repeately database corruption after some time

[Matthias Andree]

David Relson wrote:
> On Wed, 22 Mar 2006 04:15:58 +0100
> Christian Parpart wrote:
> 
>> On Wednesday 22 March 2006 03:57, David Relson wrote:

>> thanks for the fast reply. The command you gave me results into:
>>
>> mailbox_size_limit = 51200000
>>
>> As said, I'm getting an incredable amount of spam each day, and bogofilter 
>> frees me of 97% of all of them.
>>
>> What I do not understand immediately is that DB_PAGE_NOTFOUND relation to my 
>> problem. I mean, I was grepping the FAQ for my problem but didn't really know 
>> that it was directly related to postfix itself.

> DB_PAGE_NOTFOUND tends to happen when the database maximum size has
> been reached, BerkeleyDB tries to go beyond the maximum, and does the
> wrong thing.
> 
> Bogofilter has had the "cowardly refusing message message for a while"
> and that's supposed to provide protection against database corruption.
> Likely Matthias will have more to say when he comes online in the
> morning (CEST).

CET -- CEST is from last Sunday in March until last Sunday in October.

DB_PAGE_NOTFOUND happened when page splits failed, before the "cowardly
refusing..." behavior was added to bogofilter. I'm not exactly sure if the
headroom we leave below the resource limit is sufficient, at least
DB_PAGE_NOTFOUND have not been reported in a while.

> It sounds like you're using autoupdate mode, i.e. the "-u" switch.  The
> growth rate can be slowed down by using the "update_thresh" option
> (either in the config file or on the command line).  I use
> "thresh_update=0.01" so that obvious spam (0.99 or above) and obvious
> ham (0.01 or below) don't cause database updates.  You could try that
> value or a larger one.

...or by disabling "-u" mode altogether and manually training misclassified
or unsure messages.

-- 
Matthias Andree