Integration · DuckDB
Generate DuckDB test data, with foreign keys intact
Realistic, referentially-correct multi-table datasets you can load into DuckDB with one COPY. Free, no credit card.
Short answer
Design your tables in SynthForge, generate, and export the DuckDB bundle. You get
DuckDB-dialect CREATE TABLE statements with foreign keys inline, CSV data files, and an
import_duckdb.sql script that loads everything with native COPY commands. Every child row
references a real parent row by construction, and the whole thing lands in a
synthforge_import schema.
See the data
| first_name | last_name | date_of_birth |
|---|---|---|
| Tammy | Morrow | 1969-08-30 |
| Paul | Pisacane | 1913-10-03 |
| Raymond | Leung | 1951-01-30 |
| Hildreth | Lowery | 1918-05-10 |
| Joseph | Torres | 1950-04-28 |
| Robert | Mote | 1936-09-25 |
| Kaylee | Bean | 2023-03-24 |
| Marcella | Noha | 2010-03-17 |
| Sierra | Jenkins | 2000-09-11 |
| Monique | McGreevy | 2006-08-16 |
| Reese | Blair | 2016-01-18 |
| Montana | Friel | 2004-05-09 |
Person names follow US Census and SSA frequencies. First names match birth year on the same row.
Try these after you load the pack
SELECT COUNT(*) AS orphan_orders
FROM orders o
LEFT JOIN customers c ON o.customer_id = c.customer_id
WHERE c.customer_id IS NULL;SELECT last_name, COUNT(*) AS n,
ROUND(100.0 * COUNT(*) / SUM(COUNT(*)) OVER (), 3) AS pct
FROM customers
GROUP BY last_name
ORDER BY n DESC
LIMIT 10;SELECT first_name, COUNT(*) AS n
FROM customers
WHERE EXTRACT(YEAR FROM CAST(date_of_birth AS DATE)) BETWEEN 1950 AND 1959
GROUP BY first_name
ORDER BY n DESC
LIMIT 10;Why generate rather than hand-roll
If you have ever populated a DuckDB file for a demo or a benchmark, you have probably written a script with random values, then discovered your orders reference customers that do not exist, then added dedup and parent-lookup code that has nothing to do with what you were testing.
SynthForge resolves the table dependency graph, generates parents first, and draws each foreign key from the parent IDs that actually exist, so the data is referentially valid the moment it lands. A single COPY per table loads it, with no server to stand up.
How to do it
1. Define the schema
Describe it in plain English ("a customers table, an orders table with a customer_id foreign key, line_items referencing orders"), build it in the visual editor, or paste a CREATE TABLE script and let the importer pick out tables, columns, and single-column foreign keys.
2. Set distributions and cardinality (optional)
Numeric columns can follow Normal, LogNormal, Exponential, or Triangular distributions. Set per-relationship ratios like "5 to 20 orders per customer" instead of hand-typing a row count for every child table. This is the step that makes the data look real.
3. Generate and export DuckDB
Pick DuckDB as the SQL dialect. The export bundle contains import_duckdb.sql (the CREATE TABLE DDL with foreign keys declared inline, plus a COPY statement per table) and one csv/<table>.csv data file per table.
4. Load it
Run from the extracted bundle directory so the csv/ paths resolve. One command creates every table and loads every row. Everything goes into a synthforge_import schema, so running this against a database you already use will not touch your own tables:
duckdb mydb.duckdb < import_duckdb.sql
# the script runs native COPY statements like:
# COPY synthforge_import.customers FROM 'csv/customers.csv' (FORMAT CSV, HEADER);
# COPY synthforge_import.orders FROM 'csv/orders.csv' (FORMAT CSV, HEADER);
# query them by schema:
# SELECT count(*) FROM synthforge_import.customers;
#
# somewhere else? find and replace synthforge_import first.5. (Alternative) Use the Parquet export for analytics
SynthForge also exports Parquet, which DuckDB reads natively. Handy when you are benchmarking analytical queries rather than loading tables.
CREATE TABLE orders AS
SELECT * FROM read_parquet('parquet/orders.parquet');Frequently asked questions
Does DuckDB enforce the foreign keys?
How large a dataset can I generate?
Can I get the same schema for Postgres or another engine later?
Does SynthForge use my real data?
Related
Ready to get started?
Design a multi-table schema, generate referentially-intact data, and export the DuckDB bundle. No credit card.