Paging Through a Parquet File in DuckDB: file_row_number or OFFSET?
Summary
This article benchmarks paging through a large Parquet file in DuckDB, comparing LIMIT/OFFSET with a row-range approach using file_row_number. It finds a 2.53x speedup for 20M rows on a 163-row-group file and explains how row groups and optimizer rewrites influence performance, offering practical guidance on when to use file_row_number and how to determine page boundaries.