Pytest Test Categories Plugin

Enforce Google's hermetic testing practices in Python.
Block network, filesystem, and subprocess access in unit tests. Validate your test pyramid. Eliminate flaky tests.

Documentation • PyPI • Quickstart • Why?

---

Quickstart

# Install
pip install pytest-test-categories

# Mark your tests
# @pytest.mark.small  - Fast, hermetic (no I/O)
# @pytest.mark.medium - Can use localhost, filesystem
# @pytest.mark.large  - Full network access

# Enable enforcement in pyproject.toml
# test_categories_enforcement = "strict"

# Run pytest as usual
pytest

---

Table of Contents

• Why pytest-test-categories?
• Test Size Categories
• Installation
• Quick Start
• Configuration
• Enforcement Modes
• Philosophy: No Escape Hatches
• Test Distribution Targets
• Reporting
• Documentation
• Contributing
• License

---

Why pytest-test-categories?

The Problem

Flaky tests are a symptom. Hidden external dependencies are the disease.

Most test suites suffer from:

• Flaky tests - Tests that pass locally but fail in CI due to network timeouts, race conditions, or shared state
• Slow CI pipelines - No time budgets means tests grow unbounded
• Inverted test pyramid - Too many slow integration tests, too few fast unit tests
• No enforced boundaries - "Unit tests" that secretly hit the database, network, or filesystem

The root cause? Tests with hidden external dependencies that make them non-deterministic.

The Solution

pytest-test-categories brings Google's battle-tested testing philosophy (from "Software Engineering at Google") to Python:

What	How
Categorize tests by size	@pytest.mark.small, medium, large, xlarge
Enforce hermeticity	Block network, filesystem, database, subprocess in small tests
Enforce time limits	1s for small, 5min for medium, 15min for large
Validate distribution	Maintain healthy 80/15/5 test pyramid

When a small test tries to access the network, it fails immediately with actionable guidance:

======================================================================
[TC001] Network Violation
======================================================================
Category: SMALL

What happened:
  SMALL test attempted network connection to api.example.com:443

To fix this (choose one):
  • Mock the network call using responses, httpretty, or respx
  • Use dependency injection to provide a fake HTTP client
  • Change test category to @pytest.mark.medium
======================================================================

(back to top)

---

Test Size Categories

Resource	Small	Medium	Large	XLarge
Time Limit	1s	5min	15min	15min
Network	❌ Blocked	Localhost	✓ Allowed	✓ Allowed
Filesystem	❌ Blocked	✓ Allowed	✓ Allowed	✓ Allowed
Database	❌ Blocked	✓ Allowed	✓ Allowed	✓ Allowed
Subprocess	❌ Blocked	✓ Allowed	✓ Allowed	✓ Allowed
Sleep	❌ Blocked	✓ Allowed	✓ Allowed	✓ Allowed

Small tests must be hermetic - completely isolated from external resources. This eliminates flakiness at the source.

(back to top)

---

Installation

pip

pip install pytest-test-categories

uv

uv add pytest-test-categories

Poetry

poetry add pytest-test-categories

(back to top)

---

Quick Start

Basic Usage

Mark your tests with size markers:

import pytest

@pytest.mark.small
def test_unit():
    """Fast, hermetic unit test - no network, no filesystem, no database."""
    assert 1 + 1 == 2

@pytest.mark.small
def test_with_mocking(mocker):
    """Mocked tests are still hermetic - mocks intercept before the network layer."""
    mocker.patch("requests.get").return_value.json.return_value = {"status": "ok"}
    # Your code that uses requests.get() works without hitting the network

@pytest.mark.medium
def test_integration(tmp_path):
    """Integration test - can access localhost and filesystem."""
    db_path = tmp_path / "test.db"
    # ... test with local database

Or inherit from base test classes:

from pytest_test_categories import SmallTest, MediumTest

class DescribeUserService(SmallTest):
    """All tests in this class are automatically marked as small."""

    def test_validates_email_format(self):
        assert is_valid_email("user@example.com")

    def test_rejects_invalid_email(self):
        assert not is_valid_email("not-an-email")

class DescribeUserRepository(MediumTest):
    """Tests requiring database access."""

    def test_saves_user(self, db_connection):
        # ... test with real database

Enable Enforcement

By default, enforcement is off. Enable it in pyproject.toml:

[tool.pytest.ini_options]
# Resource isolation: "off" (default), "warn", or "strict"
test_categories_enforcement = "strict"

# Distribution validation: "off" (default), "warn", or "strict"
test_categories_distribution_enforcement = "warn"

Run pytest as usual:

pytest

(back to top)

---

Configuration

pyproject.toml

[tool.pytest.ini_options]
# Enforcement modes: "strict" (fail), "warn" (warning), "off" (disabled)
test_categories_enforcement = "strict"
test_categories_distribution_enforcement = "warn"

Command-Line Options

CLI options override pyproject.toml settings:

# Enforcement modes
pytest --test-categories-enforcement=strict
pytest --test-categories-distribution-enforcement=warn

# Reporting
pytest --test-size-report=basic      # Summary report
pytest --test-size-report=detailed   # Per-test details
pytest --test-size-report=json       # JSON output
pytest --test-size-report=json --test-size-report-file=report.json

Note: Time limits are fixed per Google's testing standards and cannot be customized. This ensures consistent test categorization across all projects.

(back to top)

---

Enforcement Modes

Mode	Behavior	Use Case
off	No enforcement (default)	Initial exploration
warn	Emit warnings, tests continue	Migration period
strict	Fail tests on violations	Production enforcement

Recommended Migration Path

# Week 1: Discovery - see what would fail
test_categories_enforcement = "off"

# Week 2-4: Migration - fix violations incrementally
test_categories_enforcement = "warn"

# Week 5+: Enforced - violations fail the build
test_categories_enforcement = "strict"

(back to top)

---

Philosophy: No Override Markers

pytest-test-categories intentionally does not provide per-test override markers like @pytest.mark.allow_network.

If a test needs network access, filesystem access, or other resources, it should be marked with the appropriate size:

# Wrong: Trying to bypass restrictions
@pytest.mark.small
@pytest.mark.allow_network  # This marker does not exist!
def test_api_call():
    ...

# Correct: Use the appropriate test size
@pytest.mark.medium  # Medium tests can access localhost
def test_api_call():
    ...

# Or: Mock the dependency for small tests
@pytest.mark.small
def test_api_call(httpx_mock):
    httpx_mock.add_response(url="https://api.example.com/", json={"status": "ok"})
    ...

Why no escape hatches?

1. Flaky tests are expensive - escape hatches become the norm, defeating the purpose
2. Categories have meaning - if a "small" test can access the network, it's not really a small test
3. Encourages better design - mocking and dependency injection lead to more testable code

See the Design Philosophy documentation for the full rationale.

(back to top)

---

Test Distribution Targets

Size	Target Percentage	Tolerance
Small	80%	± 5%
Medium	15%	± 5%
Large/XLarge	5%	± 3%

When distribution enforcement is enabled, pytest will warn or fail if your test distribution falls outside these ranges.

(back to top)

---

Reporting

Terminal Reports

# Basic summary report
pytest --test-size-report=basic

# Detailed report with per-test information
pytest --test-size-report=detailed

JSON Report Export

For CI/CD integration and custom tooling:

# Output JSON report to terminal
pytest --test-size-report=json

# Write JSON report to a file
pytest --test-size-report=json --test-size-report-file=report.json

JSON Report Structure

{
  "timestamp": "2025-11-29T12:00:00.000000Z",
  "summary": {
    "total_tests": 150,
    "distribution": {
      "small": {"count": 120, "percentage": 80.0, "target": 80.0},
      "medium": {"count": 22, "percentage": 14.67, "target": 15.0},
      "large": {"count": 6, "percentage": 4.0, "target": 4.0},
      "xlarge": {"count": 2, "percentage": 1.33, "target": 1.0}
    },
    "violations": {
      "timing": 0,
      "hermeticity": {
        "network": 0,
        "filesystem": 0,
        "process": 0,
        "database": 0,
        "sleep": 0,
        "total": 0
      }
    }
  },
  "tests": [
    {
      "name": "tests/test_example.py::test_fast_function",
      "size": "small",
      "duration": 0.002,
      "status": "passed",
      "violations": []
    }
  ]
}

(back to top)

---

Documentation

For comprehensive documentation, visit pytest-test-categories.readthedocs.io:

• Getting Started - Installation and first steps
• User Guide - Test sizes, isolation, timing, and distribution
• Configuration - All configuration options
• Examples - Common patterns and CI integration
• API Reference - Markers, fixtures, and error messages
• Architecture - Design philosophy and ADRs

Project Resources

• CONTRIBUTING.md - Contribution guidelines
• CODE_OF_CONDUCT.md - Community standards
• SECURITY.md - Security policy
• ROADMAP.md - Project vision and milestones
• GitHub Discussions - Questions and ideas
• Issue Tracker - Bug reports and feature requests

(back to top)

---

Contributing

We welcome contributions! Please read our CONTRIBUTING.md for detailed guidelines.

Quick Start for Contributors

1. Fork and clone the repository
2. Create an issue describing what you plan to work on
3. Create a feature branch from main
4. Make your changes following our coding standards
5. Run pre-commit hooks to ensure quality: uv run pre-commit run --all-files
6. Open a pull request linking to your issue

(back to top)

---

License

This project is licensed under the MIT License.

---

Happy testing!
_{Built with the belief that flaky tests are a solved problem.}