which is better for virtual domains

["Николай Г. Петров" <ni...@cplus.ru>]

Hey guys!
If there is a lot of virtual domains with many virtual users in it, 
which is the better variant of configuration spamassassin about spam/ham:
   - individual database for each of users
   - or the same database for all of supported domains

And If I can I have one more question: is there any opportunity about 
virtual users to manage their white_lists? (alternatively, via the web 
interface)?

Re: which is better for virtual domains

[Jonas Eckerman <jo...@truls.org>]

Please keep discussions on-list.

On 2012-04-23 20:44, "Николай Г. Петров" wrote:

> My mail system:
> OS: FreeBSD
> МТА: sendmail
> MDA: maildrop
> database: ldap (openldap)
> pop/imap: courier-imap

I still have no idea how you call spamassassin or spamc or if you use 
some other method to connect to spamd.

>> -l -c -i 127.0.0.1 -m 3 --max-conn-per-child=5 --round-robin -u vmail
>> -x --virtual-config-dir='/corpmail/%d/.spamassassin/' -d -r ${pidfile}
>> -s /var/log/spamd.log

> , but in log I have a:

>> spamd[7256]: spamd: using default config for root:
>> /corpmail//.spamassassin//user_prefs
> Why 'root'?

Maybe because you haven't succesfully told spamd what user mail address 
to scan the mail for, so it falls back to the default.

> Why domains is not apear?

Maybe because spamd don't know the domain.

> My question is: privious, you say that you save a awl in mysql - what is
> it 'awl' - auto-white-lists?

Yes, AWL is short for Auto White-List. (Wich is a bad name for it.)

> And may I save in ldap?

I don't know. I've never used SA with LDAP.

> I read manual about
> ldap database: I don't understand atribute:
>> spamassassin: add_header all Foo LDAP read
> What they mean from this example?
> It's a 'awl' or 'user_prefs' or 'somthing else'?

I have no idea where in what man-page you found that, so I have no 
context at all.

> If I right understand I try to reach level on my mail system like this
> (please, if something is wrong, critic) ):

> /corpmail/domain1/.spamassassin/bayes_seen
> /corpmail/domain1/.spamassassin/bayes_toks
> /corpmail/domain1/.spamassassin/auto-whiltelist
> /corpmail/domain1/user/Maildir/user_prefs - (optionaly)

AFAICT you need to skip the optional one since you cant't keep multiple 
user-dirs for one user, and in your scheme the domain is used insetad of 
the user.

> I can automate train spamassassin from individual MDA filter per each of
> users, but normal mesage possible goto 'ham'.

I don't know what "normal mesage possible goto 'ham'" means here.

> Re-learn I think to configure with forward message to
> [spam|nospam]@doman[1|2].ru for each of domains, and by cron put some
> script which re-learn from folder spam|nospam on domains.

> How do you think it will work? Or may be some better idea?

I've done something similar myself. How good it works depends a lot on 
your users.

/Jonas
-- 
Jonas Eckerman
http://www.truls.org/

Re: which is better for virtual domains

["Николай Г. Петров" <ni...@cplus.ru>]

23.04.2012 15:40, Jonas Eckerman написал:
> On 2012-04-23 12:23, "Николай Г. Петров" wrote:
>
>> If there is a lot of virtual domains with many virtual users in it,
>> which is the better variant of configuration spamassassin about 
>> spam/ham:
>>    - individual database for each of users
>>    - or the same database for all of supported domains
>
> You forgot one option:
>  - individual database for each domain
>
> The answer depends on the situation (I assume you're asking about the 
> bayes database(s)).
>
> If the users have separate bayes databases, will they actually train 
> them? If the users doesn't train their databases, a common database 
> could work a lot better than individual databases.
>
> How much does the mail streams for the domains have in common? If they 
> have a lot in common it makes sense to have a common bayes database 
> for them. Otherwise separate databases for each domain migh be better.
>
> Regards
> /Jonas
Thank you for your detailed response!
--

- yes, I mean bayes (possible user's 'whitelists from' in user_prefs)
- About individual database for each domains, I forgot
       - If we mean a separate databases for each of domains, 
configuration like a?:

            -x - virtual-config-dir = '/corpmail/%d/.spamassassin/'

- count of users are not so many:
   domains: ~20
   users per domain ~40

- Is there actual possibility to users for train their bayes database if 
they not realy access on the server? May be you know some client 
interface(like web)?

Re: which is better for virtual domains

[Jonas Eckerman <jo...@truls.org>]

On 2012-04-23 12:23, "Николай Г. Петров" wrote:

> If there is a lot of virtual domains with many virtual users in it,
> which is the better variant of configuration spamassassin about spam/ham:
>    - individual database for each of users
>    - or the same database for all of supported domains

You forgot one option:
  - individual database for each domain

The answer depends on the situation (I assume you're asking about the 
bayes database(s)).

If the users have separate bayes databases, will they actually train 
them? If the users doesn't train their databases, a common database 
could work a lot better than individual databases.

How much does the mail streams for the domains have in common? If they 
have a lot in common it makes sense to have a common bayes database for 
them. Otherwise separate databases for each domain migh be better.

Regards
/Jonas
-- 
Jonas Eckerman
http://www.truls.org/