Mail gateway server with amavis on CentOS 8

[Nikolaos Milas]

On 28/12/2020 1:35 μ.μ., Nikolaos Milas wrote:

> Why do the two servers produce so different results in terms of BAYES 
> filtering (BAYES_999=2.7, BAYES_99=3.7 vs BAYES_00=-1.9)?

Here is an update regarding my latest efforts. I tried:

1. to move content of /etc/mail/spamassassin/local.cf to 
/var/spool/amavisd/var/.spamassassin/user_prefs

2. To clear and re-create the bayes db:

    sh-4.4$ sa-learn --dbpath /var/spool/amavisd/var/.spamassassin/ --clear
    sh-4.4$ ls -la var/.spamassassin/
    total 12
    drwx------. 2 amavis amavis 4096 Dec 28 23:18 .
    drwx------. 4 amavis amavis 4096 Dec 16 16:53 ..
    -rw-r--r--. 1 amavis amavis  299 Dec 28 23:15 user_prefs
    sh-4.4$
    sh-4.4$ sa-learn --dbpath /var/spool/amavisd/var/.spamassassin/
    --restore bayes-mailgw3.db
    sh-4.4$ ls -la var/.spamassassin/
    total 8624
    drwx------. 2 amavis amavis    4096 Dec 28 23:19 .
    drwx------. 4 amavis amavis    4096 Dec 16 16:53 ..
    -rw-------. 1 amavis amavis  659456 Dec 28 23:19 bayes_seen
    -rw-------. 1 amavis amavis 9490432 Dec 28 23:19 bayes_toks
    -rw-r--r--. 1 amavis amavis     299 Dec 28 23:15 user_prefs

Then I tried re-sending the test email; however, none of the above 
changed anything. The test mail got through with the same scoring.

I am wondering: Could it be that the Bayes db is NOT in fact recoverable 
onto another machine? I mean that, although it technically gets 
recovered, in essence it may not be usable on a different box, i.e. on a 
different mail server, in which case we would need to clear the bayes db 
and train using original spam and ham messages (still from other mail 
servers). Could that make a difference?

Thanks,
Nick