How to query a Parquet file with SQL

By Mykola-Bohdan Vynnytskyi · Updated November 7, 2026

Open the file in Parquet Reader, type a query in the SQL Query panel and press ⌘Return. The file is available as a table called parquet_table. You do not need Python, Spark or a database server, and SQL is free. On the command line, DuckDB does the same in one line.

Run SQL in Parquet Reader

1. Open the file. The panel is already filled in with SELECT * FROM parquet_table LIMIT 100.
2. Replace the query with your own.
3. Press ⌘Return or click Run Query.
4. The result replaces the table. You can search it, copy it or export it. Show Original or ⌘⇧R brings the file back.

Filter and sort

SELECT *
FROM parquet_table
WHERE country = 'DE'
ORDER BY amount DESC
LIMIT 100

Group and aggregate

SELECT country, count(*) AS orders, sum(amount) AS revenue
FROM parquet_table
WHERE status = 'paid'
GROUP BY country
ORDER BY revenue DESC

┌─────────┬────────┬─────────┐
│ country │ orders │ revenue │
│ varchar │ int64  │ double  │
├─────────┼────────┼─────────┤
│ DE      │      1 │   120.0 │
│ UA      │      1 │    49.9 │
│ PL      │      1 │    NULL │
└─────────┴────────┴─────────┘

Group by month

SELECT date_trunc('month', order_date) AS month, count(*) AS orders
FROM parquet_table
GROUP BY month
ORDER BY month

Join two files

1. Open the first file, click Save as table and name it, here orders.
2. Open the second file and save it the same way, here countries.
3. Use both names in one query.

SELECT c.name, count(*) AS orders, sum(o.amount) AS revenue
FROM orders o
JOIN countries c ON o.country = c.code
GROUP BY c.name
ORDER BY orders DESC, c.name

┌─────────┬────────┬─────────┐
│  name   │ orders │ revenue │
│ varchar │ int64  │ double  │
├─────────┼────────┼─────────┤
│ Germany │      2 │  200.25 │
│ Ukraine │      2 │    65.4 │
│ Poland  │      1 │    NULL │
└─────────┴────────┴─────────┘

A saved table points at the file. It is not a copy, so the file has to stay where it is. Saved tables are still in the list after a restart, and Save as table keeps a query result too, so you can query it like any other table.

What the app accepts

• A query must start with SELECT. WITH is rejected, so write a subquery in FROM instead.
• One statement at a time, read-only. Nothing you type can change the file.
• A query sees your tables only. parquet_table is the table on screen, and saved tables go by their names. Functions that read other files, such as read_parquet, are blocked.

SELECT *
FROM (
  SELECT country, sum(amount) AS revenue
  FROM parquet_table
  GROUP BY country
)
WHERE revenue > 50

Other ways to run SQL on Parquet

DuckDB command line

Use the file name where the table name would go:

duckdb -c "SELECT country, count(*) AS orders, sum(amount) AS revenue FROM 'orders.parquet' WHERE status = 'paid' GROUP BY country ORDER BY revenue DESC"

pandas

This one is not SQL, but it answers the same question:

import pandas as pd

df = pd.read_parquet("orders.parquet", columns=["country", "status", "amount"])
paid = df[df["status"] == "paid"]
print(paid.groupby("country")["amount"].agg(["count", "sum"]))

         count    sum
country
DE           1  120.0
PL           0    0.0
UA           1   49.9

Look at PL. Its only paid order has no amount. SQL counts one order and returns a NULL sum. This pandas version counts non-null values, so it reports zero orders and a sum of 0.0. If a report moved from one to the other, this is where the numbers change.

Reading less of the file

pd.read_parquet loads the columns you ask for into memory, and the filter runs afterwards. DuckDB's read_parquet and polars' scan_parquet push the filter and the column list down into the scan, so they can skip data the query does not touch.

What the free version does not do

SQL is free, without a time limit or a query limit, and so are joins between files, with any number of tables. Reading a whole folder as one table is Pro. See the guide to partitioned folders.