[Pdns-users] MongoDB Backend merged Re: Mongo DB and PowerDNS part 3: Now with DNSSEC

[Nick Milas]

On 26/5/2011 3:44 ΌΌ, fredrik danerklint wrote:

> Nick,
>
> To answer all of your questions, please take a look at the source file
> 'communicator.cc'. At the end of that file there is a function called
> "void CommunicatorClass::mainloop(void)" that checks slave and master every
> other 'slave-cycle-interval' seconds (parameter taken from the pdns.conf
> file). I assume that this function is running in a seperate thread.
>
> This is how PowerDNS knows when to send a update to other nameservers.
>

Actually, I am afraid things do not work like that. The 
'slave-cycle-interval' parameter is only used by slaves and only when 
they (the slaves) are in undetermined state, i.e. at launch. "Once a 
domain has been checked, it will not be checked before its SOA refresh 
timer has expired."

What I was asking is how the *Master* knows that the serial in the SOA 
of one of its zones has changed. If you have not implemented some 
solution, Master will never know that a serial (in its own backend!) has 
been changed, unless you manually let it know. The PowerDNS 
documentation states: "Some backends may be able to detect zone changes, 
others may choose to let the operator indicate which zones have changed 
and which haven’t. Consult the documentation for your backend to see how 
it processes changes in zones." The usual logical solution is the use of 
triggers (if your backend supports them).

Of course, for the Master it doesn't really matter to know that a serial 
has changed, because it directly refreshes its data (which is retrieved 
from the database). [ I don't know what happens with cached data, if 
they are used - LDAP backend doesn't use cached data; the LDAP server 
takes care of that. ] But the Master will not be able to Notify 
slaves... They will wait until their refresh interval (specified in the 
SOA) expires and only then they will ask the Master if serial has changed.

So, in fact you don't have what pdns calls Master operation, unless the 
backend on the Master provides a mechanism to detect serial changes and 
send Notify messages to slaves. If it doesn't, you must manually or 
semi-manually send Notify messages, as I am also now doing with LDAP 
backend, using a cron job to detect externally (i.e. not within the 
backend) serial changes and to send, when such changes are detected, 
Notify messages to slaves.

But I guess, when you don't have triggers, you could embed in the 
backend the above procedure. You could define a time parameter (perhaps 
changeable in the config) which would cycle domains (zones), 
automatically, as part of the backend process and detect serial changes. 
As I have explained earlier (in other threads), this works fine for a 
moderate number of zones. If the number of zones is high, however, it 
doesn't scale well.

Nick.