Notify "storms"

[Bryan Irvine]

On Mon, Jan 18, 2010 at 1:27 PM, Todd <canadaboy at gmail.com> wrote:
> Good day all,
>
> We've run into a problem with our DNS servers.  The way we update our
> masters is via a CVS Checkout and reload of the zones modified.
> Sometimes though, we need to reload the whole config for big
> changs/etc.  When that happens, all 6 masters (I know, we're getting
> rid of some) send notifies to all 80+ (I know, we're getting rid of
> some) slaves for all 1800 zones.  This causes all the slaves to verify
> all 1800 zones on 6 masters, which then delays the changes we made
> from actually getting to the slaves.  Right now it's about 2.5 hours
> for all slaves to do all zones.
>
> We would like to make this better.  We're trying to figure out what
> mechanism might be limiting the rate at which the slave does SOA
> checks against the master so it can perform that step quicker.  We
> have looked at the zone transfer limits on the master/slave, but that
> is related to the transfer mechanism, not the SOA query.
>
> Can anyone help with ideas on this?  Are we missing something obvious?

Might not be what you are looking for but sounds like some of the
ideas presented at infrastructures.org might help.

-B