Solution:
If the system is capable of IPv6 but has no IPv6 interface configured, run named with the -4 option (IPv4 only).
For example. Add the -4 flag to the ExecStart named command in "/usr/lib/systemd/system/named.service".
Like this: ExecStart=/usr/sbin/named -4 -u named -c ${NAMEDCONF} $OPTIONS
Or preferably by adding the option to /etc/sysconfig/named. Like this: OPTIONS="-4"
For good measure. In addition also remove/comment out all things IPv6 in the BIND configuration.
o This will prevent named from trying to use IPv6 to access the root servers.
o Don't think it is necessary to remove the AAAA records from the root hints. Think they will just be ignored.
o Don't know whether or not forwarders are afflicted by this.
A few symptoms of this issue are:
1) Slow initial/intermittent recursion lookups. About 10 seconds was the experience here on this system.
2) A lot of "resolver: info: resolver priming query complete" log entries.
3) Slow to get the cache primed after start up.
System:
KVM VPS
CentOS Linux release 7.8.2003 (Core)
Derived from Red Hat Enterprise Linux 7.8 (Source)
BIND 9.11.4-P2-RedHat-9.11.4-16.P2.el7_8.3 (Extended Support Version) <id:7107deb>
Conclusion:
Hopefully this has been fixed by default in later versions. But for those not there yet...
Enjoy the fruits of my labor, as I am doing, with snappy recursive resolving and a more stable BIND implementation.
