[GRASS5] Spamfilter added for RT Bugtracker mails

GRASS GRASS developer
1

To get rid (hopefully) of most spams arriving in 'grass5'
mailing list via bugtracker, I have added a filter rule
to pipe RT mails through 'bogofilter':

Reply-To: guest user via RT <grass-bugs@intevation.de>
From: guest user via RT <grass-bugs@intevation.de>
To:
Cc: grass5@grass.itc.it
X-Request-ID: 2174
X-RT-Loop-Prevention: bug
X-Sender: guest
X-Managed-By: Request Tracker 1.0.7 (http://www.fsck.com/projects/rt)
X-Bogosity: No, tests=bogofilter, spamicity=0.000000, version=0.15.13.1
   -> this line was added

Subject: [GRASS5] [bug #2174] (grass) Mac OSX startup: ERROR: Invalid return code
Errors-To: grass5-admin@grass.itc.it
X-BeenThere: grass5@grass.itc.it
X-Mailman-Version: 2.0.5
List-Help: <mailto:grass5-request@grass.itc.it?subject=help>
List-Post: <mailto:grass5@grass.itc.it>
List-Subscribe: <http://grass.itc.it/mailman/listinfo/grass5&gt;,

The same filter has been applied also to 'weblist' (mailing list
dedicated to receive web site comments and requests).

Let me know in case of problems

Markus

2

On Mon, Jan 12, 2004 at 10:03:33AM +0100, Markus Neteler wrote:

To get rid (hopefully) of most spams arriving in 'grass5'
mailing list via bugtracker, I have added a filter rule
to pipe RT mails through 'bogofilter':

Note that we already have a spam filter running
here at Intevation before we accept entries in the bugtracker.
We are running spamassasin.
It also contains a bayesian spam filter component
in additions to other methods.
As you can see there is no total protection.
However we tuned it to minimise false positives.

The same filter has been applied also to 'weblist' (mailing list
dedicated to receive web site comments and requests).

Let me know in case of problems

Let us know the number of false positives over the next month.
I'd rather have 10 false negatives than 1 false positive.

3

On Mon, Jan 12, 2004 at 10:46:32AM +0100, Bernhard Reiter wrote:

On Mon, Jan 12, 2004 at 10:03:33AM +0100, Markus Neteler wrote:
> To get rid (hopefully) of most spams arriving in 'grass5'
> mailing list via bugtracker, I have added a filter rule
> to pipe RT mails through 'bogofilter':

Note that we already have a spam filter running
here at Intevation before we accept entries in the bugtracker.
We are running spamassasin.
It also contains a bayesian spam filter component
in additions to other methods.
As you can see there is no total protection.
However we tuned it to minimise false positives.

Maybe that's the problem - it's not catching some obvious
mails. Note that I have additionally implemented some 60
rules in procmail to catch further spams (including some
missed from RT spamassasin).

Started in August 2002, these procmail rules have been catching
more than 2700 spams. A major help for me as I didn't had to
eliminate them from mailman's admin tool (otherwise mailman
would catch them except for RT mails).

> The same filter has been applied also to 'weblist' (mailing list
> dedicated to receive web site comments and requests).

On 'weblist' list, bogofilter has caught already 8 mails since
installation some hours ago.

> Let me know in case of problems

Let us know the number of false positives over the next month.
I'd rather have 10 false negatives than 1 false positive.

Will do. It was trained on 6000 spams (filtered from my personal box)
and some thousand hams. So far I had 2 false positives which is
an exciting good ratio. I'm pretty happy with bogofilter
and want to let you participate :slight_smile:

So please don't feel offended, RT is configured well. I don't
see a problem to filter again, just to minimize the remaining
rubbish passing RT spamassasin in that current configuration.

