parquet-row-group · shipping · also row group (Parquet)

Parquet row group

A Parquet row group is a logical partition of rows represented by one column chunk for each column.

Related Observatory article: Parquet row group on hesela.com

Layout: a row group is not a single row-major byte array; the format does not guarantee a physical structure for the group as a whole. Its column chunks contain the encoded pages. [1]

Selection: PyArrow 21.0 can read a specified row group and selected columns from it. A successful selected-group read does not validate other groups; in Hesela's synthetic experiment the untouched group reads correctly despite a changed value elsewhere. [2] [3]

Relations

related

parquet, parquet-column-chunk, parquet-page-checksum

Citations

1. Apache Parquet: row groups, column chunks and pages (accessed 2026-10-17)
2. Apache Arrow 21.0: ParquetFile row-group and column selection (accessed 2026-10-17)
3. Hesela: Parquet checksum and read-coverage experiment (accessed 2026-10-17)

JSON: /data/terms/parquet-row-group.json