How to convert Parquet to CSV in your browser
Parquet is a binary, column-based file format used by Spark, pandas, DuckDB, Athena, BigQuery and many data exports. Excel and most text editors cannot open it. To convert parquet to csv, this page reads the file with a JavaScript Parquet reader, shows what is inside and writes the rows out as text.
The Example button builds a five-row customer file. Converted with the default options it gives:
id,zip,name,joined,last_seen,balance,active,visits,country
1,02134,Ana Souza,2024-01-15,2024-03-01T09:30:00.000Z,1250.5,true,12,BR
2,10001,Ben Carter,2024-02-03,2024-03-02T18:05:10.000Z,0,false,3,US
3,00501,Chloé Martin,2023-11-30,2024-02-28T07:00:00.000Z,-87.25,true,40,FR
4,94105,Dev Patel,2024-03-09,,310,true,,IN
5,00000,Eli Cohen,2022-07-21,2024-03-05T23:59:59.250Z,19999.99,false,7,NAText columns such as the zip codes keep their leading zeros, dates are written as plain dates, timestamps as ISO 8601 text in UTC, and the two missing values are left empty.
Open and view a Parquet file online
Because the summary and preview appear as soon as the file is read, the page works as a Parquet viewer too. You do not need to export anything. The summary strip answers the first questions people ask about a file: how many rows, which columns, whether it is compressed and by what, and which program wrote it.
The Columns tab lists each column with its Parquet type, such as STRING, INT64, DECIMAL(10,2) or TIMESTAMP (micros), and whether it can hold nulls. The preview shows the first 100 rows of up to 40 columns. Columns past the 40th are not shown but are still exported.
Exact numbers: big integers, decimals and timestamps
Many viewers read every number as a JavaScript double, which rounds whole numbers above 9,007,199,254,740,991. Here a 64-bit integer is read as a big integer and written digit for digit. A column declared as DECIMAL(18,3) is rebuilt from its stored integer and scale, so 123456789012345678 with scale 3 becomes 123456789012345.678 instead of a rounded value.
- Timestamps keep their unit: millisecond, microsecond and nanosecond values are written with 3, 6 or 9 fractional digits, in UTC.
- Dates become YYYY-MM-DD and times of day become HH:MM:SS with fractions.
- FLOAT columns print the shortest text that reads back as the same 32-bit value, so 0.1 stays 0.1.
- Binary columns are written as text when the bytes are valid UTF-8 and as 0x hex otherwise.
In JSON output, whole numbers beyond the safe range and wide decimals are written as text strings, because a JSON reader would round them as numbers.
Nested columns: structs, lists and maps
A CSV cell holds one value, so a struct, list or map column is written as JSON text inside the cell. The Columns tab marks these columns. Numbers inside nested values go through JSON, so a decimal that sits inside a struct is not rebuilt with the exact scale the way a top-level decimal column is.
id,tags
1,"[""new"",""vip""]"If you need those values as real columns, export JSON or JSON Lines and flatten them in your own tool.
CSV options explained
- Separator: comma, semicolon or tab. Values that contain the separator, a quote or a line break are put in double quotes.
- Header row: write the column names on the first line, or leave them out.
- Quote all: put every value in double quotes.
- UTF-8 BOM: adds a byte-order mark at the start, which helps Excel on Windows show accents correctly.
- Null values: write nothing, or the word NULL. A CSV cannot tell a null from an empty string, so pick NULL when that difference matters.
Lines end with a carriage return and line feed (CRLF), which Excel and most CSV readers accept.
Compression, size limits and errors
Parquet compresses each column, so the CSV is usually several times larger than the Parquet file. Snappy, Gzip and uncompressed files were tested for this page. Brotli, Zstandard and LZ4 files are handled by the reading library and use the same code path.
The page opens files up to 1 GB. The file is read one row group at a time, but the finished CSV is built in your device's memory, so a very large export can still fail on a small device. If the browser runs out of memory, the page says so and you can choose fewer columns or a smaller file.
A file that is not Parquet, is cut short, has a damaged footer or uses an encrypted footer is rejected with a message that says which. Parquet files with encrypted columns and the INTERVAL and BSON column types are not supported.
Convert Parquet to CSV with pandas or DuckDB
- pandas: pd.read_parquet("data.parquet").to_csv("data.csv", index=False)
- DuckDB: COPY (SELECT * FROM 'data.parquet') TO 'data.csv' (HEADER, DELIMITER ',');
Use code when the file has hundreds of millions of rows or when the job must run on a schedule. Use this page for a quick look, for a file you received by email, or on a computer where you cannot install Python.
Privacy
The file is read and converted in your browser and nothing is sent to a server. The Privacy Policy has the details.
Related tools
To go back, use CSV to Parquet. To join several CSV files, try Merge CSV Files. To read JSON Lines output, open JSONL Viewer, and to show a CSV as a web table, use CSV to HTML Table.