On Wed, Dec 14, 2022 at 10:43:14AM -0500, Wietse Venema wrote:
> > > query = update send_counts ...
> >
> > How exactly do you imagine integrating such a query into Postfix?
> >
> > [ Hint using this as a transport table may not do what you think
> > it does. Transport lookups happen per-recipient and in both
> > smtpd(8) and qmgr(8) where lookups happen each time a message
> > enters the active queue (and before actual delivery). ]
> >
> > And is sending a small batch of messages and then stopping really a more
> > effective approach to avoid problems due to lack of "reputation" with
> > new new IPs?
>
> I suppose that the idea is to rate-limit traffic based on IP
> reputation. Just using randmap to send 1% of a million could still
> be too much if the deliveries happen in a short time.
>
> This would take a transport map whose response depends on how much
> mail has been delivered (over an interval of time). The number of
> queries is an estimator that may be off by a factor but that would
> be OK if that factor is constant.
Perhaps, but I'd rather expect the operator to have a good idea of the
expected full traffic rate, and if needed choose initial fractions that
are less than 1% (the prototype Postgres RNG can easily be tuned for
more fine-grained initial sampling, 0.1% or whatever).
Counting messages feels too complex to me, but sure so long as one
understands that transport queries over-estimate deliveries, whatever
rocks one's boat.
--
Viktor.
