On Wed, May 12, 2021 at 04:48:29PM +0900, yuzuko wrote: > Hello, > > > Thank you all for the feedback and insights. > > > > Yes, the intention is to *replace* expand_inherited_rtentry() in the same > > way planner_hook replaces standard_planner(). > > > > This patch is really useful. We are working on developing hypothetical > partitioning as a feature of HypoPG[1][2], but we hit the same problem > as TimescaleDB. Therefore we would also be thrilled to have that hook. > > Hypothetical partitioning allows users to define multiple partitioning > schemes on real tables and real data hypothetically, and shows resulting > queries' plan/cost with EXPLAIN using hypothetical partitioning schemes. > Users can quickly check how their queries would behave if some tables > were partitioned, and try different partitioning schemes. HypoPG does > table expansion again according to the defined hypothetical partitioning > schemes. For this purpose, we used get_relation_info hook, but in PG12, > table expansion was moved, so we cannot do that using > get_relation_info hook. This is exactly the same problem Erik has. > Therefore the proposed hook would allow us to support hypothetical > partitioning.
Sorry for missing that thread until now. And yes as Hosoya-san just mentioned, we faced the exact same problem when implementing hypothetical partitioning, and eventually had to stop as the changes in pg12 prevented it. So +1 for introducing such a hook, it would also be useful for that usecase.
