Greg Stein wrote: >On Thu, Jan 10, 2002 at 09:10:13PM -0800, Brian Pane wrote: > >>... >>Is there any reason why the ptrans pool needs to be a child of >>tpool? If we can use a null parent for ptrans, this problem should >>go away. >> > >null parents are ugly. isn't there another pool that has the right lifetime >for that pool? >
There isn't a pool with the right lifetime in the current worker design. The safe pool to use as the parent would be the worker thread's pool, but the listener thread doesn't have a way to use that pool for the accept. In Aaron's "time-space tradeoff" variant of the worker design, it's possible to use the worker thread's pool as the parent of the transaction pool, but the "time-space tradeoff" code hasn't fared as well as the current design in performance testing. --Brian
