Inconsistent results for domains following cache flushes

[Nick Jones]

We are seeing weird issues with customers domains:
www.andra.com.au and www.wsid.com.au

Flushing the cache then doing a lookup for one then the other, often one
will succeed and the other will fail.

[NBRS-3.0 myhost named]# host www.andra.com.au
www.andra.com.au is an alias for andra.com.au.
andra.com.au has address 203.84.232.7
andra.com.au mail is handled by 10 mail.andra.com.au.

[NBRS-3.0 myhost named]# host www.wsid.com.au
Host www.wsid.com.au not found: 2(SERVFAIL)

Flush the cache again and repeat the lookup, and the result may swap.

Putting in a forwarder for those two domains resolves the issue.

The domain is a little weird in that the whois shows the NS records as
ns3.danielbeard.com and ns4.danielbeard.com, but the authoritative
servers list ns.andra.com.au. Also, www.andra.com.au is a cname for
andra.com.au.

A trace shows:

26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): create
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): join
26-Jun-2009 09:30:29.991 fetch 0xa0ddd60 (fctx
0xaf24578(www.andra.com.au/A)): created
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): start
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): try
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'):
cancelqueries
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'):
getaddresses
26-Jun-2009 09:30:29.991 dns_adb_createfind: name 0x9db43c8 is an alias
(cached)
26-Jun-2009 09:30:29.991 dns_adb_destroyfind on find 0xa77d208
26-Jun-2009 09:30:29.991 dns_adb_createfind: name 0x9db4310 is an alias
(cached)
26-Jun-2009 09:30:29.991 dns_adb_destroyfind on find 0xa77d208
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): no
addresses
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): done
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'):
stopeverything
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'):
cancelqueries
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): sendevents
26-Jun-2009 09:30:29.991 fetch 0xa0ddd60 (fctx
0xaf24578(www.andra.com.au/A)): destroyfetch
26-Jun-2009 09:30:29.991 fctx 0xaf24578(www.andra.com.au/A'): shutdown

Is this issue caused by bind caching the glue result from the root
servers, then ignoring the authoritative result or something like that?

The bind version is: 9.3.5

Thanks for any suggestions

  Nick