BIND servfail from caching server

[Justin Krejci]

When doing a recursive query for MX supernet.com against a caching BIND
server, the BIND server responds back with the answer.  The TTL is 300.

After the TTL expires the following recursive query for the same record
returns a SERVFAIL from the caching server.

If I do a +trace on the same query to the same caching server for the
same data it is able to respond with the answer yet a standard recursive
query still gives a SERVFAIL.

Queries for other domains are working fine on this caching server. Other
3rd party DNS caching servers are responding fine for the same record
above even after the TTL expires, tried @8.8.8.8 and @208.67.220.220

If if flush the cache on the caching server it successfully returns the
answer to the query but only for the up the TTL's life then goes back to
SERVFAIL again. (tried doing a full stop-and-start of named as well).

This particular server is running BIND 9.7.0-P2 but this exact same
behavior is also happening on a server running 9.5.1-P2.1 as well.

So I noticed when doing a trace that the NS servers are different
between the gtld and the actual authoritative servers.

<snip>
com.                    172800  IN      NS      l.gtld-servers.net.
com.                    172800  IN      NS      e.gtld-servers.net.
;; Received 502 bytes from 192.36.148.17#53(i.root-servers.net) in 2987
ms

supernet.com.           172800  IN      NS      ns2.earthlink.net.
supernet.com.           172800  IN      NS      ns3.earthlink.net.
;; Received 111 bytes from 192.54.112.30#53(h.gtld-servers.net) in 119
ms

supernet.com.           300     IN      MX      5
onemain-mx.earthlink.net.
supernet.com.           3600    IN      NS      dns1.earthlink.net.
supernet.com.           3600    IN      NS      dns2.earthlink.net.
;; Received 172 bytes from 207.217.120.43#53(ns3.earthlink.net) in 54 ms



Is this just a bug that upgrading BIND will fix or is there something
else going on here?