On Mon, Oct 10, 2022 at 10:27:37AM +0200, juan smitt wrote:
> How can one directly monitor the effectiveness of the increased number
> of Postfix SMTP server processes?
> (https://www.postfix.org/STRESS_README.html)
>
> Is it true that this scales up dynamically until the limit,
Yes, a new smtpd(8) process is started on demand, if a new connection
comes in while all extant processes are busy. Once started, each
process handles up to $max_use connections, exiting after that, or
when idle for at least $max_idle.
> which means I can monitor it indirectly by counting the number of
> processes (and also checking the delay in the mail log)?
The delay times in the mail log are not a direct indication of either
too many or too few smtpd(8) processes.
Instead, you can for smtpd(8) processes with "-o stress=yes", these
should not be too frequent. To measure whether inbound mail is delayed
before it gets to your system you'd need to parse the timestamp in
the most recent remote "Received:" header and (assuming that timestamp
is reasonably accurate) infer the SMTP hop delay.
I'm not recommending you immediately plan to do that, just saying that's
what it would take. The signal would be noisy, remote delays aren't
always your fault, there could be local congestion on the remote end.
--
Viktor.
