Docker Dataset

Pre-populated sample databases as Docker images — ready-to-run PostgreSQL, MySQL, CockroachDB, SQLite, DuckDB, ClickHouse, Apache Druid, and Apache Pinot containers loaded with real, valid sample data (Chinook, Northwind, Sakila/Pagila, World, AdventureWorks, Stack Exchange, and more). Ever needed a database already populated with valid data — to practice SQL, run tests, demo an app, or benchmark — without hand-crafting rows or hunting for a usable dump? Every image ships exactly one dataset, so you just docker run and connect.

Quick start

Run the PostgreSQL world image, wait for it to initialize, run a real query, and clean up:

docker run -d --name pg-ds-world aa8y/postgres-dataset:world
# The first start loads the dataset; give it a few seconds. Watch for the
# second "database system is ready to accept connections":
docker logs -f pg-ds-world
docker exec -it pg-ds-world psql -d world -c 'SELECT name, population FROM city ORDER BY population DESC LIMIT 5'
docker rm -f pg-ds-world

That is the whole shape of it: pull a tag from the matrix below, docker run, connect. Each engine connects a little differently (server engines take a client over a port; SQLite and DuckDB open a file), and each has per-dataset notes — see its guide: PostgreSQL · MySQL · CockroachDB · SQLite · DuckDB · ClickHouse · Apache Druid · Apache Pinot.

Dataset support matrix

Each cell is the image tag to pull for that dataset on that engine; — means it isn't shipped there (yet). The dataset name links to its upstream source when every engine pulls from the same one; where engines use different upstreams, the source link is on the individual tag instead. All images are published for linux/amd64 and linux/arm64.

Dataset	PostgreSQL	MySQL	CockroachDB	SQLite	DuckDB	ClickHouse	Apache Druid	Apache Pinot
AdventureWorks	adventureworks	—	—	—	—	—	—	—
Airlines	airlines	—	—	airlines	airlines	—	—	—
Chinook	chinook	chinook	chinook	chinook	chinook	chinook	—	—
Dell DVD Store	dellstore	dellstore	dellstore	dellstore	dellstore	dellstore	dellstore	dellstore
Employees	employees	employees	employees	employees	employees	—	—	—
French Towns	frenchtowns	frenchtowns	frenchtowns	frenchtowns	frenchtowns	frenchtowns	frenchtowns	frenchtowns
GeoNames	geonames	geonames	geonames	geonames	geonames	geonames	geonames	geonames
ISO 3166	iso3166	iso3166	iso3166	iso3166	iso3166	iso3166	iso3166	iso3166
MoMA	moma	moma	moma	moma	moma	moma	moma	moma
Northwind	northwind	northwind	northwind	northwind	northwind	—	—	—
NYC Taxi Trip Records	—	—	—	—	nyc-taxi	nyc-taxi	—	—
OMDb	omdb	—	—	—	—	—	—	—
OpenFlights	openflights	openflights	openflights	openflights	openflights	openflights	openflights	openflights
PGExercises	pgexercises	pgexercises	pgexercises	pgexercises	pgexercises	pgexercises	pgexercises	pgexercises
Sakila / Pagila	pagila	sakila	sakila	sakila	sakila	—	—	—
SportsDB	sportsdb	sportsdb	sportsdb	sportsdb	sportsdb	—	—	—
Stack Exchange¹	stackexchange-<site>	stackexchange-<site>	stackexchange-<site>	stackexchange-<site>	stackexchange-<site>	stackexchange-<site>	—	—
USDA	usda	usda	usda	usda	usda	usda	usda	usda
World	world	world	world	world	world	world	world	world

¹ <site> is one of beer, coffee, poker, woodworking, chess, cooking, outdoors, boardgames (e.g. stackexchange-chess).

Every engine also publishes a latest tag. It is an alias, not a dataset of its own: it opens the same dataset one of the named tags does — world on PostgreSQL, MySQL, Apache Druid, and Apache Pinot; chinook on CockroachDB, SQLite, and DuckDB; and nyc-taxi on ClickHouse.

Tag naming

A tag names the dataset an image carries. How that dataset is addressed once the container is up depends on the engine:

• PostgreSQL, MySQL, CockroachDB, SQLite, DuckDB, ClickHouse put the dataset in a named database (or, for SQLite/DuckDB, a named file). The name is the tag minus any stackexchange- prefix (e.g. stackexchange-beer → beer). Two things are not databases named after the tag: the latest alias opens its target dataset's database (e.g. world, not latest), and on ClickHouse the nyc-taxi tag's database is nyc_taxi, because the base image interpolates the name into SQL unquoted and a hyphen does not survive that.
• Apache Druid and Apache Pinot have no per-dataset database to name — Druid exposes the dataset's tables as datasources in its single druid schema, and Pinot as a flat namespace of tables — so you query the tables directly. Each engine's guide shows how.

Documentation

• Engine guides: PostgreSQL · MySQL · CockroachDB · SQLite · DuckDB · ClickHouse · Apache Druid · Apache Pinot — each has the client invocation, host-connection details, and per-dataset notes for its engine.
• Building images — dave, custom images, and build caching
• Testing — structure tests and integration smoke tests
• Dataset attribution and licenses

Docker Hub repositories: aa8y/postgres-dataset · aa8y/mysql-dataset · aa8y/cockroach-dataset · aa8y/sqlite-dataset · aa8y/duckdb-dataset · aa8y/clickhouse-dataset · aa8y/druid-dataset · aa8y/pinot-dataset

Dataset licenses

This repository's own software and packaging are MIT licensed. Each bundled dataset keeps its upstream license — see docs/ATTRIBUTION.md for the per-dataset sources, licenses, and required attributions (notably the Stack Exchange dumps, which are CC BY-SA 4.0).

Future Work

The remaining matrix gaps, and what each one is waiting on:

• The last matrix gaps. adventureworks and omdb stay PostgreSQL-only by design (their upstreams lean on PostgreSQL-specific machinery — see Datasets not ported to MySQL). airlines on MySQL and CockroachDB is a volume problem, not a dialect one: 10.7M rows replayed as init-time INSERTs would blow the smoke test's readiness budget, so it needs a bulk-load path first (MariaDB's LOAD DATA INFILE, CockroachDB's IMPORT INTO as the employees tag already does) and a ~500 MB data payload in the image.
• More OLAP datasets. ClickHouse carries the pgFoundry family, chinook, nyc-taxi, the CSV datasets and the Stack Exchange sites; its remaining gaps need per-dataset work (northwind's bytea columns, sportsdb's 107 mostly-empty tables — see Datasets not ported to ClickHouse). Druid needs an input format that tolerates embedded newlines (the Stack Exchange sites) or per-tag JVM sizing (nyc-taxi) — see Datasets not ported to Druid. Pinot's obvious next tag is nyc-taxi, which needs its Parquet ingestion job rather than the controller's synchronous CSV endpoint — see Datasets not ported to Pinot.
• More Parquet-native datasets. DuckDB and ClickHouse both ship nyc-taxi from Parquet already (DuckDB reads it at build time; ClickHouse loads it at container start), and the open-data world publishes plenty more sources in that shape.
• More free data sources across every engine.