multiple CNAMEs revisited

[Kevin Darcy]

CS wrote:

>This is a testy subject, and I'm fully aware that what I want/need to
>do is not allowed.
>
>That said, I need to replicate the following "bogus" configuration:
>
>spamd   IN      CNAME   spamd1
>spamd   IN      CNAME   spamd1
>spamd   IN      CNAME   spamd2
>spamd   IN      CNAME   spamd2
>spamd   IN      CNAME   spamd3
>spamd   IN      CNAME   spamd4
>
>spamd1  IN      A       10.0.0.1
>spamd2  IN      A       10.0.0.2
>spamd3  IN      A       10.0.1.1
>spamd4  IN      A       10.0.1.2
>
>This hack let you work around BIND collapsing identical A records.  If
>it's not obvious, I'm working with a spamassassin cluster, and spamc
>with the above dns hackery and a local copy of dnscache with the
>round-robin patch currently splits the load between two servers very
>well.  And it solves the "what if a server is down" problem by just
>trying the next server in line if it doesn't get a response.
>
>root at helga[/tmp]# grep "spamd1 spamd" /var/log/spamd|wc -l
>   20872
>root at helga[/tmp]# grep "spamd2 spamd" /var/log/spamd|wc -l
>   20870
>
>I need to add two more smaller servers to this cluster, so I'd like to
>alter the weighting in DNS.  Again, I understand DNS is not meant to
>solve these problems, but in this case, it could solve them very well
>if the above config was legal (or if there was a config option to
>allow the "illegal" behaviour).
>
>So given the current constraints, what is a good workaround?  I've
>looked at load-balancing tools, and frankly this is a better solution;
>I don't want to rely on one piece of software; DNS in this case has a
>good deal of built-in redundancy.
>
>If this is not feasible with bind, are there any other servers I
>should be looking at?
>
This is not feasible in BIND, unless you want to hack some functionality 
into the code. I hacked a $WEIGHT directive into BIND 8 a number of 
years ago, but things were much simpler then, and I think I just punted 
on the interaction with Dynamic Update (i.e. what happens to the 
weighting if someone adds or deletes A records from the RRset?). 
Nowadays, one would have to consider the interaction with DNSSEC as 
well. Of course, being just a zonefile directive, this weighting 
metadata is not something that would be propagated via AXFR/IXFR (an 
alternate replication mechanism, such as rsync or scp, could get around 
this).

I'm not sure of any other DNS software that implements this feature. I 
assume you've already looked at djbdns, since you mentioned "dnscache"...

                                                                         
                                                      -Kevin