On Wed, Oct 14, 2009 at 6:06 PM, Dan Kennedy <danielk1...@gmail.com> wrote: > >> The problem I think may be people using TerminateThread(); that's how >> you hard kill a thread. It seems that can interrupt an I/O >> operation -- >> ie an operation that writes more than one cluster at a time. Meaning, >> synch = full may have nothing to do with it. If you have to say write >> more than one cluster (disk block), TerminateThread looks like it can >> abort the IO op in the middle of a multi-block op? I'm trying to run >> that down but can't yet find anything that verifies this. > > Even if it does, which seems quite plausible, the only way > I can see this causing corruption is if you are in persistent > journal mode and a (weak) checksum gives you a false positive > on the last, corrupted, record in the journal file.
This is quite possibly happening in our case. Any way to prove the theory? What should we look for? The problem is that journal_mode=persist was the only usable journaling mode on Windows due to the way file deleting is handled (in SQLite). Using journal_mode=delete is problematic on any machine with TortoiseSVN/TortoiseGIT or other programs installed. Now that journal_mode=truncate exists we can try switching to that. Best regards, Filip Navara _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
