Strange problem

[Clenna Lumina]

Måns Nilsson wrote:
> --On mC%ndag, mC%ndag 21 maj 2007 23.40.41 +0200 Jean-FranC'ois Leroux
> <leroux.jeanfrancois at gmail.com> wrote:
>>>
>>> Doubled notifies should not be a problem; they're just discarded or
>>> NOOP's.
>>> Make certain that you see a notify from the server you have in
>>> masters {}; directive -- the slave will by default discard notifies
>>> from servers not in
>>> masters {}; (But also look at "allow-notify").
>>
>> Yes, the slave discards notifies, as I've seen in the log. Now can I
>> (or should I) allow notification from my internal slave, either by
>> adding it as a master for that zone or by adding it in my 'notifers'
>> acl, which is used inside the zone statement?
>> What I'd like to do is giving redundancy, but I'm not sure this is
>> the correct way to do it.
>
> Having multiple masters is one idea. If played right, you can
> automatically avoid problems caused by losing any one node (except
> perhaps the hidden master; there, you might need to do manual work.)
> by having multiple paths.

Would this be a good scheme, just out of curiosity?

(I jotted this down and it seems to make sense:

[Internal/Hidden]
Master-A:   IPs: 10.0.0.2
 Slave-A:   IPs: 10.0.0.3                 masters { 10.0.0.2; };

[External/Public]
Master-B:   IPs 10.0.0.4, 12.123.100.44   masters { 10.0.0.2; 
10.0.0.3; };
 Slave-B:   IPs 10.0.0.5, 12.123.100.45   masters { 10.0.0.4; 10.0.0.2; 
10.0.0.3; };


If Master-A goes down, Slave-A can (using it's backup zonefile copy) 
temporarly serve Master-B. Slave-B can still update from Master-B; 
Both -B's can still serve the world and local network.

If both Master-A and Slave-A go down, Slave-B can still get data from 
Master-B; Both -B's can still serve the world and local network.

If Master-B should go down, Slave-B can still pull from the "-A" 
servers. Slave-B can still serve the world and local network.

If both -A's go down and Master-B, Slave-B can still serve to the 
world/lan using it's backup copy.



Any flaws or problems here, or something I missed? (I'm asking this for 
educational benefit, as one never knows if they'll done day need it :)

--
CL


>>> significant? It might be so that the notify arrives from S1 to S4
>>> and is processed before the zone has been properly transfered to S2.
>>
>>
>> Well, no, the zone isn't really significant ( about 100 lines), but
>> the idea was fine. Actually, reinstalling bind9 solved this, so I
>> guess something had gone wrong when updating from 9.3.2 to 9.3.4.
>>
>> Thanks for your answer too.