Converting Parquet Columns Back to JSON or NDJSON
API fixtures, Elasticsearch seeders, document stores, and LLM evaluation harnesses expect row-oriented JSON rather than binary columnar Parquet blocks. Switching output modes between formatted JSON Array (.json) and newline-delimited JSONL (.jsonl) lets you reuse the same local Parquet source for both REST payloads and streaming CLI pipelines.
This browser workbench decodes Snappy, Zstd, Gzip, Brotli, and LZ4 columns without server uploads, safely converting BigInt integers to string scalars and JavaScript Date objects to ISO-8601 timestamps so JSON.stringify never crashes on analytics exports.
Step-by-Step Workflow & Local CLI Commands
- Load your
.parquetfile or click Try sample .parquet (50 rows) above. - Toggle between JSON Array (.json) and JSONL / NDJSON (.jsonl) and inspect the live formatted output.
- Click Download JSON / JSONL, or switch to the SQL Query tab first to extract a filtered subset.
For multi-hundred-megabyte full-file conversions on your machine, use DuckDB CLI:
# Export Parquet to newline-delimited JSONL locally via DuckDB
duckdb -c "COPY (SELECT * FROM 'orders.parquet') TO 'orders.jsonl' (FORMAT JSON, ARRAY false);"
# Or export as a single JSON array
duckdb -c "COPY (SELECT * FROM 'orders.parquet') TO 'orders.json' (FORMAT JSON, ARRAY true);"Known Limitations & Precision Checks
- Large file guard: Interactive browser materialization loads up to 1,000 rows so wide nested schemas preview and convert instantaneously without hitting browser tab RAM ceilings.
- Binary BYTE_ARRAY fields: Non-UTF-8 raw byte columns render as compact descriptors rather than corrupting output UTF-8 text.
Frequently Asked Questions
What is the difference between JSON Array and JSONL / NDJSON output?
JSON Array wraps all records inside [...] so standard JSON parsers read the whole file at once. JSONL (newline-delimited JSON) puts one valid JSON object per line without commas between lines, making it ideal for log pipelines and LLM datasets.
How does the converter handle 64-bit BigInt columns from Parquet?
Standard JSON.stringify throws a TypeError on JavaScript BigInt values. This workbench converts BigInts cleanly to strings so timestamp nanosecond columns and snowflake IDs never fail export.
Are Parquet files ever sent to an API during JSON export?
No. All decompression and JSON formatting execute strictly inside browser memory after loading the page.