Stefan Osterlitz wrote:
> Hello Users,
> i have a problem with the autolearning function on my server. Mails
> submitted to spamd get autolearn=failed, but mail sent to spamassassin
> or sa-learn will get learned correctly.
>
> I am running SA 3.2.3 with Bayes and AWL on the MySQL backend.
> The server is running as user vscan with start options -d -m 10 -x -q
> --socketpath=/var/run/spam -u vscan
>
>
> [EMAIL PROTECTED]:~/spam_ <mailto:[EMAIL PROTECTED]:%7E/spam>> sa-learn
> --clear
> [EMAIL PROTECTED]:~/spam_ <mailto:[EMAIL PROTECTED]:%7E/spam>> sa-learn --sync
> [EMAIL PROTECTED]:~/spam_ <mailto:[EMAIL PROTECTED]:%7E/spam>> sa-learn
> --dump magic
> 0.000 0 3 0 non-token data: bayes db version
> 0.000 0 0 0 non-token data: nspam
> 0.000 0 0 0 non-token data: nham
> 0.000 0 0 0 non-token data: ntokens
>
> [EMAIL PROTECTED]:~/spam_ <mailto:[EMAIL PROTECTED]:%7E/spam>> spamc -c
> --headers
> -U /var/run/spam < 1390002.msg.msg
> 8.8/4.0
> [EMAIL PROTECTED]:~/spam_ <mailto:[EMAIL PROTECTED]:%7E/spam>> sa-learn
> --dump magic
> 0.000 0 3 0 non-token data: bayes db version
> 0.000 0 0 0 non-token data: nspam
> 0.000 0 0 0 non-token data: nham
> Nothing has happened in the database using spamc/spamd..
That alone isn't a concern. Note that not every spam will be
autolearned, and with a score of 8.8, it's probably not going to be.
SA requires at least 3.0 points of header tests AND 3.0 points of body
tests, regardless of what the total score is.
It also recalculates the score as if bayes were disabled, so the total
score the learner uses is likely to be lower.
Also of note, that spamc call is not compatible with your local.cf below
and could result in mail corruption. To use the --headers parameter to
spamc you need to use "report_safe 0" in your local.cf.
from man spamc under --headers:
Note that this only makes sense if you are using "report_safe
0" in
the scanning configuration on the remote end; with
"report_safe 1",
it is likely to result in corrupt messages.
>
> ep 6 11:03:04 h9902 spamd[9924]: spamd: result: Y 10 -
> BAYES_95,DRUGS_DIET,DRUGS_ERECTILE,DRUGS_MUSCLE,FB_CIALIS_LEO3,FB_GVR,FROM_LOCAL_NOVOWEL,HS_INDEX_PARAM,HTML_MESSAGE
> scantime=1.0,size=4368,user=(unknown),uid=65,required_score=4.0,rhost=localhost,raddr=127.0.0.1,rport=/var/run/spam,mid=<[EMAIL
> PROTECTED]>,bayes=0.980617,autolearn=unavailable
> I do see nothing special there, apart from the autolearn=unavailable.
> There are no errors due to permissions, vscan homedir amd /tmp are
> writable, the database works and is writable by the user.
And the autolearn=unavailable isn't concerning either.. It usually means
the message has already been learned.
If there were permissions issues you'd most likely get autolearn=failed,
and if it was disabled you'd get "autolearn=disabled".
See also:
http://wiki.apache.org/spamassassin/AutolearningNotWorking
> This is my local.cf (score changes left out) - it does --lint!
>
> trusted_networks 81.169.178.43
> internal_networks 81.169.178.43
> required_score 4
> score ALL_TRUSTED 0
Why did you zero out ALL_TRUSTED? Was it hitting spam? If so, fix your
trusted_networks setting, don't zero out this one rule. You have by far
more problems than just this one rule..
See also:
http://wiki.apache.org/spamassassin/TrustPath
> Do you have any ideas what I might try? How does the spamd server work
> differently from a manual call of spamassassin? Is there any more
> debugging I can get from the daemon?
Are you sure SA is *never* autolearning?
ie, assuming your spamd logs to maillog
grep "autolearn=yes" /var/log/maillog
>
> Cheers,
> Stefan
>
