I wrote a page about this: https://duckdb.org/docs/guides/performance/file_formats https://duckdb.org/docs/guides/performance/file_formats TL;DR: if your workl

I wrote a page about this: https://duckdb.org/docs/guides/performance/file_formats https://duckdb.org/docs/guides/performance/file_formats TL;DR: if your workload is aggregation-heavy, Parquet is pretty good. If you have many joins, the lack of fine-grained statistics will likely result in bad join plans. (Disclaimer: I work at DuckDB Labs)