How to Open a Parquet File Without Python
Someone sent you a .parquet file, or a pipeline produced one, and you want to see what is inside. Double-clicking does nothing useful: a text editor shows binary noise and Excel does not open it. Most guides then tell you to install Python, pandas and PyArrow. You do not need any of them.
Why a text editor cannot open Parquet #
Parquet does not store rows of text like CSV. It stores each column together, compresses it (Snappy, ZSTD or gzip are common), and writes the column names, data types and row counts into a footer at the end of the file. The file starts and ends with the four bytes PAR1. Reading it means decoding that footer and then the compressed columns, which needs a Parquet reader.
Option 1: open it in your browser #
The Parquet Viewer reads the file with DuckDB compiled to WebAssembly, inside the browser tab. Nothing is uploaded and nothing is installed.
- Open the Parquet Viewer and drop the file on it (or press Try a sample file).
- The Data tab shows the rows and the total row count. Search any column, sort by any column, and scroll to load more.
- The Schema tab lists every column with its type:
BIGINT,DECIMAL(10,2),TIMESTAMP, or nestedSTRUCT,LISTandMAPcolumns, which are shown as JSON. - The Metadata tab shows row groups, compression, compressed and uncompressed size, and the library that wrote the file.
- The SQL tab runs DuckDB SQL against the file as the table
data. Export any view or query result as CSV or JSON.
This is the fastest way to answer "what is in this file?" without setting anything up.
Option 2: the DuckDB command line #
If you work in a terminal, DuckDB is a single program with no dependencies (brew install duckdb on macOS, winget install DuckDB.cli on Windows). It queries Parquet files directly:
# First rows
duckdb -c "SELECT * FROM 'orders.parquet' LIMIT 10"
# Column names and types
duckdb -c "DESCRIBE SELECT * FROM 'orders.parquet'"
# Row count, row groups and the writer, from the footer
duckdb -c "SELECT num_rows, num_row_groups, created_by FROM parquet_file_metadata('orders.parquet')"
# Convert to CSV
duckdb -c "COPY (SELECT * FROM 'orders.parquet') TO 'orders.csv' (HEADER)"
Spark and Hive often write a folder of part files instead of one file. DuckDB reads them all with a glob: SELECT COUNT(*) FROM read_parquet('output/*.parquet').
If you do have Python #
For completeness, the usual Python route is pandas with PyArrow installed:
import pandas as pd
df = pd.read_parquet("orders.parquet")
print(df.dtypes)
print(df.head())
pyarrow.parquet.read_metadata("orders.parquet") prints the footer (row groups, row count and writer) without reading the data.
What to check in an unfamiliar Parquet file #
- Row count. It is stored in the footer, so it is exact and instant.
- Types. A column you expected to be a number may be a string, and dates are sometimes stored as plain integers. Check the schema before loading the file elsewhere.
- Nested columns.
STRUCTandLISTcolumns do not flatten into CSV cleanly. They are written as JSON text. - Compression and row groups. Very small row groups (thousands of rows) make large files slow to query. Very large ones use more memory.
- The writer.
created_bytells you whether pandas, Spark, DuckDB or another tool produced it, which helps when types look surprising.
Getting the data into Excel #
Excel cannot open .parquet files directly, so convert to CSV first. For a quick look, export the rows from the Parquet Viewer. For a whole large file, the Parquet to CSV converter writes the CSV locally in your browser.
When a Parquet file will not open #
| Problem | Likely cause |
|---|---|
| "Not a Parquet file" | The file is CSV or JSON that was renamed to .parquet. Renaming does not convert it. |
| The footer cannot be read | The download or copy stopped early; the file must end with PAR1. |
_SUCCESS or .crc files fail |
These are Spark marker and checksum files. Open the part-*.parquet files instead. |
| Out of memory while sorting | Sorting reads every row. Filter first, or use the DuckDB command line, which can spill to disk. |
Summary #
You can read a Parquet file without Python: in the browser with the Parquet Viewer, or in a terminal with DuckDB. Both show the rows, the schema and the footer metadata, and both can convert the file to CSV.
Try the tools from this post
Free and private: your files stay in your browser.