BIND 9.7 Serial Number Decrease Problem

[John Wobus]

Barry Finkel wrote:
> I ran a test this morning on one of the Solaris 10 slave servers.
> A query to the server showed serial numbers:
>
>      _tcp   1238
>      _udp    842
>
> Both of these match the zone on the MS Windows DNS Server.
> I checked the zone files on the slave server:
>
>      _tcp   1239
>      _udp    843
>
> Both of these are increased by one from what BIND returns in
> response to a query.
>
> The two zones have NO .jnl files.
>
> I did
>
>      ./rndc stop
>      <<Wait for the "exiting" message.>>
>      /etc/init.d/named.anl start;tail -f /var/adm/messages
>
> Once BIND started, the serial numbers were INCREASED, as I
> expected they would be, given the lack of .jnl files.
>
> And a few minutes later BIND complained about the serial
> number on the master being less than that on the slave
> for both zones.  I consider this a bug in BIND 9.
> What further diagnostics do I need to get?
>
> I have another Solaris 10 slave on which, I assume, I can
> duplicate this.  And from past experience, in one day, after
> the zone has expired and been refreshed, I will be in the same
> state on this slave.

Do bind slave instances EVER make up or increment serial
numbers?  This just seems like such an unlikely bug
that bind would start doing that.  Could it be that
the supposed slave instance is accepting dynamic updates?

I'd be tracing/tracking SOA files on the master, and communications
between the dns instances very closely before I'd even
give such a potential bug much thought. Perhaps there are
bind functions that I'm not aware of and I'm wrong.

John