pytest-postgresql

What is this?

This is a pytest plugin, that enables you to test your code that relies on a running PostgreSQL Database. It allows you to specify fixtures for PostgreSQL process and client.

How to use

Warning

Tested on PostgreSQL versions >= 10. See tests for more details.

Install with:

pip install pytest-postgresql

You will also need to install psycopg. See its installation instructions. Note that this plugin requires psycopg version 3. It is possible to simultaneously install version 3 and version 2 for libraries that require the latter (see those instructions).

Plugin contains three fixtures:

• postgresql - it's a client fixture that has functional scope. After each test it ends all leftover connections, and drops test database from PostgreSQL ensuring repeatability. This fixture returns already connected psycopg connection.
• postgresql_proc - session scoped fixture, that starts PostgreSQL instance at it's first use and stops at the end of the tests.
• postgresql_noproc - a noprocess fixture, that's connecting to already running postgresql instance. For example on dockerized test environments, or CI providing postgresql services

Simply include one of these fixtures into your tests fixture list.

You can also create additional postgresql client and process fixtures if you'd need to:

from pytest_postgresql import factories

postgresql_my_proc = factories.postgresql_proc(
    port=None, unixsocketdir='/var/run')
postgresql_my = factories.postgresql('postgresql_my_proc')

Note

Each PostgreSQL process fixture can be configured in a different way than the others through the fixture factory arguments.

Sample test

def test_example_postgres(postgresql):
    """Check main postgresql fixture."""
    cur = postgresql.cursor()
    cur.execute("CREATE TABLE test (id serial PRIMARY KEY, num integer, data varchar);")
    postgresql.commit()
    cur.close()

Pre-populating the database for tests

If you want the database fixture to be automatically pre-populated with your schema and data, there are two lewels you can achieve it:

1. per test in a client fixture
2. per session in a process fixture

Both fixtures are accepting same set of possible loaders:

• sql file path - which will load and execute sql files
• loading functions - either by string import path, actual callable. Loading functions will receive host, port, user, dbname and password arguments and will have to perform connection to the database inside. Or start session in the ORM of your choice to perform actions with given ORM. This way, you'd be able to trigger ORM based data manipulations, or even trigger database migrations programmatically.

Per test in a client fixture

Client fixture loads are performed the database each test. Are useful if you create several clients for single process fixture.

from pathlib import Path
postgresql_my_with_schema = factories.postgresql(
    'postgresql_my_proc',
    load=[Path("schemafile.sql"), Path("otherschema.sql"), "import.path.to.function", "import.path.to:otherfunction", load_this]
)

Warning

The database is dropped after each test and before each test using this fixture, it will load the schema/data again.

Warning

client level pre-population is deprecated. Same functionality can be achieved by intermediary fixture between client and test itself.

Per session in a process fixture

The process fixture pre-populates the database once per test session, and loads the schema and data into the template database. Client fixture then creates test database out of the template database each test, which significantly speeds up the tests.

from pathlib import Path
postgresql_my_proc = factories.postgresql_proc(
    load=[Path("schemafile.sql"), Path("otherschema.sql"), "import.path.to.function", "import.path.to:otherfunction", load_this]
)

Additional benefit, is that test code might safely use separate database connection, and can safely test it's behaviour with transactions and rollbacks, as tests and code will work on separate database client instances.

Defining pre-populate on command line:

pytest --postgresql-populate-template=path.to.loading_function --postgresql-populate-template=path.to.other:loading_function --postgresql-populate-template=path/to/file.sql

Connecting to already existing postgresql database

Some projects are using already running postgresql servers (ie on docker instances). In order to connect to them, one would be using the postgresql_noproc fixture.

postgresql_external = factories.postgresql('postgresql_noproc')

By default the postgresql_noproc fixture would connect to postgresql instance using 5432 port. Standard configuration options apply to it.

