On Mon, Jan 26, 2009 at 11:02 AM, Mark Martinec <mark.martinec+ama...@ijs.si> wrote: >> It claims to have found an entry for u...@example.com "w/o IP address" >> (ip="none", I assume) > > Yes, that would normally be an: u...@example.com|ip=none > At that point the AutoWhitelist.pm should also be > removing such database entry and replacing it with > a new one containing the ip part.
It seems like this isn't always happening, or that two messages are being processed simultaneously for the same user and they're stomping on each other's result. >> even though searching the database confirms that >> an entry like that does not exist (I looked at all entries for >> u...@example.com). > > There may be some other reason why AutoWhitelist thinks a > record "u...@example.com|ip=none" exists. Try enabling > DBI module tracing (SQL database access), see: man DBI. > For example (assuming Bourne shell or alike): > > su vscan -c 'DBI_TRACE=2 spamassassin -t -D <0.lis' >0.log 2>&1 Thanks for that... that helped quite a bit. >> SA claims the weight of this nonexistent entry is -187.24 which explains >> why the AWL is subtracting a large amount. > >> It then goes and adds the >> new /16 to the db with a high score (38.844) as it should. > > At that point the old record (assuming it existed) should > be deleted, so with a next attempt with a same message the > matching on "w/o IP address" should not occur. > Does it or does it not? Before I ran my test run, I found an ip=none entry sitting there. (It wasn't there 15 minutes ago.) I quickly ran SA from the commandline. The test run did indeed delete the ip=none entry, but it also used it to score the message. AWL subtracted 39 points, and I made sure that this was a new netblock (I changed it in the headers). I am running multiple servers against one database, each with many amavisd processes. It is not unexpected that there may be multiple messages inbound to the same recipient at the same time -- so perhaps there is a bit of a race condition here? How can the ip=none entry be made per-process or at least per-machine specific? >> The AWL MyISAM table checks OK, is cleaned nightly and is >> analyzed/optimized. (If this has gone into SA-land and you consider it no >> longer relevant to amavisd, please let me know and I'll try there...) > > The SA 'users' ML would be more appropriate if you don't find the > solution soon. Understood, and thanks for your continued help. ------------------------------------------------------------------------------ This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword _______________________________________________ AMaViS-user mailing list AMaViS-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/amavis-user AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3 AMaViS-HowTos:http://www.amavis.org/howto/
