On Tue, Jul 14, 2015 at 09:19:02PM -0700, E.B. wrote:
> > However, all the requests from a single smtpd(8) service will go
> > to the same policy service instance. So in memory caching can be
> > used to construct per-transaction state (if, for example, the policy
> > service fires for each recipient and also at data and/or end of
> > data).
>
> That's guaranteed? thats very useful! I saw mailing list threads having
> that problem - connecting the state where recpients are known to the
> state where total number of recipients is known - so was this recent
> change?
The only exception would be a premature exit by a spawned policy
service (make sure to set a large enough maximum time limit so that
spawn does not kill these prematurely). If the policy service
disconnects unilaterally, smtpd(8) might then reconnect to some
other policy service mid-transaction. Otherwise, IIRC the same
policy service connection is re-used for the lifetime of the
process.
--
Viktor.
