Overview
The Warden Bayes filter requires a minimum of 200 trained spam messages before it becomes active. While training typically occurs automatically through real-time learning, auto-learning, or the Plesk nightly script, administrators can initiate manual training to accelerate system initialization.
Manual Training from Message Directories
Use the /usr/bin/sa-learn utility to process existing message directories. Execute the appropriate command based on the intended classification:
- Train as Spam: Process a directory containing unwanted messages.
sa-learn --spam --username=amavis --use-ignores --local --progress /var/qmail/mailnames/example.com/johndoe/Maildir/.Spam/cur/- Train as Ham: Process a directory containing legitimate messages.
sa-learn --ham --username=amavis --use-ignores --local --progress /var/qmail/mailnames/example.com/johndoe/Maildir/curViewing Learning Statistics
Monitor the current state of the Bayes database via the command line or through the Warden dashboard (Anti-spam Learning Statistics widget).
sa-learn --dump magicMigrating Bayes Data Between Servers
Administrators can export and import Bayes training data to synchronize filters across environments. Ensure Warden is fully installed on the destination server before proceeding.
- Export Data: On the source server, generate a backup of the Bayes database.
sa-learn -u amavis --backup > /tmp/amavis-bayes.db- Transfer File: Securely copy the /tmp/amavis-bayes.db file to the destination server.
- Import Data: On the destination server, restore the database into Warden.
sa-learn -u amavis --restore /tmp/amavis-bayes.dbCommon Issues & Notes
- Import Duration: Restoring large Bayes databases is resource-intensive. Importing a dump containing approximately 2 million tokens typically requires two hours or more.
- Activation Threshold: The filter will remain inactive until the required message/token thresholds are met. Use the statistics command above to verify progress before expecting classification changes.
