On Sat, Nov 30, 2013 at 5:15 PM, L. Wood <lwoo...@live.com> wrote: > Thanks, that makes sense. > > But I can pretty much rule this case out. In my case, it's highly unlikely > that the folder would be moved during the brief instant between _open() and > the actual opening of the "-journal" file. > > Are there other possible cases of corruption? >
(1) Process A opens database at /first/path/to/database.db (2) The directory is renamed to /second/path/to/database.db (3) Process B begins writing to the database, creating a rollback journal at /second/path/to/database.db-journal (4) Process B crashes without committing or rolling back. (5) Process A tries to read the database. It checks for a hot journal file at /first/path/to/database.db-journal, finds none (since the hot journal is now in a different place) and therefore does not know that it needs to recover the database file. It sees the database as being corrupt. Or if A tries to write to the database, it makes the corruption worse, such that even a third process that does see the hot journal is no longer able to recover. -- D. Richard Hipp d...@sqlite.org _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
