On 27/03/2019 20:27, Marc Pujol wrote:
A 27/03/2019 15:39, Nick Howitt escrigué:
I am not sure how to read this but the TIMING and TIMING-SA figures
are huge with tests_pri_-90: 299180, b_finish: 297503 and
expire_bayes: 297448 jumping out to me. I am not aware of using bayes
filtering.
You most probably are using bayes filtering. This is controlled by the
"use_bayes" directive in your spamassassin (not amavis) configuration.
Furthermore, you have auto_expire enabled (think of expiration as a
"periodic cleanup" that spamassassin must do to the bayes database).
This is known to cause large delays.
You can disable auto-expiration by adding "bayes_auto_expire 0" to
your spamassassin configuration.
Then perform an expiration periodically in a cron file (make sure it
runs under the same user amavis is running). Expirations are triggered
through sa-learn (again, make sure you run that under the same user
amavis is running):
sa-learn --force-expire
What is weird is I've been down this route. "grep bayes /etc/* -r"
returned one commented line in /etc/mail/spamassassin/v320.pre:
# and create a header containing ASN data for bayes tokenization.
and that is it. I also tried an "sa-learn --sync" and "sa-learn
--force-expire" and got no improvements. I am really puzzled.
