pytest-neon

Pytest plugin for Neon database branch isolation in tests.

Each test gets its own isolated database state via Neon's instant branching and reset features. Branches are automatically cleaned up after tests complete.

Features

Isolated test environments: Each test runs against a clean database state
Fast resets: ~0.5s per test to reset the branch (not create a new one)
Automatic cleanup: Branches are deleted after tests, with auto-expiry fallback
Zero infrastructure: No Docker, no local Postgres, no manual setup
Real database testing: Test against actual Postgres with your production schema
Automatic DATABASE_URL: Connection string is set in environment automatically
Driver agnostic: Bring your own driver, or use the optional convenience fixtures

Installation

Core package (bring your own database driver):

pip install pytest-neon

With optional convenience fixtures:

# For psycopg v3 (recommended)
pip install pytest-neon[psycopg]

# For psycopg2 (legacy)
pip install pytest-neon[psycopg2]

# For SQLAlchemy
pip install pytest-neon[sqlalchemy]

# Multiple extras
pip install pytest-neon[psycopg,sqlalchemy]

Quick Start

Set environment variables:

export NEON_API_KEY="your-api-key"
export NEON_PROJECT_ID="your-project-id"

Write tests:

def test_user_creation(neon_branch_isolated):
    # DATABASE_URL is automatically set to the test branch
    import psycopg  # Your own install

    with psycopg.connect() as conn:  # Uses DATABASE_URL by default
        with conn.cursor() as cur:
            cur.execute("INSERT INTO users (email) VALUES ('[email protected]')")
        conn.commit()
    # Branch automatically resets after test - next test sees clean state

Run tests:

pytest

Fixtures

Which fixture should I use?

Fixture	Scope	Reset	Best For	Overhead
`neon_branch_readonly`	session	None	Read-only tests (SELECT) - enforced	~0s/test
`neon_branch_dirty`	session	None	Fast writes, shared state OK	~0s/test
`neon_branch_isolated`	function	After each test	Write tests needing isolation	~0.5s/test
`neon_branch_shared`	module	None	Module-level shared state	~0s/test

Quick guide:

Use neon_branch_readonly if your test only reads data (SELECT queries). This uses a true read-only endpoint that enforces read-only access at the database level. Any write attempt will fail with a database error.
Use neon_branch_dirty if your tests write data but can tolerate shared state across the session. Fast because no reset or branch creation per test.
Use neon_branch_isolated if your test modifies data and needs isolation from other tests. Resets the branch after each test.

Architecture

The plugin creates a branch hierarchy to efficiently support all fixture types:

Parent Branch (configured or project default)
    └── Migration Branch (session-scoped, read_write endpoint)
            │   ↑ migrations run here ONCE
            │
            ├── Read-only Endpoint (read_only endpoint ON migration branch)
            │       ↑ neon_branch_readonly uses this (enforced read-only)
            │
            ├── Dirty Branch (session-scoped child, shared across ALL workers)
            │       ↑ neon_branch_dirty uses this
            │
            └── Isolated Branch (one per xdist worker, lazily created)
                    ↑ neon_branch_isolated uses this, reset after each test

`neon_branch_readonly` (session-scoped, enforced read-only)

Use this fixture for tests that only read data. It uses a true read_only endpoint on the migration branch, which enforces read-only access at the database level. Any attempt to INSERT, UPDATE, or DELETE will result in a database error.

def test_query_users(neon_branch_readonly):
    # DATABASE_URL is set automatically
    import psycopg
    with psycopg.connect(neon_branch_readonly.connection_string) as conn:
        result = conn.execute("SELECT * FROM users").fetchall()
        assert len(result) >= 0

        # This would fail with a database error:
        # conn.execute("INSERT INTO users (name) VALUES ('test')")

Use this when:

Tests only perform SELECT queries
Tests don't modify database state
You want maximum performance with true enforcement

Session-scoped: The endpoint is created once and shared across all tests and workers.

