[Gllug] Continued spamassassin strangeness

[Adrian McMenamin]

On Sat, 2004-12-25 at 01:03, Nix wrote:
> On Fri, 24 Dec 2004, Adrian McMenamin suggested tentatively:
> > It is running as 'nobody' - I wasn't aware that had an impact on
> > permissions, I thought nobody was just another shell-less user. The
> > files are now root.nobody but it is still not working :(
> 
> OK. Things we need to know before the cause is clear:
> 
> - are you running spamd/spamc, or spamassassin(1)?
> - who is the owner of the shared bayes files, and what
>   are the permissions on those files?
> - what does `file' report on those files? (This should
>   report Berkeley DB version and such things)
> - if you're running spamd and spamc, how are you invoking
>   each of them?
> - what user does spamd end up running as?
> 
> Generally, for non-database-config configurations, you want spamd to run
> as `root', so that it can read per-user configuration files and write
> per-user bayes databases; the best way to do shared Bayes databases
> these days is to put them in a MySQL database (or a PostgreSQL one in
> SpamAssassin 3.1). That way, you have no permissions problems.
> 

Was up to 3am fixing this :( But its done now.

Spamd was being invoked by spamass-milter.

Spamd was running as root (ie dropping down to 'nobody' when processing
mail). I really had two choices: let it use the "static" learning I
produced with sa-learn as root, or dynamically learn from my incoming
email on a per-user basis.

I settled for the per-user basis, which means that spamd is learning
again and not reporting any bayesian scores just now.

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug