> Ernie Dunbar wrote: >>>>Oct 5 13:26:39 pop spamd[19660]: Cannot open bayes databases >>>>/home/spamc/.spamassassin/bayes_* R/O: tie failed: Permission denied >>>> Oct >>> >>>The typical cause of this is that the file ownership changed on the >>>bayes files when you converted them to 3.0. At least this was the >>> problem >>>I encountered. >>> >>>When I did all my conversions following the UPGRADE procedure as root, I >>>had to go back in and convert the files back to their correct >>> ownerships. >>>Check for that. >> >> >> That's kind of the odd thing. I set the ownership back to spamc, but it >> was changed to root for bayes_journal and bayes_toks shortly therafter. >> >> What I did instead of UPGRADEing, was nuke the whole database and >> rebuild >> it from scratch using some new spam. >> > > This can happen if you have any scripts or process that you use to train > the bayes database. Awhile ago I ran into this issue a few times when I > was logged in as root and manually trained a message (cat msg | sa-learn > --spam), which would occasionally change the ownership of the > journal/etc to root. Fixed that by not doing it as root, though now I > have bayes in SQL so it doesn't matter. >
About the only processes I can find that might be doing this are the spamc children. For some odd reason they run as root, despite that I use the '-u spamc' switch with spamd. The parent runs as spamc though, which is pretty confusing. Is this the way it's supposed to run? Or did I mess up the compile?
