Andrew Dunstan wrote:
> I'm not convinced it would be a huge gain anyway. Switching madly in
> and out of the perl interpreter at least is a known performance
> problem, IIRC
Returning control to the backend for every row returned would likely be
excessive, but you could return once every k rows and get most of the
benefits of both approaches (k might be on the order of 1000). The
problem with the current approach is that it makes returning large
result sets from PL functions very expensive, since they need to be
spooled to disk.
As for using threads, that's pretty much a non-starter: we can't safely
allow calls into the backend from multiple concurrent threads, and I
doubt that will chance any time soon.
-Neil