Parquetastic (https://parquetastic.dev) is a web app to inspect the internal structure of Apache Parquet files. Immediately understand your row groups, column chunks, pages and other metadata without first downloading a CLI tool or writing custom code!
Data never leaves your device. The Parquet file is fully processed locally in your browser.
- Hierarchical metadata view — browse row groups, column chunks, and individual pages (if your file has a page index)
- File layout diagram — visual byte-level map of how the file is structured on disk
- Page and column chunk level details — types, encodings, compression, min/max statistics
- Two layout modes — rows (column chunks as horizontal rows) and columns (vertical strips showing page alignment across row groups)
- Large file support — reads only the metadata, so even 10 GB+ Parquet files load instantly
- Light and dark mode with system preference detection
To inspect your Parquet files, simply drop them into https://parquetastic.dev. To develop locally, ensure you have npm / Node.js installed and then run
npm install
npm run devto install dependencies and start a development server. This will print a locahost URL on which you can find the running application. Your changes will reflect instantly.
Parquetastic reads Parquet files entirely in the browser:
- Magic bytes — validates the file starts and ends with
PAR1 - Footer — reads the Thrift-encoded
FileMetaDatafrom the end of the file - Page indexes — if available, read the
ColumnIndexandOffsetIndexstructures for each column chunk - Visualization — render the parsed metadata as interactive diagrams
The Thrift deserialization code uses auto-generated type definitions and parsing code based on the Parquet format spec.
- React 19
- Vite
- Tailwind CSS 3
Contributions are welcome and I will review them as time permits. Please open an issue first if you're planning a larger change.
Kudos to Jan Finis, my colleague at Salesforce, who was instantly a fan of this tool and who encouraged me to open-source it.
Big thanks to my good friends Opus and Sonnet for helping me build this tool by writing most of the code.