Recently I had the problem, that my dyndns hosts could not be resolved for backup. Monitoring showed me, that changeip.com has some short blackouts from time to time. Only for a few seconds, but when my internal bind9 server was not able to fetch the dynip, it cached the negative answer for 3 hours.

The magic keyword is: max-ncache-ttl

This is the solution to change this behavior:

named.conf

options {
        listen-on-v6 { none; };
        auth-nxdomain no;    # conform to RFC1035
        directory       "/var/named";
        dump-file       "/var/named/data/cache_dump.db";
        statistics-file "/var/named/data/named_stats.txt";
        memstatistics-file "/var/named/data/named_mem_stats.txt";
        allow-query     { any; };
        recursion yes;
        max-ncache-ttl 15;  // 3600 seconds is one hour

        bindkeys-file "/etc/named.iscdlv.key";
};