Jason L Tibbitts III <tibbs@hpc.uh.edu> writes:
> can certainly pick up any bounces then. Once we see any kind of bounce
> from an address, we can give it special treatment during delivery and
> always do a VERP for it.
Wouldn't it suffice to do this after we get an unparseable probe for that
address?
In addition to the above, I'd like to send subscribers who missed any mail
from any list one warning message (per day) saying for which messages from
which lists bounces were received during the last 24 hours (and when the
functionality is available, also give them a token that gives an easy way
for getting a digest of all missed messages). Sending such warning messages
also helps a lot with designing a good procedure for deciding when an address
is put into the BOUNCE class. (Otherwise we have a problem on announcement
lists which send only one message every N days, where N is bigger than the
timeout for bounce data.)
> The decision about whether to put an address in the BOUNCE class is a tough
> one; I'd say that some number of messages in sequence should be good
> enough. (If you're not bouncing every message, you're not really
> bouncing.)
The flaw with this procedure is that it assumes mail transport from the
machine with the bouncing address to the mailing list host to be reliable.
Such an assumption would be particularly fatal in the case that a machine
is flooded with so many bounces that it cannot process them all. Since
there'd never be enough bounced messages in sequence to warrent moving the
bad addresses to the BOUNCE class, the situation would only be getting
worse all the time.
> NB> As I said before, I'm willing to follow whatever specifications you may
> NB> want to give me describing requirements for the bounce-handler of Mj2.
>
> I'd of course like to hear your ideas on the matter; I have to keep the
> whole of the package in my head while you're only concentrating on one
> piece so it makes sense that you'd have better ideas about bounce stuff
> than I would.
Ok... I'll write up some ideas, please don't hold your breath though...
right now I'm very busy and I can hardly find time to move that Bouncefilter
version (which I've been promising for a while now) forward enough so that I
can give it out with good conscience. But as soon as that is done I'll work
on a draft specification of what the bounce-handler of Mj2 should be like.
(I don't really care whether it'll be called Bouncefilter or not as long as
it shares most of its modules with the stand-alone version of Bouncefilter.)
-- NB.
References:
|
|
