Bayesian filter across all cluster

[Ch@rlus]

Hey guys !

I'm facing an issue on a fairly recent Proxmox Mail Gateway installation.

We have a really simple installation with 3 servers, configured in cluster directly in Proxmox.

It looks like the bayesian filter are not sync accross all the nodes in the cluster. I'm receiving tons of the same SPAM, and they are successfully placed in quarantine (score 6) on my server no 1 (which is the master in the cluster at the moment), but they are not blocked (score 1) on my two other servers.

After looking a the logs, it seems that the only difference is "BAYES_99,BAYES_999" in the logs.. I'm assuming the bayesian filters/training is not sync across all the nodes.. If this a normal behaviour ? If there anything I could do to ensure that all node get a similar spam detection ?

Thanks !

 

[dietmar]

Yes, that is expected behavior. We do not sync the bayes database. Normally, you receive most spam mails more than once, so it is very likely that all cluster nodes learn and filters them. But yes, the databases are different.

 

[Ch@rlus]

Thanks for your reply

Is there any downside to sync theses databases ? It looks like my first MX server is getting much more email than my 2 others (although they all have the same priority), therefore it seems to have a better spam detection ratio.

Or maybe, is there anything I could do to "help" my two other server to complete their Bayes training ? Does the "blacklist" button on the quarantine send the mail to Bayes ? And what about spam that didn't even get into the quarantine ? Is there any way to send them to Bayes ?

 

[DerDanilo]

Out would be good if the db would be synced if technically possible.
How can we achieve this?

 

[blonjon]

I suppose it is possible to configure bayes to use a MySQL or Redis backend and use it on for all nodes. I am looking for a similar architecture. I'll let you know if I find anything relevant.