The mailing list for listmasters using Sympa

[Chris Hastie <address@concealed>]

On Thu, 12 Aug 2004, Olivier Salaun - CRU <address@concealed> wrote:

> > Anyway, after looking through the code I have some questions about
> > whether I've interpreted the actions correctly in a couple of areas.
> >
> > Firstly, the Type Rate. Am I right in thinking that a user could get
> > 50 4xx bounces followed by one 5xx and find the whole 51 multiplied by 1?
>
> You're right. The alternative would be to reset the bounce information
> (related to a subscriber) if we receive a bounce with a type different
> than the previous one.
>
> > Secondly, regularity rate and digest readers. If a subscriber receives
> > (or rather doesn't) a daily digest on a list with, say, 30 posts per
> > day and ALL his digests bounce, will his regularity rate appear to be
> > 0.033 despite the fact that everything bounces? In which case, it will
> > be a long while before digest bouncers get removed!
>
> Yes it will take longer to get him removed but :
>
>    1. we have fewer bouncing data for this person
>    2. we have "wasted" less ressources for this bouncing user
>
> Do you have a proposal to improve any of these aspectes of the bounce
> management system ?

Thanks for clarifying that Oliver. The only firm ideas I have right now relate
to documentation - it might be useful if these issues were mentioned, and in
particular the issue that drove me to poke around in the code in the first
place - the units for the minimum_bouncing_period paramater.

From my point of view 10 days is a rather high value for the default for
minimum_bouncing_period. A user could have produced 200 bounces before they
were counted. But I guess if I was running a weekly newsletter it would be
about right. So picking a default that makes sense to everybody is not really
possible.

I rather like the basic algorithm used. I guess it could be refined by, for
example, keeping separate tallies of permanent and transient failures and
calculating the regularity rate for digest / summary users against the numbers
of digests or summaries rather than the number of normal messages. But from my
initial look at the code I would say both refinements require rather a lot of
effort for marginal improvements. May be if I find myself really short of
things to do...

The one thing easy that I might be tempted to change is the type rate for
transient failures. I might just push this down a bit - I often see users
produce large volumes of transient failures for a couple of days whilst some
temporary routing problem exists, but all comes back fine eventually.

Related to this, is there any mechanism to reset the bounce count / score if a
user hasn't produces any bounces for a while? Or is it just allowed to drop
down as the regularity rate drops?

--
Chris Hastie