Parquet 129 is the columnar file format of the data lake. Twitter and Cloudera built it in 2013 on the record-shredding ideas of Google's Dremel paper, and it became a top-level Apache project in 2015. Spark 129 , Trino 403,499 , DuckDB 61,228 , Snowflake, BigQuery 1 and pandas 16,086 all read it, and Iceberg 129 and Delta Lake 237,929 tables (Lakehouses, Data Quality and Governance) are directories of Parquet files plus a log. The format specification is at 2.14.0 (11 September 2026); the files here come from Arrow 129 's C++ writer through pyarrow 25.0.1 129 and are checked with DuckDB 1.5.5. Row vs Columnar Layout showed why columns win; this section opens the file to show how.