Wouldn't it be nice if Postgresql could gain a native new storage engine which serializes columnar data as parquet files? Does this even make sense?
DuckDB already has fine Parquet support and I find it really helpful.
DuckDB already has fine Parquet support and I find it really helpful.
It's a matter of how much you want to keep using postgres versus if you want interactive-level performance.
I'm not saying one is better or worse than the other but they are definitely not the same.
Yes, but may not be practical. A non-trivial amount of code makes assumptions about the underlying data layout, so things like the cost-based optimizer and MVCC would basically need to be rewritten.
Surely that's exactly what parquet_fdw is?