pytest_pg 0.0.29

A tiny plugin for pytest which runs PostgreSQL in Docker

pytest-pg

A pytest plugin that provides session-scoped fixtures for running PostgreSQL inside Docker containers. It automatically spins up a container, waits for PostgreSQL to become ready, exposes connection details (host, port, user, password, database) via a PG dataclass, and tears the container down after the test session. Pre-built fixtures are available for PostgreSQL versions 14 through 18 as well as latest.

Readiness checks work with any of the common drivers — asyncpg, psycopg2, or psycopg3.

To speed up tests, pytest-pg does the following tweaks:

1. fsync=off
2. full_page_writes=off
3. synchronous_commit=off
4. jit=off
5. bgwriter_lru_maxpages=0
6. data directory is mounted to a tmpfs

How to use?

You can use the following fixtures:

• pg – the latest PostgreSQL image available
• pg_14 – PostgreSQL 14
• pg_15 – PostgreSQL 15
• pg_16 – PostgreSQL 16
• pg_17 – PostgreSQL 17
• pg_18 – PostgreSQL 18

import asyncpg


async def test_asyncpg_query(pg):
    conn = await asyncpg.connect(
        user=pg.user,
        password=pg.password,
        database=pg.database,
        host=pg.host,
        port=pg.port,
    )

    await conn.execute("CREATE TABLE test_table (id serial PRIMARY KEY, value text);")
    await conn.execute("INSERT INTO test_table (value) VALUES ($1)", "hello")
    row = await conn.fetchrow("SELECT value FROM test_table WHERE id = $1", 1)

    assert row["value"] == "hello"

    await conn.close()

Also run_pg context manager is available, you can use it to create your own fixture, using docker image you need:

import os

import pytest
import pytest_pg


@pytest.fixture(scope='session', autouse=True)
def postgres_env_vars() -> Generator[None]:
    docker_image = 'postgres:18'
    with pytest_pg.run_pg(docker_image) as pg:
        os.environ['POSTGRES_USER'] = pg.user
        os.environ['POSTGRES_PASSWORD'] = pg.password
        os.environ['POSTGRES_HOST'] = pg.host
        os.environ['POSTGRES_PORT'] = str(pg.port)
        os.environ['POSTGRES_DBNAME'] = pg.database
        yield


# or like so:
@pytest.fixture(scope='session', autouse=True)
def postgres_env_vars(pg_18: pytest_pg.PG) -> Generator[None]:
    os.environ['POSTGRES_USER'] = pg_18.user
    os.environ['POSTGRES_PASSWORD'] = pg_18.password
    os.environ['POSTGRES_HOST'] = pg_18.host
    os.environ['POSTGRES_PORT'] = str(pg_18.port)
    os.environ['POSTGRES_DBNAME'] = pg_18.database
    yield

Using a custom registry

By default the built-in fixtures pull postgres:<version> from Docker Hub. To pull from a mirror instead (for example, to avoid Docker Hub rate limits), set the base image in pyproject.toml. The fixtures append the version tag to it, so the mirror must carry the matching tags: <base>:14 … <base>:18 for the versioned fixtures, and <base>:latest for the pg fixture. Mirrors of pinned versions often omit latest; if yours does, use a versioned pg_NN fixture instead of pg.

[tool.pytest.ini_options]
pg_docker_image_name = "eu.gcr.io/anna-money/postgres"