examples

drizzle-saat examples

Six runnable projects, small to large, each showing how complex, reproducible data comes from very little fixture code. Every example uses Bun's built-in bun:sqlite, so they run with zero native setup:

cd examples/<name>
bun install
bun run seed      # seed and print a report
bun run verify    # seed twice with the same seed → assert byte-identical data

#	Example	Size	Fixture LOC	Rows	Highlights
1	basic-blog	small	~25	~125	count+data, keyed rows, ref().random()
2	ecommerce-store	mid	~70	~350	FK chains + topo-order, ref().where(), multi-file fixtures
3	saas-multitenant	mid-large	~80	~125	tenant scoping, composite-PK join table
4	social-network	large	~90	~1100	8 namespaces, nested refs in JSON, threaded comments
5	analytics-events	mid-large	~60	50–5050	--scenario datasets, high volume
6	testing-with-drizzle-saat	practical	~25	per-test	deterministic fixtures in a bun test suite

The pitch, in one screen

// social-network/drizzle-saat/content.ts — 120 posts, each with a nested ref in JSON
{
  table: posts, namespace: "post", count: 120,
  data: () => ({
    authorId: ref("user").random(),
    body: faker.lorem.paragraph(),
    metadata: { lang: "en", pinnedBy: ref("user", "ada") }, // ← resolved to an id
    createdAt: faker.date.recent({ days: 30, refDate: "2026-06-01" }).toISOString(),
  }),
}

That's the whole posts dataset. drizzle-saat resolves the refs (top-level and nested), orders inserts by dependency, and produces the exact same rows for a given seed.

Type checking

Every example is wired for full fixture type-safety. After bun install (which generates .drizzle-saat/types.d.ts via a postinstall hook), run:

bun run typecheck   # regenerates types, then `tsc --noEmit`

This catches, at compile time:

ref("dwad")                       // ✗ "dwad" is not a known namespace
priority: "URGENT"                // ✗ not in the column's enum
title: ref("user").random()       // ✗ a number ref can't fill a string column
data: () => ({ title: "t" })      // ✗ missing required NOT NULL columns

Each example's tsconfig.json lists ".drizzle-saat/types.d.ts" explicitly — TypeScript's include globs skip dot-directories, so ".drizzle-saat" alone is silently ignored and namespace checking won't work. (See the README's "Type safety" section.)

Reproducibility note

A fixed seed makes runs byte-identical — except for fakers anchored to the clock. Always pass a fixed refDate to faker.date.* helpers, or those columns will differ every run. See ecommerce-store for the details.

Using a real database

Each example uses bun:sqlite purely so it runs anywhere. In your own project you don't write a runner — point drizzle.config.ts at your database and use the CLI (npx drizzle-saat, npx drizzle-saat --scenario load, npx drizzle-saat --seed 7). The fixtures are identical; switch dialect to postgresql or mysql and nothing else changes.