locking [was: database rebuild]

[David Relson]

At 03:41 AM 2/8/03, Nick Simicich wrote:

>Thanks for the db_verify suggestion, that got me pointed down the "road to 
>recovery".

When BerkeleyDB was chosen for bogofilter, the goal was to have the perfect 
database.  Unfortunately the way we chose to use it wasn't perfect.  The 
result has been database corruption for some people.  Hopefully we have a 
working locking protocol to ensure database integrity.

Significant work has gone into the locking protocol.  The regression tests 
run by "make check" include two tests for locking.  One of those works hard 
to reveal any flaws in the locking protocol now used.  The tests have been 
passing on all our tested platforms with the only known issue being some 
flavors of NFS.