Re: [mlmmj] Problems with microsoft

[Ben Schmidt <mail_ben_schmidt@yahoo.com.au>]

Hi, Christian,

Here are some thoughts:

> I use mlmmj to send newsletters, and I have the following problem, but
> now more serious since Microsoft have blocked the server.

That's annoying. :-)

> Problem 1.
> Our mail recipients sometimes receives a bunch of mails with the
> following message "some messaged could not be delivered. If you see
> this things are back to normal." They get many of these, so they
> complain.
> I understand that they get one mail telling them that everything
> Works, but why 10-40?

It would be helpful to find out where those mails originate--is Mlmmj
sending a lot of mail to Postfix, or is Postfix sending a lot of mail to
the next server? There are a few things to look at: (1) your mail logs;
you should be able to find the messages going out, (2)
mlmmj.operations.log in the relevant listdir, (3)
mlmmj-maintd.lastrun.log in the relevant listdir (if you get to it
quickly enough after it happens), (4) the Message-ID and other headers
of the received 'duplicate' messages.

Perhaps you could furnish us with some of that information. De-identify
it by making some small modifications to the email/IP/list addresses in
it if necessary.

> Is it a configuration problem between mlmmj and postfix?

Possibly.

> I have the following settings for postfix and mlmmj that I think is
> relevant for the problem. But i don't really understand how they
> interact, could there be some configuration error so mlmmj fills
> postfix with a queue due to lack of respone.
>
> * /etc/postfix/main.cf
> bounce_queue_lifetime = 2d
> minimal_backoff_time = 1800s

This could possibly be relevant if Postfix is trying to send mail to
mlmmj-receive, succeeding, but receiving a failure response; Postfix
will keep trying for 2 days to deliver the bounce message to Mlmmj,
which will keep receiving it and keep thinking the address is bouncing,
and keep sending bounce probes.

You should be able to determine from your mail logs if this is happening
(you will see a lot of failed messages from Postfix itself--postmaster,
or mail_daemon or whatever it uses--to list+bounces addresses).

It would also be helpful to know how Postfix and Mlmmj are linked? What
do you have in your config files to facilitate delivery of messages to
Mlmmj?

> * in  'tunables' bouncelife
> 2592000
> (30 days)

This shouldn't be too relevant; it's how long Mlmmj waits before giving
up and unsubscribing the user. If you changed this, bounce probes would
just turn into unsubscriptions; the cause of the problems wouldn't be
addressed.

> Problem 2
> Microsoft think I am doing 'namespace mining', I know I don't. but
> maybe it is somehow connected to the problem above ?

If this is truly the case, it is probably unrelated to the problem
above. If you are namespace mining, you are trying lots of addresses
@hotmail.com (or wherever), hoping to find real ones. In fact, you are
probably actually getting a lot of bounces for nonexistent addresses.
So, if, as a responsible mail host, I want to detect if you're namespace
mining, I would use the number of bounces due to nonexistent addresses
as a heuristic, and block you if you get a lot of them.

A nonexistent address can't receive a lot of probe messages! It can't
receive anything. So it's probably not related to the problem above.

However, it could be related to your bouncelife tunable. If an address
ceases to exist, because it's deleted; or in some cases, if an address
with wrong spelling is added to the list (e.g. without requiring
confirmation), Mlmmj is going to receive a bounce message about the
non-existent address. However, Mlmmj doesn't know whether that bounce
message is a permanent error or a temporary error (and in fact,
sometimes, due to misconfiguration, errors that seem permanent are
actually temporary, so best retried anyway). Therefore Mlmmj will keep
retrying the address--possibly every 2 hours (however often mlmmj-maintd
runs; I'm not sure if Mlmmj throttles delivery or not) for *30 days*,
and every time receive a bounce message due to the non-existent address.
That many bounces for non-existent addresses would definitely make you
look like you're namespace mining (if the watchdog software that uses
the heuristic isn't smart enough to realise they're all for the same
address or few addresses).

You can check whether this is happening by looking in your listdir. The
last bounce for each currently-bouncing address is stored in the bounce
subdir, so you can read them, and see how many addresses (including how
many from Microsoft) are bouncing, and why.

One reason for doing automatic bounce processing is to minimise
unnecessary bounces, by unsubscribing users before bounces to them
become suspicious or waste too much bandwidth; by making your bouncelife
so high, you've reduced the effectiveness of the feature.

> Best Regards and I hope you can help me.

No trouble. If you need more help, please furnish us with more
information: mail logs, mlmmj logs, message headers, configuration. All
this information is useful and necessary for properly tracking down
these problems.

Ben.