[dns-operations] A record round-robin behavior

[Jim Reid]

On Apr 21, 2006, at 10:01, Mark K. Pettit wrote:

>> I am not saying that randomizing data is a bad thing I am saying that
>> client should not count on an RR set being randomly shuffled.
>
> Yeah, obviously.  I"m just trying to attempt to predict the  
> behavior of BIND.

What's the point? If you agree that a client shouldn't rely on a  
randomly ordered RRset, you're really agreeing that clients should  
not expect any style of RRset ordering in DNS responses. In other  
words you're trying to predict something that's inherently  
unpredictable. Even though BIND's behaviour wrt RRset ordering is, as  
a general rule, deterministic.

> In investigating this whole thing, I came across the "rrset-order"
> option in BIND 8, which I hadn't heard of before.
>
> This makes things even more confusing.  According to the man page, the
> default rrset-order is "cyclic".  To quote from the BIND 8.4.6 man  
> page:

...snipped...

> This implies to me that if I don't have an "rrset-order" statement  
> in my
> named.conf file, A record round-robins should be returned in a  
> cyclical
> fashion.  That is not the behavior I see.
>
> Anyone have any idea why that might be?

Perhaps it's because the server you're querying (or the servers  
they're querying) don't run BIND 8.4.6. Or something else is making  
queries for the same name/class/type tuple in between the times you  
make those queries. Or maybe there's some firewall/DNS proxy between  
your client and the server that's interfering with the DNS traffic?