On 5 April 2016 at 14:18, Peter Geoghegan <pg@heroku.com> wrote:
I rather agree that an in-core system that solved some of the basic problems would be a huge step forward, and would motivate people to work on the harder problems. It's surprising to me that we don't seem to be much closer to that then we were when 9.4 was released.
I think we are, actually. For one thing a lot of groundwork has gone in, even if it doesn't seem especially highly visible.
Also, pglogical_output might've been able to get into 9.6 if we'd got it submittable a bit earlier, it'd had useful review in the first few weeks (months?), and I'd had more time to more responsively react to review comments by the time it did get review.
Even there, there are a bunch of places (like the relation cache/invalidations) that should really be moved "up" into logical decoding its self. Which while architecturally superior means yet more groundwork patches to write and get committed before a new revision of pglogical_output can go into 9.7.
I thought that getting pglogical_output into 9.6 might help us move closer to that goal, but folks seemed to think it was useless without the downstream client. I disagree, both on the principle of iterative development and because it's plenty useful for other things by its self, but didn't make any progress.
I don't think it helps that there's a real community split between people who want to see the actual logical rep engine live out-of-tree as an extension and those who think it's not interesting until it's in core.