[Gushi]

Hey all,

I'm really not sure where to go on this for more information.  I've got
a couple of named processes on two different boxes that will
periodically deadlock and all recursive queries will return SERVFAIL.
Authoritative queries still work.

The named process will no longer respond to any signals, and will only
be take-downable by a kill -9.

I'm running the latest version of named, on the latest version of the
OS (Freebsd 6 in this case).

At this point, the kind of answer I'm looking for is more of an "what
do I do to debug this" type of answer -- what kinds of things can I
turn on in logs to trace this down.

My experience with BIND logging seems to have been that the various
compile-time defaults on every OS are different (i.e. the RPM version
of BIND under linux seems to want to log different things than does
what I'm used to), and I'm running with a stock logging configuration,
so I'm not sure about the best debuglevels and such.

Anyone can advise here?  At this point the only solution is rebooting
the machines periodically, which is far from acceptable.  Would anyone
find a corefile useful?