DNS name resolution question

Kevin Darcy kcd at daimlerchrysler.com
Wed Sep 11 23:55:32 UTC 2002


Aaravali Karavali wrote:

> I have a question regarding DNS name resolution capabilites.
>
> Here is my situation.  We have a DNS name which is currently being
> resolved using round-robin to two IP addresses.  If I do a nslookup on
> the same name successively, I get two different IP addresses.  For
> example:
>
> 1. nslookup weblogic-cluster.
>
> Address:  216.33.240.47
>
> 2. nslookup weblogic-cluster.
>
> Address:  216.33.240.12
>
> Incidently, these IP addresses are associated with the DNS names
> weblogic-cluster-mb and weblogic-cluster-ma respectively.
>
> Is it possible to setup a record type in DNS that would return back
> the following when the name is resolved
>
> nslookup weblogic-cluster.
>
> 216.33.240.47, 216.33.240.12

Returning all A records in a response is the *default* BIND behavior,
whenever one has multiple A records owned by the same name. Returning a
single, differing A record for successive queries is *not* within BIND's
capabilities, short of changing the record back and forth, and (unless
the change was made via Dynamic Update) reloading the containing zone,
between each query.

Is the nameserver which serves the "weblogic-cluster." name a
BIND nameserver? Rather, I suspect you're talking to some sort of
load-balancer. If you undelegate/re-delegate that name, so that a
BIND nameserver is serving the relevant zone, then what you want to do --
returning all of the A records -- can be achieved rather trivially.

Now, if you want to always return the A records in a particular *order*;
that may be a little trickier. Because of caching complications, you'd
have to configure *all* servers which may be answering that query --
regardless of whether they are authoritative or merely caching servers --
with an "rrset-order" option, or the non-BIND equivalent. If some of
those servers are not in your control, or are non-BIND implementations
which lack an equivalent feature, this may end up not being feasible.


- Kevin




More information about the bind-users mailing list