Re: ns delegation, unix dig vs dnsjava dig vs dnsjava Lookup

[Brian W. <bwe...@xb...>]

On Wed, 18 Jul 2007, Stefano Bagnara wrote:

> Brian Wellington ha scritto:
>>> We use dnsjava to lookup MX records for SMTP delivery: what do you think
>>> is the behavior required by DNS+SMTP specification? Currently we simply
>>> run the Lookup using the dnsservers configured by the user (or the one
>>> autodiscovered by dnsjava) and take its result as is. In this
>>> "uudial.ch." domain and 194.246.118.118 configuration the dnsjava
>>> "Lookup" does not return the MX record: is this intended? Should we do
>>> something different? Can you answer this or give me a pointer to some
>>> doc I can read about?
>>
>> I don't know what 194.246.118.118 is, but when I query it, it's not
>> doing recursion.  dnsjava's Lookup class is a stub resolver, so it
>> should only be sending queries to servers that perform recursion.  If I
>> instead send the query to a server that does perform recursion, I get
>> the desired MX record.
>
> Maybe I'm confusing iterative vs recursive terms.
>
> I tried googling this but I didn't find an answer:
>
> is a server allowed to reply non-recursively to a recursive question?

Yes.  There is no such thing as a recursive question - there is a question 
where recursion is desired.  There's no way to require recursion.

> what does a client should (have to) do when a server replies without the
> "ra" flag?

There's nothing a client can do.  A client which is incapable of 
performing recursion itself should not be configured to point at servers 
which do not perform recursion.  Anything else is a configuration error on 
the client.

> Should we keep resolving the data to the NS server received?
>
> If so, can you suggest an approach to the problem using dnsjava?

Continuing the resolution would be the correct thing to do if you were 
writing a name server.  From the context of a client wishing to lookup MX 
records, I believe the right thing to do is simply fail.

> Would this behavior be defined "iterative" resolution or iterative vs
> recursive both describe only server side resolution?

I'm not really sure how iterative resolution would differ from recursive 
resolution.  There are basically two kinds of resolvers - full resolvers 
and stub resolvers.  A full resolver would use the complete resolution 
algorithm, talking to authoritative name servers and following NS records. 
A stub resolver would ask a full resolver to do the recursion for it.

Pretty much every client application uses a stub resolver.  It's possible 
to have a full resolver that's not part of a caching name server, but I 
don't know of any.

> What do other other clients do wrt non recursive servers?

They would generally fail to perform the desired lookups.

Brian