Parquet viewer: see the schema and rows of a .parquet file
Drop a Parquet file to see its row count, its schema (physical and logical types, nesting, which columns can be null), every row group with its size and codec, and the rows themselves. Only the footer is read up front. Rows are decoded straight from the file as you scroll, so the file's size does not matter.
The file is read on your device. It is never uploaded.
What it shows
- Schema: each column's physical type (INT32, INT64, DOUBLE, BYTE_ARRAY …), logical type (STRING, DATE, TIMESTAMP(MICROS, UTC), DECIMAL(10,2) …) and repetition (required, optional or repeated). Fields inside structs and lists are indented under their parent.
- Row groups: rows, uncompressed size, size on disk and codec for each group. A row group is the unit query engines read, so thousands of tiny groups often explain why a file is slow to scan.
- Writer and metadata: the
created_bystring (for exampleparquet-cpp-arrow version 15.0.0) and the names of extra metadata keys such aspandasorARROW:schema. - Rows: decoded in windows of 5,000 rows, only when you scroll to them. Dates show as 2024-01-31, timestamps in UTC ISO form, and lists, maps and structs as JSON. Click a cell to see a long value in full.
Compression it decodes
| Codec | Decoded? | Where it is common |
|---|---|---|
| SNAPPY | yes | Default in pandas, pyarrow and Spark |
| ZSTD | yes | Polars' default; smaller files than snappy |
| GZIP | yes | Older Hadoop and Hive jobs |
| BROTLI | yes | Occasionally chosen for the smallest files |
| LZ4, LZ4_RAW | yes | Speed-focused pipelines |
| UNCOMPRESSED | yes | Test data, some streaming writers |
| LZO | no | Rare, legacy Hadoop |
If a file uses a codec viewhack cannot decode, the page names that codec and still shows the schema and row groups, which live in the uncompressed footer. The decoding is done by hyparquet 1.31.1, fzstd 0.1.1 and hyparquet-compressors 1.1.2, all permissively licensed JavaScript (MIT; the brotli decoder is also Apache-2.0) served from this site.
What it cannot open
- LZO-compressed data: the schema is shown, the rows are not.
- Encrypted Parquet (files starting with
PARE) is recognised and named, but it cannot be read without the key. - Multi-file datasets: open one file at a time. Partition columns that exist only in folder names (
year=2024/) do not appear. - Delta Lake and Iceberg tables: each data file opens, but deletes and updates recorded in the table's log are not applied, so you see the raw file.
- Truncated files: the footer sits at the very end, so an interrupted download cannot be opened. The page tells you when the closing
PAR1is missing. - Sorting and filtering are not available for Parquet yet. Rows appear in file order.
- Very wide files: a 5,000-row window of a file with thousands of columns or very large text values can need a lot of memory, especially on a phone.