Migrating our Debian Maia Anti-Spam Anti-Virus Gateway Email Server's Bayes database to MySQL


This document assumes you have installed Maia using one of the Debian documents at http://verchick.com/mecham/public_html/spam/.

Navigate to where your Bayes data is currently stored:
cd /var/lib/amavis/.spamassassin

As the 'amavis' user (or the user that currently own the Bayes data), we need to create a backup file of our Bayes data that will later be restored to our SQL database. This process may take a while:
su amavis -c 'sa-learn --sync --force-expire'
su amavis -c 'sa-learn --backup > backup.txt'

There are two files we need to grab from the source code:
wget http://svn.apache.org/repos/asf/spamassassin/branches/3.1/tools/convert_awl_dbm_to_sql
wget http://spamassassin.apache.org/gtube/gtube.txt

Now we will configure SpamAssassin to use MySQL. Once we do this, don't restart amavisd-new until we are completely finished. If you have to restart amavisd-new before the process is complete, set local.cf back the way it was. Note that Debian places local.cf in the /etc/spamassassin directory while many other OSs place it in /etc/mail/spamassassin. Also note that we will comment out bayes_path and auto_whitelist_path:
cp /etc/spamassassin/local.cf /etc/spamassassin/local.cf-pre-sql
vi /etc/spamassassin/local.cf

Insert text below, and comment out items as noted:
bayes_store_module              Mail::SpamAssassin::BayesStore::SQL
bayes_sql_dsn                   DBI:mysql:maia:localhost
bayes_sql_username              amavis
bayes_sql_password              passwd

auto_whitelist_factory          Mail::SpamAssassin::SQLBasedAddrList
user_awl_dsn                    DBI:mysql:maia:localhost
user_awl_sql_username           amavis
user_awl_sql_password           passwd

#this next part will store all data in just one user's table, no matter who runs sa-learn
bayes_sql_override_username amavis

#auto_whitelist_path /var/lib/amavis/.spamassassin/auto-whitelist
#bayes_path  /var/lib/amavis/.spamassassin/bayes

It is required we initialize the database by learning a message:
su amavis -c 'sa-learn --spam gtube.txt'

We are now going to load in the data from our old auto-whitelist file. Run these commands to show what arguments are required to run the special program convert_awl_dbm_to_sql. You should still be in the /var/lib/amavis/.spamassassin directory where we downloaded this script:
chmod +x convert_awl_dbm_to_sql
./convert_awl_dbm_to_sql

Now that you have read and understood what is needed, you may run the program. Edit if necessary; make sure you select, copy and paste the entire text:
./convert_awl_dbm_to_sql --username amavis --dsn DBI:mysql:maia:localhost --dbautowhitelist /var/lib/amavis/.spamassassin/auto-whitelist --sqlusername amavis --sqlpassword passwd --ok

Now log in to MySQL and issue these commands so we can see if it worked:
mysql -u root -p

USE maia;
SELECT email,count,totscore FROM awl;

Assuming you have been using auto-whitelist for at least a little while, you should see some data. Make a note of how many rows are in the set. Now we can quit:
quit

OK, now let's load in the Bayes data; this may take 5 minutes, or more than an hour and don't be surprised if your CPU load increases substantially during this process:
su amavis -c 'sa-learn --restore backup.txt'

On a healthy system that has been in use for a while there will be between 100,000 and 150,000 rows of data in the Bayes data tables. If you would like to view our progress, open another PuTTY session, log into MySQL, and monitor the progress:
mysql -u root -p

USE maia;
SELECT COUNT(*) spam_count FROM bayes_token;

You can use the [up-arrow] to recall commands. While you wait, commands that provide interesting data:
SHOW DATABASES;
SHOW TABLES;
DESCRIBE bayes_token;
DESCRIBE bayes_vars;
SELECT username FROM bayes_vars;
(should show 'amavis')

If everything went well, we can quit:
quit

And check the database:
su amavis -c 'sa-learn --dump magic'


The data conversion is complete, now we can test amavisd-new:
amavisd-new stop
amavisd-new debug-sa

Send a test mail through, and observe. You should see:
debug: bayes: Using username: amavis
debug: bayes: Database connection established
debug: bayes: found bayes db version 3

and if use_auto_whitelist is enabled (rather, is not disabled):
debug: SQL Based AWL: Connected to DBI:mysql:maia:localhost
...
debug: Post AWL score: 0.0520000000000005

If everything looks OK, stop debugging with [Ctrl]+c and start amavisd-new:
amavisd-new start

