Parquet is a columnar file format designed for efficient data storage and retrieval. On disk, it is organized around row groups, column chunks, and pages. Along with that, each file also has a footer that describes how everything fits together. A Parquet reader that understands this layout can avoid a lot of work during the scan, such as skipping entire row groups, column chunks, and pages, and decoding only the values that matter.

This article uses a single sample Parquet file to explain exactly what happens during column reads and some common optimization techniques.