On Wed, 15 Sep 2010 19:42:44 -0400 Dennis German <dger...@real-world-systems.com> wrote:
> On Sep 15, 2010, at 1:42 PM, RW wrote: > > > On Wed, 15 Sep 2010 11:18:20 -0400 > > Dennis German <dger...@real-world-systems.com> wrote: > >> I believe that bayes_seen is a perl hash and will not be reduced > >> in size by deleting entries. The only way to reduce it's size is > >> to have a program read the current file, entry by entry and output > >> to a new file. This will not copy deleted entries and the output > >> will be significantly smaller. ... > >> Dennis German > >> > > It's straightforward to do it with backup and restore, but the > > problem is that that there is no time field. You might just as well > > delete the file periodically. > > Thanks for the info however after running backup & restore: > Before: > 41,619,456 Sep 15 19:04 bayes_seen > 2,543,616 Sep 15 19:04 bayes_toks > After: > 43,511,808 Sep 15 19:26 bayes_seen > 2,560,000 Sep 15 19:26 bayes_toks autodelete doesn't remove signatures at all, so there's no point in compacting. I misunderstood what you saying and thought you were talking about removing entries. My point was that you can do that with sa-learn, but there is no basis for selecting which entries to delete. If you want to do it, it presumably could be done by switching to SQL and adding a date field. The trouble is that signatures shouldn't really be removed on a multi-user system unless there is a centralized method for leaning - otherwise there may be people who are relying on them to be kept indefinitely e.g. they may occasionally run sa-learn --ham on a read-mail folder.
