On Mon, Mar 15, 2010 at 11:18 AM, Scott Hess <sh...@google.com> wrote:
> AFAICT, the operation to copy the pages back _is_ journaled, and the
> journal will get any pages which are overwritten in the front of the
> main database.  If the initial database has half of the pages used, it
> seems like the journal could range from almost nothing (if the used
> pages were mostly at the end) to almost half the database size (if the
> used pages were mostly at the front), so to be safe one would need 2x
> the final database size available.

As an aside, it occurs to me that one could implement an alternate
VACUUM which wrote the new database back to the main database file,
journalling the original versions of those pages.  The main database
would be locked for longer, but that might be a fair tradeoff for some

sqlite-users mailing list

Reply via email to