bogoupgrade

[Matthias Andree]

Rodrigo Bernardo Pimentel <rbp at isnomore.net> writes:

>> Well, you could have some success with db_dump -r and db_load, but you'd
>> better check with bogoutil -d after that to see how much data is
>> missing. db_stat -d on the old and repaired .db file may give you an
>> idea of how much has been gone - provided that db_stat terminates.
>
>         BTW, why are bogofilter databases so often corrupted? For a time,
> everytime I upgraded bogofilter (which was *not* on every bogofilter
> release) I trained it again and stored the previous db files. When I saw a
> previous post about corrupt databases I went to check mine and *all* of them
> (4, actually) were corrupted. Ok, so I upgraded to 0.14.3-1 (Debian) and
> retrained it. That was about a week ago (9 days, I think). And yesterday I
> noticed wordlist.db is corrupt again!

bogofilter versions before 0.10 and 0.14.0 ... 0.14.2 had locking bugs
that could corrupt the data base when concurrent access occurs. No such
bugs are known for 0.14.3 and newer and for 0.10 to 0.13.X. However, if
your data base is corrupt, it doesn't fix itself, but must be repaired.

>         Now, this shouldn't be happening so often, right? At the very least,
> we should be getting some sort of warning from bogofilter ("Warning: corrupt
> database, please retrain" or something).

bogofilter isn't getting complaiants from DB, else it would be reporting
it.

>         And what causes corruption? A message bogofilter doesn't
>         understand?

Possible reasons are running a version with broken locking, although
0.14.3 should be safe, and I/O errors or "disk full" conditions during
write access. Whether an application crash causes corruption, I cannot
say for sure.

-- 
Matthias Andree