how to seed test databases reliably

TL;DR

Seed per test with factories, keep seeds deterministic (fixed values, not random), and reset between tests. Reliable seeds make tests hermetic and repeatable.

Error

(Not an error; a data strategy. The failure it prevents: tests depending on ambient database state.)

Steps

1. Use factories (factorybot, factoryboy, fishery) to build only what each test needs. Expected: minimal, explicit data.
2. Make seeds deterministic: fixed names, dates, and IDs where asserted. Expected: repeatable runs.
3. Seed in the test or its setup, not in global fixtures. Expected: tests declare dependencies.
4. Reset between tests (transactions or truncation). Expected: no leakage.
5. Version seed data with the schema; update seeds in the same PR as migrations. Expected: no drift.

When to use

• Setting up test data strategy.
• Flaky or order-dependent data failures.

When not to use

• Production or staging seeding.
• Static reference data (seed once, not per test).

Tool compatibility

• Any ORM; factory libraries per ecosystem.

Variant phrasings

Test database seeding best practices

The general topic; factories plus determinism.

Reliable test fixtures

The goal; per-test, deterministic, reset.

Why it happens

Ambient database state is the top source of test interdependence. Explicit per-test seeds remove it.

Edge cases

• Random data breaks assertions on exact values; use fixed values where asserted.
• Large seed sets slow every test; build the minimum.
• Time-sensitive seeds need clock control too.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_SFKbHHClgeSQq5uHo9a6AA