Performance: ~1.5s initial setup per session, no per-test overhead. For 10 read-only tests, expect only ~1.5s total overhead.

`neon_branch_dirty` (session-scoped, shared state)

Use this fixture when your tests write data but can tolerate shared state across the session. All tests share the same branch and writes persist (no cleanup between tests). This is faster than neon_branch_isolated because there's no reset overhead.

def test_insert_user(neon_branch_dirty):
    # DATABASE_URL is set automatically
    import psycopg
    with psycopg.connect(neon_branch_dirty.connection_string) as conn:
        conn.execute("INSERT INTO users (name) VALUES ('test')")
        conn.commit()
    # Data persists - subsequent tests will see this user

def test_count_users(neon_branch_dirty):
    # This test sees data from previous tests
    import psycopg
    with psycopg.connect(neon_branch_dirty.connection_string) as conn:
        result = conn.execute("SELECT COUNT(*) FROM users").fetchone()
        # Count includes users from previous tests

Use this when:

Most tests can share database state without interference
You want maximum performance with minimal API calls
You manually manage test data cleanup if needed
You're combining it with neon_branch_isolated for specific tests that need clean state

Warning: Data written by one test WILL be visible to subsequent tests AND to other xdist workers. This is truly shared - use neon_branch_isolated for tests that require guaranteed clean state.

pytest-xdist note: ALL workers share the same dirty branch. Concurrent writes from different workers may conflict. This is "dirty" by design - for isolation, use neon_branch_isolated.

Performance: ~1.5s initial setup per session, no per-test overhead. For 10 write tests, expect only ~1.5s total overhead.

`neon_branch_isolated` (function-scoped, full isolation)

Use this fixture when your test modifies data and needs isolation from other tests. Each xdist worker has its own branch, and the branch is reset to the migration state after each test.

def test_insert_user(neon_branch_isolated):
    # DATABASE_URL is set automatically
    import psycopg
    with psycopg.connect(neon_branch_isolated.connection_string) as conn:
        # Guaranteed clean state - no data from other tests
        # (only migration/seed data if you defined neon_apply_migrations)
        result = conn.execute("SELECT COUNT(*) FROM users").fetchone()
        initial_count = result[0]

        conn.execute("INSERT INTO users (name) VALUES ('test')")
        conn.commit()

        # Verify our insert worked
        result = conn.execute("SELECT COUNT(*) FROM users").fetchone()
        assert result[0] == initial_count + 1
    # Branch resets after this test - next test sees clean state

Use this when:

A test modifies database state (INSERT, UPDATE, DELETE)
Test isolation is important
You're combining it with neon_branch_dirty for most tests but need isolation for specific ones

SQLAlchemy Users: If you create your own engine (not using the neon_engine fixture), you MUST use pool_pre_ping=True:

engine = create_engine(DATABASE_URL, pool_pre_ping=True)

Branch resets terminate server-side connections. Without pool_pre_ping, SQLAlchemy may reuse dead pooled connections, causing SSL errors.

pytest-xdist note: Each worker has its own isolated branch. Resets only affect that worker's branch, so workers don't interfere with each other.

Performance: ~1.5s initial setup per session per worker + ~0.5s reset per test. For 10 write tests, expect ~6.5s total overhead.

`NeonBranch` dataclass

All fixtures return a NeonBranch dataclass with:

branch_id: The Neon branch ID
project_id: The Neon project ID
connection_string: Full PostgreSQL connection URI
host: The database host
parent_id: The parent branch ID (used for resets)
endpoint_id: The endpoint ID (for cleanup)

`neon_branch_readwrite` (deprecated)

Deprecated: Use neon_branch_isolated instead.

This fixture is an alias for neon_branch_isolated and will emit a deprecation warning.

`neon_branch` (deprecated)

Deprecated: Use neon_branch_isolated, neon_branch_dirty, or neon_branch_readonly instead.

