[bind10-dev] datasource list vs zone list
Evan Hunt
each at isc.org
Fri Jan 15 16:48:58 UTC 2010
> > If you consider the latter to be a kind of "virtual data source", then
> > that's two data sources being accessed in series.
>
> That is a sexy idea. A very sexy idea. GENIUS!
>
> I mean, it may not work, but it's so elegant I secretly suspect you of
> being a physicist or perhaps even a mathematician!
Thank you (though I prefer the madder branches of Science).
But I think I should clarify, I wasn't necessarily suggesting that
we actually *implement* the resolver/validator as a data-source or
even write a data-source API front-end to it (though, yeah, that would be
interesting). I just meant we shouldn't overfocus the API semantics
on the specific needs of authoritative zone databases. We should try
instead to address the general problem of "getting data from several
different places".
There are two general ways to do that, serial and parallel; I'd like to
see a framework that supports both. Something like: "query data sources
{A, B, C} simultaneously, wait for all of them to answer or timeout,
select the best answer; if there is no satisfactory answer, query data
source D; if there is still no satisfactory answer, call function E,"
and so forth.
If the resulting design is flexible enough that it *could* be applied to
views and resolvers as well as to an authoritative server, then I think
we'll have done our jobs right. We may not ultimately choose to implement
things that way, but we'll have a design that can take whatever we do throw
at it.
eh
More information about the bind10-dev
mailing list