Markus

4

On Mon, Jan 12, 2004 at 11:17:27AM +0100, Markus Neteler wrote:

On Mon, Jan 12, 2004 at 10:46:32AM +0100, Bernhard Reiter wrote:
> On Mon, Jan 12, 2004 at 10:03:33AM +0100, Markus Neteler wrote:
> > To get rid (hopefully) of most spams arriving in 'grass5'
> > mailing list via bugtracker, I have added a filter rule
> > to pipe RT mails through 'bogofilter':
>
> Note that we already have a spam filter running
> here at Intevation before we accept entries in the bugtracker.
> We are running spamassasin.
> It also contains a bayesian spam filter component
> in additions to other methods.
> As you can see there is no total protection.
> However we tuned it to minimise false positives.

Maybe that's the problem - it's not catching some obvious
mails. Note that I have additionally implemented some 60
rules in procmail to catch further spams (including some
missed from RT spamassasin).

Oh nice.
I didn't know that you were putting so much effort into the tuning. :slight_smile:

Started in August 2002, these procmail rules have been catching
more than 2700 spams. A major help for me as I didn't had to
eliminate them from mailman's admin tool (otherwise mailman
would catch them except for RT mails).

I've implemented something similiar for the FFII and its mailman
2.0.x installation. Hold spam message will automatically be removed.
2.1.4 versions might be able to do this themselfs.

> Let us know the number of false positives over the next month.
> I'd rather have 10 false negatives than 1 false positive.

Will do. It was trained on 6000 spams (filtered from my personal box)
and some thousand hams. So far I had 2 false positives which is
an exciting good ratio. I'm pretty happy with bogofilter
and want to let you participate :slight_smile:

So please don't feel offended, RT is configured well. I don't
see a problem to filter again, just to minimize the remaining
rubbish passing RT spamassasin in that current configuration.

No problem.

5

On Mon, Jan 12, 2004 at 02:29:13PM +0100, Bernhard Reiter wrote:

On Mon, Jan 12, 2004 at 11:17:27AM +0100, Markus Neteler wrote:
> On Mon, Jan 12, 2004 at 10:46:32AM +0100, Bernhard Reiter wrote:
> > On Mon, Jan 12, 2004 at 10:03:33AM +0100, Markus Neteler wrote:
> > > To get rid (hopefully) of most spams arriving in 'grass5'
> > > mailing list via bugtracker, I have added a filter rule
> > > to pipe RT mails through 'bogofilter':
> >
> > Note that we already have a spam filter running
> > here at Intevation before we accept entries in the bugtracker.
> > We are running spamassasin.
> > It also contains a bayesian spam filter component
> > in additions to other methods.
> > As you can see there is no total protection.
> > However we tuned it to minimise false positives.
>
> Maybe that's the problem - it's not catching some obvious
> mails. Note that I have additionally implemented some 60
> rules in procmail to catch further spams (including some
> missed from RT spamassasin).

Oh nice.
I didn't know that you were putting so much effort into the tuning. :slight_smile:

Well, it's better than deleting thousands of messages manually
from the mailman queue...

> Started in August 2002, these procmail rules have been catching
> more than 2700 spams. A major help for me as I didn't had to
> eliminate them from mailman's admin tool (otherwise mailman
> would catch them except for RT mails).

I've implemented something similiar for the FFII and its mailman
2.0.x installation. Hold spam message will automatically be removed.
2.1.4 versions might be able to do this themselfs.

Sounds good - but as a more recent Mailman version needs python 2.x
this would require a complete reinstallation of 'grass.itc.it'.
I hesitate to reinstall it (kernel is updated if needed of course).
Never touch ...

[...ok ... ]

Markus

6

On Mon, Jan 12, 2004 at 02:40:38PM +0100, Markus Neteler wrote:

On Mon, Jan 12, 2004 at 02:29:13PM +0100, Bernhard Reiter wrote:

> I've implemented something similiar for the FFII and its mailman
> 2.0.x installation. Hold spam message will automatically be removed.

If you are interested I can explain how I did this for Mailman 2.0.x.
It is cronjob, just moving the marked emails away from the data directory.

> 2.1.4 versions might be able to do this themselfs.

Sounds good - but as a more recent Mailman version needs python 2.x
this would require a complete reinstallation of 'grass.itc.it'.
I hesitate to reinstall it (kernel is updated if needed of course).
Never touch ...