What This Browser Parquet Viewer Inspects
Apache Parquet stores analytical tables column-by-column rather than row-by-row, ending every file with a Thrift metadata footer and the four-byte magic signature PAR1. Because metadata lives at the end of the file, inspecting a dataset locally does not require transmitting the file across an external API.
When you open a .parquet file here, the reader uses File.slice() to decode only the trailing footer bytes first, then extracts column definitions, logical and physical types (INT32,DOUBLE, BOOLEAN, STRING, nested structs), repetition flags, row-group counts, and the compression codec recorded for each column chunk (Uncompressed, Snappy, Gzip, Zstd, Brotli, or LZ4). Once loaded, up to 1,000 preview rows are materialized on demand so you can export clean CSV or JSON files or run ad-hoc SELECT, WHERE, and GROUP BY queries inside an isolated WebAssembly SQLite engine.
Step-by-Step Workflow & Local CLI Alternatives
- Quick verification: Click Try sample .parquet (50 rows) or drop your own
.parquetfile above to inspect row counts, writer software, and column types immediately. - Inspect column encoding: Open the Schema tab to verify physical storage types and confirm whether Snappy or Zstd compression is enabled on each column.
- Filter & export: Switch to SQL Query (Local WASM) to filter or aggregate the loaded rows in table
data, then download results as RFC-4180 CSV or formatted JSON.
If you routinely inspect datasets larger than browser memory allows (hundreds of megabytes or partitioned folders), a native workstation CLI is faster than any web tab. Two reliable one-liners:
# Inspect schema and top 20 rows with DuckDB CLI
duckdb -c "DESCRIBE SELECT * FROM 'orders.parquet';"
duckdb -c "SELECT region, COUNT(*) FROM 'orders.parquet' GROUP BY region;"
# Or convert Parquet to CSV locally with Python + Polars
python3 -c "import polars as pl; pl.read_parquet('orders.parquet').write_csv('orders.csv')"Common Errors & Troubleshooting
parquet file invalid (footer != PAR1): Almost always caused by a truncated S3 download, an empty zero-byte export, or an unclosed temporary file written by an interrupted Spark/Pandas job. Check file size on disk first.- Large 64-bit integers & high-precision decimals: JavaScript numbers safely represent integers up to 253−1. This workbench stringifies
BigIntfields during CSV/JSON serialization so snowflake IDs and nanosecond timestamps never silently round. - Browser tab memory pressure: Avoid loading hundreds of megabytes of wide string columns in a mobile browser. The built-in 1,000-row preview guard keeps inspection responsive while still exposing exact total row counts from the file footer.
Frequently Asked Questions
Does this Parquet viewer upload my file to a cloud server?
No. When you select a file or drop it onto the workbench, only byte ranges needed to decode the PAR1 footer and requested row groups are sliced via the browser File API. Nothing leaves your computer over HTTP.
Which Parquet compression codecs are supported in the browser?
Uncompressed, Snappy, Gzip, Zstandard (Zstd), Brotli, and LZ4 columns are decompressed directly inside the browser using pure-JavaScript decoders bundled with hyparquet and hyparquet-compressors.
Why does local SQL mode limit rows on huge multi-gigabyte Parquet files?
Browser tabs enforce strict JavaScript heap and WebAssembly linear memory limits. Loading up to the first 1,000 rows lets you check column types, spot malformed strings, aggregate samples in SQLite WASM, and export clean slices without crashing your browser tab.
How should I query multi-gigabyte Parquet partitions locally?
For full-dataset aggregations across hundreds of megabytes or partitioned directories, use native DuckDB CLI or Python Polars on your workstation as shown in the CLI section above.