After retraining and setting spam assassin for wide site all looks good. Spam gets bayes99 and non spam is bayes00. So far i did not get any spam. Thank you all for your help.
>> >> >> Am 31.10.2015 um 16:06 schrieb j...@lexoncom.com: >>> So after initial learning it looks better now. (BAYES_50) >> >> BAYES_50 is not really good for clear spam >> > yep i though that bayes was used but it seems like it was all useless > >>> When sendmail sends email to procmail and procmail passes it to spam >>> assassin, does spam assassin runs as root user or as the user the email >>> is destined to? >> >> depends on how SA is called in detail, normally it should switch to that >> unix-user and hence training as root makes no sense, *nothing* should >> proceed potentially dangerous input as root at all - inbound mailcontent >> is by definition that sort of "don#t do that" input >> >>> I run the sa-learn as root user >> >> oh my god....... > i run it through the crontab > yes i can create new user and force sa-learn to use that user >> >>> and it seems like this is the data based >>> that is being used so it would be global data base used for all mail >>> users? >> >> https://wiki.apache.org/spamassassin/SiteWideBayesSetup > > i switched to global setup > now all users should use same db > and i will use the manual learning process >> >>> X-Spam-Flag: YES >>> X-Spam-Level: ************ >>> X-Spam-Status: Yes, score=12.9 required=5.0 >>> tests=BAYES_50,FROM_12LTRDOM, >>> >>> HTML_MESSAGE,MIME_HTML_ONLY,RCVD_IN_BRBL_LASTEXT,RCVD_IN_MSPIKE_BL, >>> >>> RCVD_IN_MSPIKE_L5,RCVD_IN_XBL,RDNS_NONE,URIBL_BLACK,URIBL_DBL_SPAM, >>> URIBL_JP_SURBL,URIBL_WS_SURBL autolearn=disabled version=3.4.0 >> >> well, the quota of your sa-headers was enough to reject my repsonse on >> the submission spamass-milter >> >> result: Y 16 - URIBL_BLACK,URIBL_DBL_SPAM,URIBL_JP_SURBL,URIBL_WS_SURBL >> >> > no sure what this means? > >
