I wrote:
> There's no hard limit. Offhand the only thing I can think of that might
> be a bit slow is password lookup during connection --- I think that does
> a linear scan through a list of usernames. This is only an issue if you
> use passwords of course, but I suppose you probably would. (It would
> likely not be hard to improve the search algorithm, if it did become
> a bottleneck.)
BTW, I take that back --- it already is a binary search, so there
shouldn't be any problem with thousands of users. Still, I like
your idea of continuing to pool the connections better. Backend
startup is a bit expensive.
regards, tom lane