Now we will create a script to keep our auto-whitelist data trim. If you are currently using my /etc/cron.weekly/trim_whitelist_weekly script, it will need to be replaced. First let's create a new script:
vi /usr/sbin/trim-awl

and insert (always make sure you hit [Enter] at the end of the last line):
#!/bin/sh
/usr/bin/mysql -uamavis -ppasswd < /etc/trim-awl.sql
exit 0

 Save and exit the file, then make the script executable:
chmod +x /usr/sbin/trim-awl

Now create a file that the script will use for input:
vi /etc/trim-awl.sql

and insert:
USE maia;
DELETE FROM awl WHERE count = 1;

Save and exit, then create a cron job to run the script:
cd /etc/cron.weekly
rm trim_whitelist_weekly
vi trim-sql-awl-weekly

and insert:
#!/bin/sh
#
#  Weekly maintenance of auto-whitelist for Debian with amavisd-new and 
#  SpamAssassin using MySQL
test -e /usr/sbin/trim-awl && test -e /usr/sbin/amavisd-new && {
        su - amavis /usr/sbin/trim-awl >/dev/null
}
exit 0
Save and exit, then make it executable:
chmod +x trim-sql-awl-weekly

Then, to test it, run it:
./trim-sql-awl-weekly

Assuming you have been using auto_whitelist at least for a few days, log back in to MySQL, and note how many rows of data are in the awl table, compare it to the number I asked you to make a note of earlier. There should be less if the script is working (unless you previously did not have any entries with a 'count' of '1'):
mysql -u root -p

USE maia;
SELECT email,count,totscore FROM awl;

quit

 Do you have any free memory?
free

Is this maybe an opportunity to reboot?


Optional: If you would like finer control of auto-whitelist maintenance:
Log in to mysql and add a new field "lastupdate" to the awl database, and then populate it (initialize it) with today's date:
mysql -u root -p

USE maia;
ALTER TABLE awl ADD lastupdate timestamp default CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP;
UPDATE awl SET lastupdate = NOW( ) WHERE lastupdate < 1;

 Now, edit the trim-awl.sql file:
vi /etc/trim-awl.sql

and replace the simple-minded:
DELETE FROM awl WHERE count = 1;
 with more sophisticated statements:
DELETE FROM awl WHERE lastupdate <= DATE_SUB(SYSDATE(), INTERVAL 6 MONTH);
DELETE FROM awl WHERE count = 1 AND lastupdate <= DATE_SUB(SYSDATE(), INTERVAL 15 DAY);

Feel free to modify the time intervals as you see fit. Each time a record is created or modified, the lastupdate field will reflect the current date and time. These statements not only get rid of "single hit" records 15 days or older, but also records with 6 months or more of inactivity.


Last minute notes: If you upgrade to SpamAssassin 3.1 (does not apply to earlier versions), and use MySQL >= 4.1, change:
Mail::SpamAssassin::BayesStore::SQL
to:
Mail::SpamAssassin::BayesStore::MySQL

If you would like to make a backup of the Bayes data:
cd /var/lib/amavis/.spamassassin
su amavis -c 'sa-learn --sync'
su amavis -c 'sa-learn --backup' > bayes_backup.txt
chown amavis:amavis bayes_backup.txt

If you would like to see how many messages have been learned and how many tokens are in the database:
su amavis -c 'sa-learn --dump magic'


Suggested reading:
http://www.mysqlperformanceblog.com/2006/09/29/what-to-tune-in-mysql-server-after-installation/
http://www.mysql.com/news-and-events/newsletter/2003-11/a0000000269.html
Out of the box MySQL is poorly tuned for use with InnoDB tables. I suggest at least configuring innodb_buffer_pool_size and innodb_log_file_size. For example, increasing innodb_buffer_pool_size (from the default of 8M) to a conservative 128M can make a hugh difference in performance.

In order to change innodb_log_file_size you will need to shut down the server, rename the old log files and then let MySQL create new ones when it starts back up: http://dev.mysql.com/doc/refman/5.0/en/adding-and-removing.html

References:
http://wiki.apache.org/spamassassin/BetterDocumentation/SqlReadmeBayes?highlight=%28mysql%29
http://people.apache.org/~parker/presentations/
http://www.paulstimesink.com/index.php?op=ViewArticle&articleId=163&blogId=2
http://www.paulstimesink.com/index.php?op=ViewArticle&articleId=167&blogId=2
http://dev.mysql.com/doc/mysql/en/index.html
http://infocenter.guardiandigital.com/archive/amavis/2004/Dec/0319.html
http://svn.apache.org/repos/asf/spamassassin/branches/3.1/sql/README.bayes
http://svn.apache.org/repos/asf/spamassassin/branches/3.1/sql/README
https://www.maiamailguard.com/maia/wiki/SpamAssassin3SQLBayes

mr88talent at yahoo dot com 10/16/2006