It's less of a complaint rather than just a warning not to do what I did.
Version:
PostgreSQL 11.6 on x86_64-pc-linux-gnu, compiled by gcc (GCC) 4.9.3, 64-bit
The actual command list would probably be impractical to put in here just because the majority of it would just be creating a large amount of partition tables. But in summary what i've done is basically this:
Existing database has a partitioned fact table
1. Create an exact clone of that partitioned fact table which includes all the same indexes, columns, and partitioned tables
2. Change the partitioned table's partition key from an indexed date column to an indexed timestamp without timezone column
3. Do an insert into from the old partitioned fact table to the new partitioned fact table which includes all the same rows (insert into since i wanted the timestamp without timezone column to be in a new timezone)
4. Switch the names of the tables so the new one will be the one that's used
5. VACUUM FULL; ANALYZE;
For my use case which is a data warehouse star schema, this fact table is basically the base table of every report. To be more specific, the reports I've tested on varied from 2x slower to 4x slower. From what I see so far that's because the query plan is drastically different for both. An example of this test would look like this:
https://explain.depesz.com/s/6rP8 and
https://explain.depesz.com/s/cLUYThese tests are running the exact same query on two different tables with the exception that they use their respective partition keys.
