Re: [exim] Feature requests for eximstats: Receipt time and …

Steve Campbell wrote:
>
> This is the right place to post feature requests.
>
> 1) Charting receipt time for messages.
> Not sure exactly what you want here. Have you got any sample logs and
> results?

Sure, here's a sample log for a single message:

2005-04-05 10:44:02 1DIpGz-0003Tk-4i <= sender@???
H=mail.nowtoronto.com [64.26.184.36] P=esmtp S=2448364
id=5.1.0.14.2.20050405104448.00a9b8a0@???
2005-04-05 10:45:01 1DIpGz-0003Tk-4i => rcpt1@???
<rcpt-orig@???> R=dnslookup T=remote_smtp H=mx3.mail.yahoo.com
[64.156.215.6] QT=1m40s DT=59s
2005-04-05 10:45:07 1DIpGz-0003Tk-4i => rcpt2@???
<rcpt-orig@???> R=dnslookup T=remote_smtp
H=gsmtp171.google.com [64.233.171.27] QT=1m46s DT=1m5s
2005-04-05 10:45:07 1DIpGz-0003Tk-4i Completed QT=1m46s

The log was collected with +queue_time +queue_time_overall +delivery_time.

Currently, you chart as "Time spent on the queue" the difference between
the first (<=) and final (Completed) log entries for the message: 1m5s.
Exim's queue_time_overall logs on the Completed entry the total time
spent in receipt and the queue, as QT=1m46s. So the Receipt time for
the message is 1m46s-1m5s=41s. I'd like that time charted for all
messages in a duration-bucket table (count, %, cumul. %)/pie chart, just
like "Time spent on the queue" is currently. Splitting it up by remote
protocol (P=* on the <= log entry) seems like it might be a good idea,
too, just as the "Time spent on the queue" chart is currently split
between all messages and messages with at least one remote delivery.
Does that make sense?

Doing the same for the DT times logged by +delivery_time seems like a
good idea, as well (no calculation required here -- just chart 59s and
1m5s for the two deliveries respectively. These would be charted as
"all deliveries" and "remote deliveries only".

> 2) Histograms for the various rejection types and/or patterns.
> I'm not too keen on doing this by rejection type, as we would end
> up with an arbitrary number of report sections which could get
> quite big. Doing it for patterns is doable though. Can you confirm
> that we are talking about histograms by time here, the same as
> exists for messages received and delivered?

Yes, a time bucket histogram with .'s and line graph, just like messages
received per hour and deliveries per hour.

Thanks,
- Marc