Daniel Shahaf <d...@daniel.shahaf.name> writes: > I see you have now implemented this (r933938; diff appended). I assume > this solution is reasonably fast (compared to the 0.25s and 3s figures > you cited upthread)?
Yes. There is very little additional overhead taking a lock unless blocked by a writer. The problem is that the lock blocks writers for the duration of the copy, which could be some time if the rep-cache is hundreds of megabytes. The SQLite backup solution breaks the write into chunks, which allows writers to progress, but does make the copy a bit slower (but still much faster than the SELECT/INSERT). -- Philip
