zmtrainsa - not sure how to implement

[ple]

Hi,
I'd like to implement zmtrainsa but get confused. These are my questions:
1. There are two domain in my server host.domain.com and domain.com; we only use domain.com as our default domain. Can I use the built-in spam/ham accounts that belong to host.domain.com or I have to create 2 brand new spam/ham accounts for the domain.com?
2. zmtrainsa <server> <user> <password> spam [foldername] is the syntax; who should I use in the <user> field?
3. This relates to the previous question; do I have to run this syntax manually for over 2K accounts on the server?
4. Once the spam is set up; can I just ignore the ham part?
Thanks in advance,
PL

[phoenix]

You don't need to run zmtrainsa, it's run automatically every night as a cron job to train your system. What, exactly, are you trying to achieve?

[ple]

Thanks for quick responding. I've noticed that all marked Junk messages are still coming back to our Inboxes. In other words, the Junk mail is not working for us. What settings do I need to enable/disable in order for the Junk mail to work?
PL

[ewilen]

What do you mean by "coming back to our Inboxes"? You mean you're still getting similar junkmail to what you've already marked?

[ple]

That is correct. When you mark the message as Junk; Zimbra moves the message to the Junk Email folder, which is good. Next day, you will see the similar message sitting in your Inbox. For some reason, Zimbra doesn't learn that the message has been flagged as junk.
PL

[ewilen]

To really see what's going on, right-click on the "similar" emails and do View Original. Then compare the spam scoring in their mail headers. You should see progress in the sense of the Bayesian scoring getting higher over time.

However Bayesian scoring alone isn't enough to auto-recognize junk on a stock Zimbra system. The reason: on a stock system, Bayesian only scores up to 3.5, and the spam threshold is 6.6. So look again at those headers and see what the total score is. If at least 3.1 "points" come from scoring other than Bayesian, then Bayesian will eventually suffice. Otherwise you should consider tweaking the spam settings.

The easiest thing to do is just to lower the threshold via the Admin GUI (I believe it's under Global settings). For more advanced tweaks you should search the forum and the wiki.

One thing to watch out for, though, is Bug 31333 &ndash; Emails from authenticated users getting marked as spam
Bug 12277 &ndash; Add support for better antispam scoring of sasl authenticated senders

The upshot of these is that if you set the threshold too low, then mail from legitimate users may be marked as spam simply because they're connecting from external networks. See comment #10 in bug 12277 for the necessary customization.

[ple]

I'll keep watching the spam score. Let's get back to my original question. It means there is no need to run zmtrainsa <server> <user> <password> spam [foldername] on my server and the spam is working as is? please correct me if I'm wrong.
Thanks.
PL

[ewilen]

Originally Posted by ple

I'll keep watching the spam score. Let's get back to my original question. It means there is no need to run zmtrainsa <server> <user> <password> spam [foldername] on my server and the spam is working as is?

This is correct. Login as zimbra and do crontab -l. This will show you that zmtrainsa is executed at 11:00 pm every day, first to train and then to clean up the spam/ham accounts.

Also see /opt/zimbra/log/spamtrain.log

However I am seeing an anomoly which I'm going to post in another thread.