How to open a Parquet file in Excel

Power Query has a Parquet connector, but Microsoft lists it for Power BI and Power Query Online only, not Excel, and double-clicking a .parquet file doesn't open it either. Whatever route you take, the result has to fit one sheet: 1,048,576 rows by 16,384 columns.

rows.page opens .parquet and .pq files in your browser with DuckDB-WASM. Parquet isn't held to the 1.5 GB limit for text formats because it's read in place, and nothing is uploaded unless you press Share. rows.page doesn't write .xlsx, so you export a CSV and import that into Excel.

1. Drop the .parquet file on rows.page. Rows load 100 at a time as you scroll.
2. Check each column's type, empty values and min and max in the Columns panel, and hide the columns you don't need.
3. Cut the rows down to what fits in a sheet with the filter box (country:NL, price:>20, or -name:* for empty values), a histogram bar, or SQL against the data table like the query below.
4. Open the Export menu and choose CSV. A filtered view exports every matching row without the hidden columns; a SQL result exports its own columns, up to 1,000,000 rows.
5. In Excel, choose Data > Get Data > From File > From Text/CSV, pick the file, set File Origin to 65001: Unicode (UTF-8) and select Load.

SELECT * EXCLUDE (__rid)
FROM data
WHERE price > 20
LIMIT 500000

Don't double-click the CSV. Excel only assumes UTF-8 when a CSV starts with a byte order mark (BOM), and rows.page writes UTF-8 without one, so accented text turns into junk such as é. Text cells that start with =, +, -, @, a tab or a carriage return get a leading apostrophe so Excel doesn't run them as formulas.

DuckDB 1.2 and later writes .xlsx files with its excel extension. Run this in the DuckDB CLI:

INSTALL excel;
LOAD excel;
COPY (SELECT * FROM 'data.parquet') TO 'data.xlsx' (FORMAT xlsx, HEADER true);

The header counts toward the sheet limit: 1,048,575 data rows fit, and one more stops the export with Sheet row limit of '1048576' rows exceeded!. Add a WHERE clause or LIMIT 1048575 inside the SELECT to stay under it. Struct and list columns are written as text such as {'name': A, 'city': X} and [a, b].

pandas reads Parquet through pyarrow and writes .xlsx through openpyxl:

pip install pandas pyarrow openpyxl

import pandas as pd

df = pd.read_parquet('data.parquet')
df.to_excel('data.xlsx', index=False)

to_excel raises This sheet is too large! when the DataFrame has more rows or columns than a sheet holds, so filter it first. Timestamps with a time zone stop it with Excel does not support datetimes with timezones; drop the zone with df['ts'] = df['ts'].dt.tz_localize(None). Struct cells are written as Python text such as {'name': 'A', 'city': 'X'}.

Parquet is typed and can nest data, while a sheet is a flat grid. Watch for these:

• Nested columns: rows.page flattens struct columns into dot-path columns such as user.address.city, up to 8 levels deep and 200 fields per struct, and the CSV export keeps them as ordinary columns. In SQL, quote those names: "user.address.city".
• Lists: the grid shows list values as chips, and the CSV export writes them as JSON text in a single cell.
• Timestamps: the CSV holds them as text such as 2024-01-01 12:00:00, with an offset such as +00 when the column has a time zone. Check the type Power Query picks in the preview.
• Row count: no route puts 2,000,000 rows on one sheet. Aggregate with SQL GROUP BY, filter first, or load the CSV into Excel's Data Model with Load To, Only Create Connection and Add this data to the Data Model.