Proposal for using SpamAssassin in master.d.o [Was: Re: Every spam is sacred]
Hi.
I will describe the method we have implemented (almost completed, just waiting
for an IP ;) in HispaLinux for an installation with virtual users and
spam/ham/forget mboxes upload.
The system uses a standard SpamAssassin installation, based on Duncan
Findlay's debs.
It is configured to use virtual users, which store the preferences files and
bayesian databases under /home/spamassassin/%username/
That is configured using spamd with the right flags (read the doc for more
info).
For the upload to happen, an infrastructure similar to the one in ftpmaster
has been put in place:
the user uploads a user.name.tar.gz and a user.name.tar.gz.asc file containing
the signature of the original file.
An script running in the server moves the file to a place where it can be:
unpacked and verified
split in individual files
clean from previous spamassassin tags
introduced into the bayes database
all as per user basis.
Debian has almost all the infraestructure to set the process:
The gpg ring can be used to verify the uploaded files.
The rest is code... ;)
Furthermore, global or local rules can be introduced for proposals like the
one Santiago Vila proposed: use the RBL's signature to increase/maintain the
score (which can be set using LDAP or another mechanism).
The main problem this implementation has is the high load the cleaning
produces, which I really do not understand, since it takes more time to clean
than to analyse and introduce the tags... :(
I have been trying to contact some developers to discuss the issue on IRC but
seems we have opposite bed shifts.
mooch
--
Jesus Climent | Unix SysAdm | Helsinki, Finland | pumuki.hispalinux.es
GPG: 1024D/86946D69 BB64 2339 1CAA 7064 E429 7E18 66FC 1D7F 8694 6D69
----------------------------------------------------------------------
Registered Linux user #66350 proudly using Debian Sid & Linux 2.4.20
You know what the real tragedy of this day is? I'm not even supposed to
be here today!
--Dante Hicks (Clerks)
Reply to: