Patching pydantic settings in pytest

I’ve been a happy user of pydantic¹ settings to manage all my app configurations since the 1.0 era. When pydantic 2.0 was released, the settings portion became a separate package called pydantic_settings².

It does two things that I love: it automatically reads the environment variables from the .env file and allows you to declaratively convert the string values to their desired types like integers, booleans, etc.

Plus, it lets you override the variables defined in .env by exporting them in your shell.

So if you have a variable called FOO in your .env file like this:

FOO="some_value"

Then you can override it via:

export FOO="other_value"

And pydantic settings will automatically pick up the overridden values without much fuss.

This is neat but can make writing deterministic unit tests tricky. If the settings instance implicitly pulls config values from both the environment file and shell, testing functions using those values can easily become flaky. Also, it’s usually frowned upon if your unit tests depend on environment variables in general.

Consider this common instantiation workflow of the settings class. Here, we have the following app structure:

.
├── src
│   ├── __init__.py
│   ├── config.py
│   └── main.py
├── tests
│   ├── __init__.py
│   └── test_main.py
└── .env

In the src/config.py file, we define our settings class as follows:

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    # Override defaults with .env file values.
    model_config = SettingsConfigDict(env_file=".env")

    env_var_1: str = "default_value"
    env_var_2: int = 123
    env_var_3: bool = False

Then the corresponding values of the environment variables are defined in the .env file. Pydantic will automatically convert the upper-cased definitions to lower-case.

ENV_VAR_1="overridden_value"
ENV_VAR_2="42"
ENV_VAR_3="true"

Next, we instantiate the Settings class in the src/__init__.py file:

from src.config import Settings

settings = Settings()

Finally, we use the config values in src/main.py:

from src import settings


def read_env() -> tuple[str, int, bool]:
    return settings.env_var_1, settings.env_var_2, settings.env_var_3


if __name__ == "__main__":
    env_var_1, env_var_2, env_var_3 = read_env()
    print(f"{env_var_1=}")
    print(f"{env_var_2=}")
    print(f"{env_var_3=}")

From the root directory, run the main.py file with this command:

python -m src.main

This reveals that pydantic settings is doing its magic–reading the .env file and overriding the default config values:

env_var_1='overridden_value'
env_var_2=42
env_var_3=True

Fantastic! But now, testing the read_env function becomes tricky. Normally, you’d try to patch the environment variables in a pytest fixture and then test the values like this:

# tests/test_main.py

import os
from collections.abc import Iterator
from unittest.mock import patch

import pytest
from src.main import read_env


@pytest.fixture
def patch_env_vars() -> Iterator[None]:
    with patch.dict(
        os.environ,
        {
            "ENV_VAR_1": "test_env_var_1",
            "ENV_VAR_2": "456",
            "ENV_VAR_3": "True",
        },
    ):
        yield


def test_read_env(patch_env_vars: None) -> None:
    env_var_1, env_var_2, env_var_3 = read_env()
    assert env_var_1 == "test_env_var_1"
    assert env_var_2 == 456
    assert env_var_3 is True

But the test will fail because we’re initializing the Settings class in the src/__init__.py file and pydantic processes the environment file and variables before pytest can intervene.

We want our unit tests to have no dependencies on the environment variables.

You might say initializing a class in the __init__.py file like that is an anti-pattern and all this can be avoided through dependency injection. You’d be right but you’d also be surprised at how many apps with 7+ figure ARR initialize their config classes like that.

So patching the environment variables doesn’t work, what does?

The idea is to let pydantic do its magic and then reset the attributes of the Settings instance to their default values in a fixture. We also want the user of the fixture to be able to override the values of some or all of the environment variables if necessary.

Here’s what has worked well for me:

import pytest
from src.main import read_env
from src import settings, Settings

import pytest
from collections.abc import Iterator
from pytest import FixtureRequest


@pytest.fixture
def patch_settings(request: FixtureRequest) -> Iterator[Settings]:
    # Make a copy of the original settings
    original_settings = settings.model_copy()

    # Collect the env vars to patch
    env_vars_to_patch = getattr(request, "param", {})

    # Patch the settings to use the default values
    for k, v in settings.model_fields.items():
        setattr(settings, k, v.default)

    # Patch the settings with the parametrized env vars
    for key, val in env_vars_to_patch.items():
        # Raise an error if the env var is not defined in the settings
        if not hasattr(settings, key):
            raise ValueError(f"Unknown setting: {key}")

        # Raise an error if the env var has an invalid type
        expected_type = getattr(settings, key).__class__
        if not isinstance(val, expected_type):
            raise ValueError(
                f"Invalid type for {key}: {val.__class__} instead "
                "of {expected_type}"
            )
        setattr(settings, key, val)

    yield settings

    # Restore the original settings
    settings.__dict__.update(original_settings.__dict__)

Here, patch_settings is a parametrizable fixture where you can optionally pass values via pytest.mark.parametrize to override certain config attributes. If you don’t override anything, the fixture sets the attributes of the Setting instance to their default values defined in the class.

Above, first we make a copy of the original settings instance. Then we reset the attributes of the Setting instance to their default values. Next, we move on to override any values passed via the @parametrize decorator. While doing this, we also check for the correct type of the incoming values and raise an error accordingly.

Finally, we yield the patched instance and reset everything back to their original values after a test ends.

You can use the fixture like this:

def test_read_env(patch_settings: Settings) -> None:
    env_var_1, env_var_2, env_var_3 = read_env()
    assert env_var_1 == "default_value"
    assert env_var_2 == 123
    assert env_var_3 is False


@pytest.mark.parametrize(
    "patch_settings",
    [
        {"env_var_1": "patched_value", "env_var_2": 456},
        {"env_var_2": 459},
    ],
    indirect=True,
)
def test_read_env_override(patch_settings: Settings) -> None:
    env_var_1, env_var_2, env_var_3 = read_env()
    assert env_var_1 == patch_settings.env_var_1
    assert env_var_2 == patch_settings.env_var_2
    assert env_var_3 is patch_settings.env_var_3

In the first case, we’re not overriding anything. So the tests will use the Settings instance with all the default values. In the second test, we’re overriding a few values and the read_env function will use the overridden values.

Either way, the tests don’t directly depend on the environment variables and it reduces the probability of spooky actions at a distance.

Fin!

Recent posts