Re: Separate connection handling from backends

On 7 December 2016 at 22:27, Kevin Grittner <kgrittn(at)gmail(dot)com> wrote:

> I don't know how that execution model would compare to what we use
> now in terms of performance, but its popularity makes it hard to
> ignore as something to consider.

Those engines also tend to be threaded. They can stash state in memory
and hand it around between executors in ways we cannot really do.

I'd love to see a full separation of executor from session in
postgres, but I can't see how it could be at all practical. The use of
globals for state and the assumption that session == backend is baked
in way too deep.

At least, I think it'd be a slow and difficult thing to change, and
would need many steps. Something like what was proposed upthread would
possibly make sense as a first step.

But again, I don't see anyone who's likely to actually do it.

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services