Fresh test data in every CI run

Why CI test data is a problem

• Fixtures rot. One non-null column and every fixture breaks across the suite.
• Production dumps in CI are a security and GDPR liability, and they are huge.
• Stale shared data makes tests flaky: a row another test changed, an ID that drifted.

What a pipeline wants is a fresh, correct database per run, generated, not copied, and the same every time so failures are real.

One command in your pipeline

Spin up an ephemeral database, generate a deterministic dataset, load it, run your tests. A GitHub Actions job:

# .github/workflows/test.yml
jobs:
  test:
    runs-on: ubuntu-latest
    services:
      db:
        image: postgres:16
        env: { POSTGRES_PASSWORD: postgres }
        ports: ["5432:5432"]
    env:
      DATABASE_URL: postgres://postgres:postgres@localhost:5432/postgres
      SEEDBASE_TOKEN: ${{ secrets.SEEDBASE_TOKEN }}
    steps:
      - uses: actions/checkout@v4
      - run: pip install seedbase
      - name: Seed the test database
        run: |
          seedbase generate --project ${{ vars.SEEDBASE_PROJECT }} \
                            --seed 20240601 --format postgresql --wait
          seedbase pull data --target "$DATABASE_URL" --replace
      - run: pytest -q

The schema is dropped after the run (if: always()), so nothing leaks between builds. The same shape works on GitLab CI, CircleCI or any runner, the CLI is just pip install seedbase.

Deterministic by seed, reproducible across PRs

Generation is deterministic: pin SEEDBASE_SEED and every pull request gets the exact same dataset, so a failing test is a real failure, not data drift. Commit the generation config as code next to your migrations and the whole team regenerates the same data from the same schema.

Or pull it straight into your tests, with the pytest plugin

Install SeedBase next to pytest and request a fixture. It is opt-in and side-effect free at import, so it cannot break an existing suite, and it skips (not fails) if credentials are missing.

pip install seedbase pytest

def test_orders_have_customers(seeded_data):
    dump = seeded_data.decode("utf-8")
    assert "INSERT INTO orders" in dump
    assert "INSERT INTO customers" in dump

def test_generation_is_deterministic(seedbase_generation):
    assert seedbase_generation["status"] == "completed"

Fixtures: seedbase_client, seedbase_project, seedbase_generation, seeded_data. Configure via env vars, pytest.ini / pyproject.toml, or CLI flags.

Works with your stack

• Any CI: GitHub Actions, GitLab CI, CircleCI, any runner. The CLI is one pip install.
• Postgres and MySQL, export as SQL/CSV/JSON, or push straight into the database.
• Python SDK and a Node SDK, plus an MCP server if your AI agent drives the pipeline.