These are the configuration options that are working on all levels with the postgresql_noproc fixture:

Configuration

You can define your settings in three ways, it's fixture factory argument, command line option and pytest.ini configuration option. You can pick which you prefer, but remember that these settings are handled in the following order:

• Fixture factory argument
• Command line option
• Configuration option in your pytest.ini file

Configuration options

PostgreSQL option	Fixture factory argument	Command line option	pytest.ini option	Noop process fixture	Default
Path to executable	executable	--postgresql-exec	postgresql_exec		/usr/lib/postgresql/13/bin/pg_ctl
host	host	--postgresql-host	postgresql_host	yes	127.0.0.1
port	port	--postgresql-port	postgresql_port	yes (5432)	random
Port search count		--postgresql-port-search-count	postgresql_port_search_count		5
postgresql user	user	--postgresql-user	postgresql_user	yes	postgres
password	password	--postgresql-password	postgresql_password	yes
Starting parameters (extra pg_ctl arguments)	startparams	--postgresql-startparams	postgresql_startparams		-w
Postgres exe extra arguments (passed via pg_ctl's -o argument)	postgres_options	--postgresql-postgres-options	postgresql_postgres_options
Location for unixsockets	unixsocket	--postgresql-unixsocketdir	postgresql_unixsocketdir		$TMPDIR
Database name which will be created by the fixtures	dbname	--postgresql-dbname	postgresql_dbname	yes, however with xdist an index is being added to name, resulting in test0, test1 for each worker.	test
Default Schema either in sql files or import path to function that will load it (list of values for each)	load	--postgresql-load	postgresql_load	yes
PostgreSQL connection options	options	--postgresql-options	postgresql_options	yes

Example usage:

• pass it as an argument in your own fixture

  postgresql_proc = factories.postgresql_proc(
      port=8888)

• use --postgresql-port command line option when you run your tests

  py.test tests --postgresql-port=8888

• specify your port as postgresql_port in your pytest.ini file.

  To do so, put a line like the following under the [pytest] section of your pytest.ini:

  [pytest]
  postgresql_port = 8888

Examples

Populating database for tests

With SQLAlchemy

This example shows how to populate database and create an SQLAlchemy's ORM connection:

Sample below is simplified session fixture from pyramid_fullauth tests:

from sqlalchemy import create_engine
from sqlalchemy.orm import scoped_session, sessionmaker
from sqlalchemy.pool import NullPool
from zope.sqlalchemy import register


@pytest.fixture
def db_session(postgresql):
    """Session for SQLAlchemy."""
    from pyramid_fullauth.models import Base

    connection = f'postgresql+psycopg2://{postgresql.info.user}:@{postgresql.info.host}:{postgresql.info.port}/{postgresql.info.dbname}'

    engine = create_engine(connection, echo=False, poolclass=NullPool)
    pyramid_basemodel.Session = scoped_session(sessionmaker(extension=ZopeTransactionExtension()))
    pyramid_basemodel.bind_engine(
        engine, pyramid_basemodel.Session, should_create=True, should_drop=True)

    yield pyramid_basemodel.Session

    transaction.commit()
    Base.metadata.drop_all(engine)


@pytest.fixture
def user(db_session):
    """Test user fixture."""
    from pyramid_fullauth.models import User
    from tests.tools import DEFAULT_USER

    new_user = User(**DEFAULT_USER)
    db_session.add(new_user)
    transaction.commit()
    return new_user


def test_remove_last_admin(db_session, user):
    """
    Sample test checks internal login, but shows usage in tests with SQLAlchemy
    """
    user = db_session.merge(user)
    user.is_admin = True
    transaction.commit()
    user = db_session.merge(user)

    with pytest.raises(AttributeError):
        user.is_admin = False

Note

See the original code at pyramid_fullauth's conftest file. Depending on your needs, that in between code can fire alembic migrations in case of sqlalchemy stack or any other code

Maintaining database state outside of the fixtures

It is possible and appears it's used in other libraries for tests, to maintain database state with the use of the pytest-postgresql database managing functionality:

