APACHE-PARQUET

Columnar data lake storage with efficient compression.

Definition

Parquet structure: row group → column chunk → page. Encodings: dictionary, run-length, delta, byte-stream-split. Min/max statistics per column chunk for predicate pushdown. Nested schema via Dremel encoding (definition/repetition levels).

Origin

Created by Twitter and Cloudera in 2013 ; Apache top-level May 2015.

Example in context

A data lake stores 1 PB of analytics events as Parquet with ZSTD compression for 70% space saving vs JSON.

Related terms

• Arrow — in-memory twin.