This fixture is an alias for neon_branch_isolated and will emit a deprecation warning.

`neon_branch_shared` (module-scoped, no isolation)

Creates one branch per test module and shares it across all tests without resetting. This is the fastest option but tests can see each other's data modifications.

def test_read_only_query(neon_branch_shared):
    # Fast: no reset between tests
    # Warning: data from other tests in this module may be visible
    conn = psycopg.connect(neon_branch_shared.connection_string)

Use this when:

Tests are read-only
Tests don't interfere with each other
You manually clean up test data
Maximum speed is more important than isolation

Performance: ~1.5s initial setup per module, no per-test overhead.

`neon_connection_psycopg` (psycopg v3)

Convenience fixture providing a psycopg v3 connection with automatic rollback and cleanup.

Requires: pip install pytest-neon[psycopg]

def test_insert(neon_connection_psycopg):
    with neon_connection_psycopg.cursor() as cur:
        cur.execute("INSERT INTO users (name) VALUES (%s)", ("test",))
    neon_connection_psycopg.commit()

    with neon_connection_psycopg.cursor() as cur:
        cur.execute("SELECT name FROM users")
        assert cur.fetchone()[0] == "test"

`neon_connection` (psycopg2)

Convenience fixture providing a psycopg2 connection with automatic rollback and cleanup.

Requires: pip install pytest-neon[psycopg2]

def test_insert(neon_connection):
    cur = neon_connection.cursor()
    cur.execute("INSERT INTO users (name) VALUES (%s)", ("test",))
    neon_connection.commit()

`neon_engine` (SQLAlchemy)

Convenience fixture providing a SQLAlchemy engine with automatic disposal.

Requires: pip install pytest-neon[sqlalchemy]

from sqlalchemy import text

def test_query(neon_engine):
    with neon_engine.connect() as conn:
        result = conn.execute(text("SELECT 1"))
        assert result.scalar() == 1

Using Your Own SQLAlchemy Engine

If you have a module-level SQLAlchemy engine (common pattern) and use neon_branch_isolated, you must use pool_pre_ping=True:

# database.py
from sqlalchemy import create_engine
from config import DATABASE_URL

# pool_pre_ping=True is REQUIRED when using neon_branch_isolated
# It verifies connections are alive before using them
engine = create_engine(DATABASE_URL, pool_pre_ping=True)

Why? After each test, neon_branch_isolated resets the branch which terminates server-side connections. Without pool_pre_ping, SQLAlchemy may try to reuse a dead pooled connection, causing SSL connection has been closed unexpectedly errors.

Note: If you only use neon_branch_readonly or neon_branch_dirty, pool_pre_ping is not required since no resets occur.

This is also a best practice for any cloud database (Neon, RDS, etc.) where connections can be terminated externally.

Migrations

pytest-neon supports running migrations once before tests, with all test resets preserving the migrated state.

How It Works

The plugin always creates a migration branch from your configured parent:

Parent Branch (your configured parent)
    └── Migration Branch (session-scoped)
            │   ↑ migrations run here ONCE
            │
            ├── Read-only Endpoint (for neon_branch_readonly)
            ├── Dirty Branch (for neon_branch_dirty)
            └── Isolated Branches (for neon_branch_isolated)

This means:

Migrations run once per test session (not per test or per module)
Each test reset restores to the post-migration state
Tests always see your migrated schema

Setup

Override the neon_apply_migrations fixture in your conftest.py:

Alembic:

# conftest.py
import subprocess
import pytest

@pytest.fixture(scope="session")
def neon_apply_migrations(_neon_migration_branch):
    """Run Alembic migrations before tests."""
    # DATABASE_URL is already set to the migration branch
    subprocess.run(["alembic", "upgrade", "head"], check=True)

Django:

