Parquet Metadata Reader
Upload a .parquet file to surface everything stored in its footer metadata: total row count, number of row groups and their sizes, the writer's created-by string, the compression codec(s) used per column, uncompressed vs compressed size, and any custom key-value metadata. Every section is shown as readable key/value cards plus a full raw JSON dump for scripting or debugging.
Reviewed by Vincenzo Manto, Founder & Lead Developer · Last updated September 27, 2026
How it works
- 1
Upload a .parquet file from your device.
- 2
The file's footer is parsed locally to extract its full FileMetaData object.
- 3
Row group counts, sizes, codecs, and key-value metadata are summarized as cards and tables.
- 4
Expand the raw JSON view to see or copy the complete metadata object.
Every metadata field, in one place
Parquet's footer packs a surprising amount of information: schema, statistics, row group boundaries, encodings, and codecs. Most tools only surface a fraction of this. This reader exposes the whole FileMetaData object hyparquet parses, formatted for humans first and available as raw JSON second.
Useful for debugging pipelines
When a downstream job chokes on a parquet file, the metadata often tells you why: an unexpected codec, a suspiciously large row group, or a missing key-value tag a pipeline expected. Checking this before writing code saves a debugging cycle.
Frequently asked questions
What's a 'row group' in Parquet?
A row group is a horizontal partition of the file's rows, each with its own column chunks and statistics. Parquet writers split large datasets into multiple row groups so readers can skip whole groups that don't match a filter.
Can different columns use different compression codecs?
Yes. Compression is set per column chunk, so one file can legitimately mix SNAPPY, GZIP, or uncompressed columns. This tool lists every distinct codec found across all row groups.
What does 'created by' tell me?
It's a free-text string the writing library embeds in the footer, typically identifying the tool and version that produced the file, e.g. 'parquet-mr version 1.13.1' or 'DuckDB'.
What is key-value metadata used for?
It's an optional list of arbitrary string pairs writers can attach to a file, often used for things like pandas' serialized index information or custom application-level tags.
Is the compression ratio shown accurate?
It's computed directly from the sum of total_uncompressed_size and total_compressed_size across every column chunk in the file's own metadata, so it reflects exactly what's on disk.