Hi, Slightly off topic, but something that needs to be done with PMG - I have been using EFA for a few years now and have a considerable amount of Bayes Database on the EFA Servers, which may have some false positives, but has a lot of SPAM learning as well as learning of SPAM / HAM that was done by pressing the menu options there. I would like to migrate the Bayes Database to the PMG server I have setup, any help / tips on this?
Migrating Bayes Database
- Thread starter thebiggeek
- Start date
Since I assume that the thread was meant for the Mail Gateway Forum - I moved it there.
Else check out the wiki-page from SpamAssassin:
https://cwiki.apache.org/confluence/display/SPAMASSASSIN/BayesMigration
(PMG stores the bayes db on each node in /root/.spamassassin/)
I hope this helps!
Else check out the wiki-page from SpamAssassin:
https://cwiki.apache.org/confluence/display/SPAMASSASSIN/BayesMigration
(PMG stores the bayes db on each node in /root/.spamassassin/)
I hope this helps!
Thank you @Stoiko Ivanov for moving this - apologies, I made the posting at the wrong place. Thank you for your help. I wanted to check with users such as @heutger or @killmasta93 if they had done this also - to see what their response is. For now I was getting a lot of emails that went through because of Bayes so I had to stop Bayes. I will read the documentation.
Bayes DB on EFA is in MYSQL format, so will need to figure out how to move this. Thanks for your input that it is kept on each node.
Bayes DB on EFA is in MYSQL format, so will need to figure out how to move this. Thanks for your input that it is kept on each node.
In my milter-reject setup, I also copied the bayes database, that worked fine. I'm unsure, if it has different format, that may occur in problems, but if you get bayes database in format like bayes.mutex, bayes_journal, bayes_seen, bayes_toks similar to what PMG is currently using, you may be able to import your recent bayes database.
Thank you @heutger - I am still debating if I need the Bayes or not, as I see, that every time I enable Bayes based filtering a lot of spam goes through, this is because users are accepting those emails and not setting up filters. I am still investigating what will be the best manner to go forward with Bayes. I am still investigating what will be my learning mechanism. This week am deploying outbound emails to also go via PMG and hopefully that will help improve the learning a bit - but for now I am not sure on Bayes
Bayes is based on training. So that's why I already asked Proxmox team to integrate the possibility to learn spam and ham minimum via quarantine (if used) or similar to rspamd by entry fields, where raw messages can be posted like spamcop ui also does. However, the best way would to train the users to move spam to spam and move spam out of spam and use e.g. sieve to invoke scripts for sa-learn on PMG. Alternative would be a shared mailbox to copy spam there to be able to fetch spam from there and train it that way.
The "built-in" alternative is to wait years until 200 spam and 200 ham has been autolearned by the system. Maybe also using outgoing filtering may help here, however, if you have anyone on your infrastructure providing (e.g. insecure scripts) or forwarding (e.g. from external sources) spam through your infrastructure, you won't have any benefits. Also most important is to learn spam, ham autolearning is working well, but spam should be trained manually, minimum or latest from your own mailbox.