# conftest.py
import pytest

@pytest.fixture(scope="session")
def neon_apply_migrations(_neon_migration_branch):
    """Run Django migrations before tests."""
    from django.core.management import call_command
    call_command("migrate", "--noinput")

Raw SQL:

# conftest.py
import pytest

@pytest.fixture(scope="session")
def neon_apply_migrations(_neon_migration_branch):
    """Apply schema from SQL file."""
    import psycopg
    with psycopg.connect(_neon_migration_branch.connection_string) as conn:
        with open("schema.sql") as f:
            conn.execute(f.read())
        conn.commit()

Custom migration tool:

# conftest.py
import pytest

@pytest.fixture(scope="session")
def neon_apply_migrations(_neon_migration_branch):
    """Run custom migrations."""
    from myapp.migrations import run_migrations
    run_migrations(_neon_migration_branch.connection_string)

Important Notes

The _neon_migration_branch parameter gives you access to the NeonBranch object with connection_string, branch_id, etc.
DATABASE_URL (or your configured env var) is automatically set when the fixture runs
If you don't override neon_apply_migrations, no migrations run (the fixture is a no-op by default)
Migrations run before any test branches are created, so all tests see the same migrated schema

Configuration

Environment Variables

Variable	Description	Required
`NEON_API_KEY`	Your Neon API key	Yes
`NEON_PROJECT_ID`	Your Neon project ID	Yes
`NEON_PARENT_BRANCH_ID`	Parent branch to create test branches from	No
`NEON_DATABASE`	Database name (default: `neondb`)	No
`NEON_ROLE`	Database role (default: `neondb_owner`)	No

Command Line Options

Option	Description	Default
`--neon-api-key`	Neon API key	`NEON_API_KEY` env
`--neon-project-id`	Neon project ID	`NEON_PROJECT_ID` env
`--neon-parent-branch`	Parent branch ID	Project default
`--neon-database`	Database name	`neondb`
`--neon-role`	Database role	`neondb_owner`
`--neon-keep-branches`	Don't delete branches after tests	`false`
`--neon-branch-expiry`	Branch auto-expiry in seconds	`600` (10 min)
`--neon-env-var`	Environment variable for connection string	`DATABASE_URL`

Examples:

# Keep branches for debugging
pytest --neon-keep-branches

# Disable auto-expiry
pytest --neon-branch-expiry=0

# Use a different env var
pytest --neon-env-var=TEST_DATABASE_URL

pyproject.toml / pytest.ini

You can also configure options in your pyproject.toml:

[tool.pytest.ini_options]
neon_database = "mydb"
neon_role = "myrole"
neon_keep_branches = true
neon_branch_expiry = "300"

Or in pytest.ini:

[pytest]
neon_database = mydb
neon_role = myrole
neon_keep_branches = true
neon_branch_expiry = 300

Priority order: CLI options > environment variables > ini settings > defaults

CI/CD Integration

GitHub Actions

name: Tests

on: [push, pull_request]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.12'

      - name: Install dependencies
        run: pip install -e .[psycopg,dev]

      - name: Run tests
        env:
          NEON_API_KEY: ${{ secrets.NEON_API_KEY }}
          NEON_PROJECT_ID: ${{ secrets.NEON_PROJECT_ID }}
        run: pytest

How It Works

At the start of the test session, the plugin creates a migration branch from your parent branch
If defined, migrations run once on the migration branch
Test branches (dirty, isolated) are created as children of the migration branch
DATABASE_URL is set to point to the appropriate branch for each fixture
For neon_branch_isolated, the branch is reset after each test (~0.5s)
After all tests complete, branches are deleted
As a safety net, branches auto-expire after 10 minutes even if cleanup fails

Branches use copy-on-write storage, so you only pay for data that differs from the parent branch.

