Developers of Sympa

[Jeff Abbott <address@concealed>]

On Nov 29, 2005, at 8:35 PM, Adam Bernstein wrote:

> I've successfully gotten VERP working with Sympa 5.1 and Postfix  
> 2.0.18,
> although what I mean by "gotten it working" might not quite be what
> you'd think.

This is still quite intriguing for me, nonetheless.

> I haven't yet done anything to bounced.pl to get it to
> parse out the VERPed return-path for automated bounce processing,  
> which
> should just be a small bit of straightforward programming, but that
> wasn't our primary need.

Unfortunately it /is/ our primary need from VERP.  The problem we run  
into a lot is that people will subscribe their address@concealed  
alias to a list, which forwards to their address@concealed address  
internally.  This then forwards to some arbitrary address, set  
through a web interface we've got, that the user picked.  Often it's  
a hotmail.com or aol.com address, but just as often it's something  
completely random.  Because we're using Majordomo, bounce reports get  
sent to the list owner.  If they're not immediately terrified of the  
"technical junk" that they see in such a message, they'll usually try  
to figure out what address bounced and unsubscribe it from the list.   
Needless to say, this does not work as the address is not subscribed  
to the list.  They then come to me and I have to dig through several  
thousand sieve scripts that users have installed to see if I can find  
the address.  If I can't, which has happened a couple of times as the  
user was forwarding their off-site address to yet another address,  
then there's really nothing we can do.

I probably end up sorting through about 5-8 such "I'm having problems  
unsubscribing this bouncing user" a week.  So, VERP is something  
we're quite interested in.

> If someone else wants to add the few lines
> of code in bounced.pl that will read the subscribed address from the
> headers instead of trying to parse it out of the message body of a
> bounce, to eliminate all the "Subscriber <> not found in list <>"  
> errors
> and process all bounces successfully, we'd be happy not to do it
> ourselves.

If no one's been able to get to this in a couple of months, once  
we've gotten Sympa deployed here, then I'll be more than happy to  
look at it or add some additional hand/eye/brain power to the problem.

> Also, keep in mind that this means Postfix will now make a separate  
> SMTP
> connection for each recipient, instead of batching them as  
> determined by
> Sympa's nrcpt parameter and Postfix's
> default_destination_recipient_limit parameter.  So delivery takes  
> longer
> -- on our system, with nrcpt=25 and Postfix at its default, VERPed
> delivery takes about 3 times longer than before.  This is significant
> for lists of 10k or 50k subscribers (of which we have a few), but
> overall it's quite acceptable.

This would also be a problem for us, too, with some of our larger lists.

What would be ideal, and would be something I'd be willing to pitch  
in towards, is a "use_verp" per-list setting, with a site- or robot- 
level default.  This way we could set the default to "true" for the  
majority of our lists, and the small number of Big Ones we've got  
could have that setting disabled for performance reasons.  This, of  
course, makes the problem a much bigger one to solve, requiring  
changes at many levels of the software, but I think it would be worth  
the effort.  The question then is:  Is this even feasible?

Thanks a bunch for doing the early legwork on this, Adam!

Jeff

Attachment: smime.p7s
Description: S/MIME cryptographic signature