[Mimedefang] Spamassassin 3.1 and improved bayes/sql.

[David F. Skoll]

Matthew Schumacher wrote:

> Yes, we tried that.  I attached another version of the proc that gets
> rid of the looping altogether, but believe it or not, it's slower.

Weird...

> The reason why we rely on procs is because the SA code doesn't have
> transactions yet.  By passing the tokens in as an array we get
> transactions because pgsql treats each run of a proc as a trans.

If your message has only a few tokens, the SELECT * ... WHERE token in
(...) runs in a single transaction.

[...]

> You mean having a separate bdb for each key/val pair?

Ouch!  No.  A separate bdb for each user.  And in a cluster of scanning
machines, the bdb's are updated on a master and then synchronized to
each slave.

> As much as I love pgsql and think it's light years beyond mysql in
> features, stability, flexibility, and even performance (when doing
> complex queries against large tables), I am now testing mysql for my
> bayes store.  Mysql has one good thing going for it, raw speed on simple
> queries, which is really the only requirement for bayes.

I've never used MySQL, so I can't comment.  But if you want real speed,
a well-written BDB implementation will beat either Pg or MySQL.

Regards,

David.