bind 9.4.3b2 probs - "recursive-clients soft limit exceeded"

[Patrick Larkin Jr]

Hi there, needing some input, but here's the background:

I have a couple dozen recursive resolvers around the country.
A mix of Sun v210's on Sol8 and T2000's on Sol10.  All were
running bind 9.4.2, but we built and installed 9.4.3b2 on
two of each platforms (4 machines).  One of the T2000's
quickly became unresponsive with many of these errors:
   recursive-clients soft limit exceeded, aborting oldest query
That machine (of the four) traditionally takes more traffic
than the other three.  All the other machines running 9.4.3b2
(and those on 9.4.2) run fine.

We built 9.4.2-P1 and installed it on the misbehaving T2000
and it seems to run OK, but we get a few of these at times:
     socket: too many open file descriptors

root at rns-catcall[233]# grep recurs named.conf*
named.conf:     recursion yes;
named.conf:     recursive-clients 50000;
root at rns-catcall[234]#

The one machine with problems on 9.4.3b2 had status values like this:
	recursive clients: 49800/49900/50000
and now, on 9.4.2-P1 its:
	recursive clients: 410/49900/50000

Our build command for 9.4.2, 9.4.2-P1 and 9.4.3b2 were all the same:
  ./configure --prefix=/ms/svc/dns --enable-threads --disable-ipv6 
--disable-linux-caps --with-randomdev=/dev/urandom -
-without-kame --sysconfdir=/ms/svc/dns/etc --localstatedir=/ms/var 
--mandir=/ms/man && MAKE

The only other thing we had to do to get 9.4.3b2 running was to
mknod /dev/poll in our chroot.

What do you think I should do?  It looks perhaps like 9.4.3b2 is
consuming resources and not clearing them fast enough.
We preferred 9.4.3b2 due to the performance enhancements.
But we rarely exceed 5000 QPS on any given machine.

Thanks!
-- 
  Patrick Larkin Jr - Dallas Texas USA
    Earthlink Core Services Engineering
       PLarkin at corp.EarthLink.NET