OK.  I deleted the .qmail file from all accounts.
As well, I "reset" the spamassassin database using "sudo -H -u vpopmail
sa-learn --clear"
I followed this with a "qmailctl restart"
 
I am still getting a huge amount of false positives, and it is still
dropping the ***SPAM*** marked emails into the .Spam box.
Once messages are in there, it seems to be impossible to get them back
into the inbox (and stay there).
 
Any advice?
 
Thank you,
Patrick M. Ring
P. Ring Technologies
Louisiana Web Host, LLC.
985-868-4200
 

________________________________

From: Jake Vickers [mailto:j...@qmailtoaster.com] 
Sent: Wednesday, July 21, 2010 1:34 AM
To: qmailtoaster-list@qmailtoaster.com
Subject: Re: [qmailtoaster] Spambox troubles


On 07/20/2010 05:54 PM, Patrick Ring wrote: 

        I've been running the Spambox on my qmt for quite some time now.
However, lately I'm getting a huge amount of false positives, and it's a
pain trying to educate the users about how to use the Spam (or .Spam)
folder.
         
        Is there any (simple) way to turn it off?
        In the reading I've done, the only thing I can find is to
recompile/rebuild a package or two with the spambox 0 defined.
        I've also read it has something to do with the .qmail file found
in each users' account.
         
        Can anyone point me to some enlightenment about this?
        
        


If you are getting false positives, check the message headers. I'll bet
your bayes DB has been tainted with legitimate messages that were marked
as spam, which will have a snowball effect. You'll see legitimate
messages hitting the bayes score categories.

Basically when the spambox is enabled for a user, after all the spam
scans are done the message gets passed to the /etc/mail/mailfilter
script (maildrop). If the message has a spam score in the headers that
is greater than 7 (I think this is the default now) then it gets sent to
the Spam folder and also sent to sa-learn so that spamassassin can
analyze it and learn to identify messages like it as spam more
efficiently. So if a legitimate message gets sent to sa-learn as spam,
spamassassin will be tainted - this will have a snowball effect.

When you enable the spambox for users, it creates a .qmail file in the
user's maildir. This tells it to pass the messages to the mailfilter
script before delivery. You have a couple ways to fix this:

Delete the bayes DB and start over. You may see an increase in spam
initially until spamassassin can learn again (brain-wipe it), but this
is usually preferable to false-positives.

Disable the spambox by recompiling qmailadmin-toaster to disable the
option - this will not REMOVE it for those that already have the .qmail
file - it will only prevent people from turning the option on. You will
then need to go and delete the .qmail files for the users to actually
disable it for users.

Delete the bayes DB, starting over like before, and adjust the
mailfilter script to fit your environment a little better.

Write a spamassassin rule that adds a negative score for things that you
identify as legitimate message contents - eventually this will counter
the tainted bayes, but it will take a LONG time.

After checking the headers and seeing the BAYES_99 category, change it's
score. This can be used in conjunction with some of the options above.

Hope that helps some.

Reply via email to