On Mon, 2006-08-07 at 16:53 -0500, Ron Johnson wrote: > > It will be difficult to have a consistent dump though. You can't do > > that with separate transactions. (And you can't have multiple > > simultaneous readers without separate transactions.) > > Absolutely, you're right. All "threads" must run from within the > same read-only transaction. > The idea is that pg_dump already works and already creates a good backup. Why not split up the data after pg_dump produces it? Of course it should be split up in a stream fashion, like I suggested before in this thread with my multiplexing script. Regards, Jeff Davis
