Robert Schetterer:
> master.cf
> ...
> slow unix - - n - - smtp
>
> main.cf
> slow_destination_concurrency_limit = 3
> slow_destination_rate_delay = 1s
> slow_destination_recipient_limit = 6
With slow_destination_recipient_limit > 1, the scheduler controls
delays, concurrencies, etc. per domain (instead of per recipient).
Therefore, different domains are delivered in parallel, subject to
the master.cf process limit.
With slow_destination_rate_delay > 0, the per-domain concurrency
(per-domain because slow_destination_recipient_limit > 1) is always
1, otherwise there can be no delay between deliveries to the same
domain. Therefore, slow_destination_concurrency_limit has no effect.
If you set slow_destination_rate_delay > 0, then you should also
increase the slow_destination_concurrency_failed_cohort_limit
as described in QSHAPE_README.
Wietse
