[Mimedefang] sa-learn load, blocking?

[David F. Skoll]

Kevin A. McGrail wrote:

> I've had the same problem and Nice didn't help.  I think it's just grinding
> and doing so much disk i/o or something.

Probably because of hits to Berkeley DB.  I would guess that SA locks
the Bayes DB when it runs sa-learn (but I haven't actually checked the
code to be sure.)  That might cause the DB readers to block too.

In CanIt, we use our own Bayes implementation that uses a PostgreSQL
back-end.  We periodically train from the journal into the database
using the "ticker" facility of MIMEDefang.  It might be worth
experimenting with calling sa-learn from the ticker.  The nice thing
about using the ticker is that you're guaranteed never to have two
tickers in the same band executing concurrently, so you can run
sa-learn reasonably often without worrying about overlapping sa-learn
runs.  This should keep each sa-learn run reasonably small and low-impact.

(If you've ever wondered why I added the ticker facility, now you
know. :-))

Regards,

David.