Spam training question

[Jbrabander]

I was wondering, for whatever process goes through the spam email account each night, how many emails does it use each night for training?

For example, my spam training account gets a pretty fair number of mails put into it daily by users. But if I look in the account right now, it's still got some mails in it dated Dec. 1st and 2nd. There are currently about 300 messages in the spam training inbox. Half from today, maybe 75 from the 1st, and 75 from the 2nd.

Shouldn't it empty each night?

[Jesster]

hmm in your cos is this value: zimbraMailSpamLifetime

zmprov gc default zimbraMailSpamLifetime

perhaps this is what you're looking for?

[phoenix]

Originally Posted by Jbrabander

Shouldn't it empty each night?

No, it will retain messages for the 'Lifetime period' specified in the COS/Advanved page in the Admin UI or modified by the attribute Jesster has posted above.

[Jbrabander]

Hmmm, something's wrong then. If I'm looking at the right setting, the spam retention is set at 30 days, but when I look in the spam account inbox, it only has messages back to Dec. 1.

Under Email Retention Policy:

e-mail message lifetime = 0 days
trashed message lifetime = 30 days
spam message lifetime = 30 days

[phoenix]

Originally Posted by Jbrabander

Hmmm, something's wrong then. If I'm looking at the right setting, the spam retention is set at 30 days, but when I look in the spam account inbox, it only has messages back to Dec. 1.

OK, what do you mean by the 'spam account'? Do you mean the Junk folder that each user has or the real ham/spam accounts on the system?

[Jbrabander]

The ham/spam accounts on the system.

[phoenix]

They should be cleared out by the cron job that runs overnight for training, they are not affected by the lifetime attribute. You can check if your accounts are set correctly with the following:

zmprov gacf | grep SpamAccount"]zmprov gacf | grep SpamAccount

[Jbrabander]

IsNotSpam is set to the ham account, IsSpam is set to the spam account.

[Rich Graves]

There's a zmtrainsa.log (or thereabouts) that, if I recall correctly, records the number of messages learned.

As you've seen, the nightly purge job doesn't empty the spam inbox, just messages more than N days old. Feeding them to zmtrainsa again doesn't do any harm because the bayes database itself records message-ids previously seen.

IMO it would not be a bad idea to run zmtrainsa more frequently, without the --sync argument if you're concerned about load.

[mek1]

Since I was hot on the trail of the aforementioned log, is it /opt/zimbra/log/spamtrain.log ?

I have a semi-related question. If I'm training by cmd say,

/opt/zimbra/bin/zmtrainsa spam.3tgic3pi5@company.com spam inbox

is that supposed to update the spamtrain.log log or is that the cron's output?

thanks