Donovan Bray:
> Can you confirm my understanding of the following settings.
No. You don't solve problems by playing with scheduler feedback.
The entire concept of feedback is largely bogus, as shown by the
measurements in SCHEDULER_README.html.
Wietse
> initial_destination_concurrency = 5
> default_destination_concurrency_limit = 20
> default_destination_concurrency_positive_feedback = 1/2
> default_destination_concurrency_negative_feedback = 1
> default_destination_concurrency_failed_cohort_limit = 1
>
> This would cause the initial concurrency to begin at 5
> It would take two successful psuedo-cohorts before it increased the
> concurrency to 6, effectively halving the rate of growth of positive
> concurrency.
> It would take 30 successful psuedo-cohorts to reach the maximum concurrency
> of 20.
> It would take a single failed cohort to drop the concurrency by 1. or would
> it be marked dead on the first failed cohort, and restart later back at 5?
>
> Is there any way to make it drop concurrency by MORE than one? If I
> understand the docs saying the following is invalid:
>
> default_destination_concurrency_negative_feedback = 3/1
>
>
> On Tue, Aug 16, 2011 at 8:50 AM, Donovan Bray <donno...@gmail.com> wrote:
>
> > How about using initial concurrency, negative and positive feedback so that
> > dynamically it loses concurrency faster on failure than it gains on success?
> >
> > I've seen no examples of how to configure it other than the defaults
> > specified.
> >
> > On Aug 15, 2011, at 4:59 PM, Wietse Venema <wie...@porcupine.org> wrote:
> >
> > >> I have several domains that I need to deliver for; and I use
> > >> sender_dependent_default_transport_maps to map those to the correct IP's
> > to
> > >> send out of.
> > >>
> > >> I am experiencing a couple of related domains begin to defer our mail
> > with
> > >> with 450 and 421's.
> > >>
> > >> Aug 15 10:10:13 server postfix/smtp[27375]: B2A93E39647: host
> > >> scc-mailrelay.att.net[204.127.208.75] refused to talk to me: 450
> > xx.xx.xx.xx
> > >> has too many connections ( 3 ) on sccwmxc01
> > >
> > > You can limit the per-destination concurrency with:
> > >
> > > /etc/postfix/main.cf:
> > > xxx_destination_concurrency_limit = 2
> > >
> > > where xxx is the name of a mail delivery transport (first field in
> > > master.cf).
> > >
> > > Or you can just ignore the issue and let Postfix deliver the mail
> > > on the next delivery attempt.
> > >
> > > I have no quick solution for (per-sender source IP address) AND
> > > (destination-dependent concurrency limit). If you need that amount
> > > of control, then separate Postfix instances are required.
> > >
> > > Wietse
> >
