Sample Parquet File

Parquet is a columnar storage format: instead of writing records one after another the way CSV does, it writes each column’s values together. This sample stores all 30 names, then all 30 emails, and so on — grouped into a single row group, with each column chunk compressed independently using snappy.

That layout is why Parquet dominates analytics workloads:

• Column pruning. A query that needs only salary reads only the salary chunk, skipping the other six columns entirely.
• Predicate pushdown. Per-chunk statistics (min/max) in the footer let engines skip whole row groups that can’t match a filter.
• Embedded schema. The footer records every column’s name and physical type, so salary comes back as an int32 — no header-sniffing or type inference, unlike CSV.

The trade-off is that none of this is inspectable with cat. The file is binary, footer-indexed, and only meaningful to a Parquet reader — hence no raw-contents block on this page. Open it with the Parquet Viewer, convert it with Parquet to CSV, or load it in code: pd.read_parquet(...), duckdb.sql("SELECT * FROM 'sample-employees.parquet'"), or arrow::read_parquet() in R.

One honest caveat: at 30 rows, format overhead (footer metadata, dictionary pages) dominates and the file is larger per-row than its CSV twin. Parquet’s advantages compound at thousands to billions of rows; this sample is sized for testing correctness, not benchmarking.

All eleven formats in this section carry the same 30 employee records. The nearest relative is the Feather sample — also columnar and binary, but optimized for interchange speed rather than storage efficiency. To produce files like this from your own data, use CSV to Parquet.