Free Parquet sample, test and dummy files (.parquet)
Columnar data files encoded with Apache Arrow / PyArrow and Snappy compression, featuring typed integer, float, boolean, and string columns. Built for testing DuckDB, Apache Spark, Pandas, Polars, Snowflake, BigQuery, and AWS Athena data ingestion.
Quick facts: .parquet format
| Extension | .parquet |
|---|---|
| Full name | Apache Parquet Columnar Storage Format |
| MIME type | application/vnd.apache.parquet |
| Category | Data |
| Compression / Encoding | Lossless columnar encoding (Dictionary, RLE, Bit-Packing + Snappy/Zstd compression) |
| Magic bytes (signature) | 50 41 52 31 ('PAR1' at both byte offset 0 AND the final 4 bytes of the file) |
| Specification | Apache Parquet Format Specification (Thrift Compact Protocol metadata) |
Download sample Parquet files
Every file below is a spec-compliant .parquet fixture generated deterministically for automated and manual testing. Verify the exact byte count and SHA-256 digest after downloading, or open the custom-size file generator to build additional fixtures in your browser.
| File | Size | Variant & contents | SHA-256 checksum | Download |
|---|---|---|---|---|
| sample-parquet-sales-100.parquet | 5.16 KB5,283 bytes | 100-row Apache Parquet columnar file with Snappy compression (orders schema) | 45f161b81312e7c60be9d60a691be79778b67e81bd87781e848f2e6bdb0398d1 |
Download Parquet |
| sample-parquet-telemetry-1000.parquet | 10.14 KB10,386 bytes | 1,000-row Apache Parquet columnar file with typed numeric and timestamp columns | 1de1224a5d4ab19d8796e261f082a284f90e06213399476c2856064c7425c73f |
Download Parquet |
| sales-100rows.parquet | 6.67 KB6,832 bytes | 100-row sales synthetic dataset in Apache Parquet format | 2642d2aacff42af41d2890d13a4272a9155bcc37d030dc89bfb2dc14041f1229 |
Download Parquet |
| sales-1000rows.parquet | 12.44 KB12,737 bytes | 1000-row sales synthetic dataset in Apache Parquet format | c6a066ae02f5f4168c4d93d369ed0cde7660b9e567ead214a942922f02388005 |
Download Parquet |
| employees-100rows.parquet | 5.71 KB5,850 bytes | 100-row employees synthetic dataset in Apache Parquet format | 5b1565386839b134eeadd80a2ba7b3d7b42bf74fa9c6dc822050f85d183e1c2c |
Download Parquet |
| employees-1000rows.parquet | 16.02 KB16,402 bytes | 1000-row employees synthetic dataset in Apache Parquet format | ebb02c8e1e65dd7d6548940ce7589845cb3c262d33e0fd2c9adbe8ae5a80bd7c |
Download Parquet |
| weather-100rows.parquet | 6.70 KB6,863 bytes | 100-row weather synthetic dataset in Apache Parquet format | a1d90fc3e170e42adc2798373926b6b635763c68f92aeab9bee8bc1416859630 |
Download Parquet |
| weather-1000rows.parquet | 7.48 KB7,660 bytes | 1000-row weather synthetic dataset in Apache Parquet format | 351151491fcb85bcb57a1949f946386d41d5be6ae2bf155c4d860784d9f02de6 |
Download Parquet |
What is a Parquet file?
Apache Parquet is an open-source, column-oriented binary data storage format designed for analytical query engines across the Hadoop, Spark, DuckDB, and cloud data-warehouse ecosystem. Unlike row-oriented formats such as CSV or JSONL—which store all fields of record 1 followed by all fields of record 2—Parquet partitions a table horizontally into Row Groups and stores each column inside a row group contiguously as a Column Chunk divided into data pages.
This layout allows analytical engines to perform projection pushdown (reading only the 2 columns requested by a SQL SELECT query instead of scanning 100 columns) and predicate pushdown (skipping entire row groups by inspecting the min, max, and null_count statistics stored in the file footer). Every valid Parquet file begins with the 4-byte ASCII magic number PAR1 (50 41 52 31) AND ends with the exact same 4-byte PAR1 trailer right after the 4-byte little-endian footer length.
How to open a .parquet file on Windows, macOS, Linux, Android and iOS
- Python:
pandas.read_parquet('sample.parquet'),pyarrow.parquet.read_table(...), orpolars.read_parquet(...). - CLI and SQL Engines: DuckDB (
duckdb -c "SELECT * FROM 'sample-parquet-sales-100.parquet' LIMIT 10;"), ClickHouse Local, andparquet-tools inspect. - macOS, Windows, and Linux IDEs: VS Code Parquet Viewer extensions, DBeaver, JetBrains DataGrip, and Tad.
- In-Browser WebAssembly: DuckDB-Wasm and Hyparquet can query Parquet files directly in the browser via HTTP Range requests.
What to test with Parquet sample files
- Head-and-tail magic byte verification: Because Parquet writes its Thrift schema and row-group metadata in the footer at the end of the file, a truncated upload will still have
PAR1at byte 0 but will lackPAR1in the final 4 bytes. Test that your ingestion validator checks both ends of the file. - Schema evolution and strict typing: Verify how your ETL pipeline maps Parquet physical types (
INT64,DOUBLE,BOOLEAN,BYTE_ARRAYwithUTF8logical annotation) into database table columns without string-coercion precision loss. - HTTP Range request partial reads: Host the 1,000-row Parquet sample and verify that DuckDB or PyArrow can read the schema footer via a tail
Range: bytes=-8request without downloading the entire file.
Common problems when handling .parquet files
The most common failure when testing Parquet uploads is corrupting the binary payload by reading it as UTF-8 text in an API gateway or multipart form parser. Always treat application/vnd.apache.parquet (or application/octet-stream) as raw binary bytes and verify the trailing PAR1 marker before passing the buffer to Arrow or Spark.
Need to convert or inspect a file right away? Use ConvertFerry Data & JSON Formatter Tools or drop any local file into our in-browser Magic-Byte File Inspector.
Related formats and testing resources
Frequently asked questions
- Why does a Parquet file have PAR1 at both the beginning and the end?
- Writers stream column chunks sequentially and only know the exact byte offsets and min/max statistics once all rows are written. Placing the Thrift FileMetaData structure at the end followed by its 4-byte length and a trailing PAR1 magic marker lets readers seek to EOF-8, read the footer length, and parse the schema in a single seek.
- What is the difference between Apache Parquet and Apache Avro?
- Parquet is a columnar storage format optimized for analytical OLAP queries that aggregate or filter specific columns, whereas Avro is a row-oriented format optimized for Kafka event streaming and write-heavy transactional records.
- What compression codec is used in these sample Parquet files?
- They use Google Snappy block compression, which is the default compression codec across Apache Spark, PyArrow, and AWS Athena.
- What columns are included in the sample Parquet files?
- Each file contains 6 typed columns: record_id (INT64), sku_code (UTF8 string), region (UTF8 string), unit_price (DOUBLE), quantity (INT64), and in_stock (BOOLEAN).