For this import DatabaseJanitor and use its init and drop methods:

import pytest
from pytest_postgresql.janitor import DatabaseJanitor

@pytest.fixture
def database(postgresql_proc):
    # variable definition

    janitor = DatabaseJanitor(
        user=postgresql_proc.user,
        host=postgresql_proc.host,
        proc=postgresql_proc.port,
        testdb="my_test_database",
        version=postgresql_proc.version,
        password="secret_password",
    )
    janitor.init()
    yield psycopg2.connect(
        dbname="my_test_database",
        user=postgresql_proc.user,
        password="secret_password",
        host=postgresql_proc.host,
        port=postgresql_proc.port,
    )
    janitor.drop()

or use it as a context manager:

import pytest
from pytest_postgresql.janitor import DatabaseJanitor

@pytest.fixture
def database(postgresql_proc):
    # variable definition

    with DatabaseJanitor(
        user=postgresql_proc.user,
        host=postgresql_proc.host,
        port=postgresql_proc.port,
        dbname="my_test_database",
        version=postgresql_proc.version,
        password="secret_password",
    ):
        yield psycopg2.connect(
            dbname="my_test_database",
            user=postgresql_proc.user,
            password="secret_password",
            host=postgresql_proc.host,
            port=postgresql_proc.port,
        )

Note

DatabaseJanitor manages the state of the database, but you'll have to create connection to use in test code yourself.

You can optionally pass in a recognized postgresql ISOLATION_LEVEL for additional control.

Note

See DatabaseJanitor usage in python's warehouse test code https://github.com/pypa/warehouse/blob/5d15bfe/tests/conftest.py#L127

Connecting to Postgresql (in a docker)

To connect to a docker run postgresql and run test on it, use noproc fixtures.

docker run --name some-postgres -e POSTGRES_PASSWORD=mysecretpassword -d postgres

This will start postgresql in a docker container, however using a postgresql installed locally is not much different.

In tests, make sure that all your tests are using postgresql_noproc fixture like that:

from pytest_postgresql import factories


postgresql_in_docker = factories.postgresql_noproc()
postgresql = factories.postgresql("postgresql_in_docker", dbname="test")


def test_postgres_docker(postgresql):
    """Run test."""
    cur = postgresql.cursor()
    cur.execute("CREATE TABLE test (id serial PRIMARY KEY, num integer, data varchar);")
    postgresql.commit()
    cur.close()

And run tests:

pytest --postgresql-host=172.17.0.2 --postgresql-password=mysecretpassword

Basic database state for all tests

If you've got several tests that require common initialisation, you can to define a load and pass it to your custom postgresql process fixture:

import pytest_postgresql.factories
def load_database(**kwargs):
    db_connection: connection = psycopg2.connect(**kwargs)
    with db_connection.cursor() as cur:
        cur.execute("CREATE TABLE stories (id serial PRIMARY KEY, name varchar);")
        cur.execute(
            "INSERT INTO stories (name) VALUES"
            "('Silmarillion'), ('Star Wars'), ('The Expanse'), ('Battlestar Galactica')"
        )
        db_connection.commit()

postgresql_proc = factories.postgresql_proc(
    load=[load_database],
)

postgresql = factories.postgresql(
    "postgresql_proc",
)

The way this will work is that the process fixture will populate template database, which in turn will be used automatically by client fixture to create a test database from scratch. Fast, clean and no dangling transactions, that could be accidentally rolled back.

Same approach will work with noproces fixture, while connecting to already running postgresql instance whether it'll be on a docker machine or running remotely or locally.

Using SQLAlchemy to initialise basic database state

How to use SQLAlchemy for common initialisation:

def load_database(**kwargs):
    connection = f"postgresql+psycopg2://{kwargs['user']}:@{kwargs['host']}:{kwargs['port']}/{kwargs['dbname']}"
    engine = create_engine(connection)
    Base.metadata.create_all(engine)
    session = scoped_session(sessionmaker(bind=engine))
    # add things to session
    session.commit()

postgresql_proc = factories.postgresql_proc(load=[load_database])

postgresql = factories.postgresql('postgresql_proc') # still need to check if this is actually needed or not

@pytest.fixture
def dbsession(postgresql):
    connection = f'postgresql+psycopg2://{postgresql.info.user}:@{postgresql.info.host}:{postgresql.info.port}/{postgresql.info.dbname}'
    engine = create_engine(connection)

    session = scoped_session(sessionmaker(bind=engine))

    yield session
    # 'Base.metadata.drop_all(engine)' here specifically does not work. It is also not needed. If you leave out the session.close()
    # all the tests still run, but you get a warning/error at the end of the tests.
    session.close()

Release

Install pipenv and --dev dependencies first, Then run:

pipenv run tbump [NEW_VERSION]