notify option does not work in Bind 9.x, is this a known bug?

Steven Job list3 at wwwcrazy.com
Mon Feb 21 02:50:30 UTC 2005


Quoting Jim Reid <jim at rfc1035.com>:

> Yes. Switch off NOTIFY on all the servers except the master. On the
> master, use an explicit also-notify{} list so that the NOTIFYs go to
> all the servers in your anycast cloud. This list will of course
> contain the real IP addresses of the servers, not their anycast
> address(es).

This is what I have done.  But none of the servers in the "also-notify" seem to
be getting any of the NOTIFY messages. (and yes it is the real IP address and
not the anycasted. ;-) )


> Or you could have a hierarchical NOTIFY scheme if there
> are lots of servers: ie a stratum N server sends NOTIFYs to X stratum
> N+1 servers and so on.

Only 13 different locations to start.  Do you think I would need it at this
level?  Put up two systems and at the second level and have one send to six and
the other send to seven?


> This makes sense because this sort of meta-data should not be in the
> DNS. Of course that meta-date is part of the infrastructure. But it's
> not necessarily visible as resource records, just like the list of
> masters{} servers a slave might use is not visible. Remember too that
> an anycast operator will want to keep internal operational details to
> themself. There should be no need for external users to know or care
> how many servers are in the anycast cloud or where they are located.
> The whole point of anycasting is to allow the operator to add, remove
> and relocate servers within the cloud without their customers and end
> users needing to be aware of those changes.

Exactly...... but having a heck of the time getting it working with 13 right now
as the notifies are not being sent (or they do not appear to be).


> You are a bit confused. In its wider context anycasting is a routing
> trick to make the same IP address appear to be in many places at the
> same time. Anycasting does not depend on the application protocol (ie
> DNS) or an implementation of that protocol. Nominum's ANS has no
> special hooks for anycasting. [Neither does BIND. Or NSD AFAIK.]
> UltraDNS has their own DNS implementation which has an Oracle
> back-end. UltraDNS does anycasting and I believe they use Oracle's
> replication stuff to synchronise new zone data across its anycast
> nodes. When Nominum's DNS hosting business, GNS, was sold to UltraDNS
> the GNS customers were migrated to the UltraDNS platform. UltraDNS did
> not acquire ANS and don't use that.

I totally agree and I understand what IP anycast is.  I have just been having
problems with getting a good anycast setup with 13 systems now (3 visible
anycasted IPs).
And you are correct.  ANS was created after UltraDNS purchase GNS and it's
software.  It was my understanding that their main authoritative works off of
BDB, but it may be Oracle like you said.  But it was my understanding the
UltraDNS purchased their platform from Nominum.  I was wondering if they did
that since they had problems with the IP anycast on Bind.

I know that many of root name servers are anycasted but they really do not have
that many zones.  Wondering if there is a limitation in Bind with this.  Or it
is just a configuration problem on our end. ;-)

Regards,
-Steve





More information about the bind-users mailing list