What Reset Does

The neon_branch_isolated fixture uses Neon's branch restore API to reset database state after each test:

Data changes are reverted: All INSERT, UPDATE, DELETE operations are undone
Schema changes are reverted: CREATE TABLE, ALTER TABLE, DROP TABLE, etc. are undone
Sequences are reset: Auto-increment counters return to parent state
Complete rollback: The branch is restored to the exact state of the parent at the time the child branch was created

This is similar to database transactions but at the branch level.

Branch Naming

Branches are automatically named to help identify their source:

pytest-[git-branch]-[random]-[suffix]

Examples:

pytest-main-a1b2-migration - Migration branch from main
pytest-feature-auth-c3d4-dirty - Dirty branch from feature/auth
pytest-main-a1b2-isolated-gw0 - Isolated branch for xdist worker 0
pytest-a1b2-migration - When not in a git repo

The git branch name is sanitized (only a-z, 0-9, -, _ allowed) and truncated to 15 characters. This makes it easy to identify orphaned branches in the Neon console.

Parallel Test Execution (pytest-xdist)

This plugin supports parallel test execution with pytest-xdist.

# Run tests in parallel with 4 workers
pip install pytest-xdist
pytest -n 4

How it works:

Fixture	xdist Behavior
`neon_branch_readonly`	Shared across ALL workers (read-only endpoint)
`neon_branch_dirty`	Shared across ALL workers (concurrent writes possible)
`neon_branch_isolated`	One branch per worker (e.g., `-isolated-gw0`, `-isolated-gw1`)

Key points:

The migration branch is created once and shared across all workers
neon_branch_readonly and neon_branch_dirty share resources across workers
neon_branch_isolated creates one branch per worker for isolation
Workers run tests in parallel without database state interference (when using isolated)
All branches are cleaned up after the test session

Cost implications:

Running with -n 4 creates: 1 migration branch + 1 dirty branch + 4 isolated branches (if all workers use isolated)
Choose your parallelism level based on your Neon plan's branch limits
Each worker's isolated branch is reset after each test using the fast reset operation (~0.5s)

Troubleshooting

"psycopg not installed" or "psycopg2 not installed"

The convenience fixtures require their respective drivers. Install the appropriate extra:

# For neon_connection_psycopg fixture
pip install pytest-neon[psycopg]

# For neon_connection fixture
pip install pytest-neon[psycopg2]

# For neon_engine fixture
pip install pytest-neon[sqlalchemy]

Or use the core fixtures with your own driver:

def test_example(neon_branch_isolated):
    import my_preferred_driver
    conn = my_preferred_driver.connect(neon_branch_isolated.connection_string)

"Neon API key not configured"

Set the NEON_API_KEY environment variable or use the --neon-api-key CLI option.

"Neon project ID not configured"

Set the NEON_PROJECT_ID environment variable or use the --neon-project-id CLI option.

"SSL connection has been closed unexpectedly" (SQLAlchemy)

This happens when SQLAlchemy tries to reuse a pooled connection after a branch reset. The reset terminates server-side connections, but SQLAlchemy's pool doesn't know.

Fix: Add pool_pre_ping=True to your engine:

engine = create_engine(DATABASE_URL, pool_pre_ping=True)

This makes SQLAlchemy verify connections before using them, automatically discarding stale ones.

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 70 Commits
.config		.config
.github/workflows		.github/workflows
src/pytest_neon		src/pytest_neon
tests		tests
.env.example		.env.example
.gitignore		.gitignore
.neon		.neon
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml
uv.lock		uv.lock

License

ZainRizvi/pytest-neon

Folders and files

Latest commit

History

Repository files navigation

pytest-neon

Features

Installation

Quick Start

Fixtures

Architecture

neon_branch_readonly (session-scoped, enforced read-only)

neon_branch_dirty (session-scoped, shared state)

neon_branch_isolated (function-scoped, full isolation)

NeonBranch dataclass

neon_branch_readwrite (deprecated)

neon_branch (deprecated)

neon_branch_shared (module-scoped, no isolation)

neon_connection_psycopg (psycopg v3)

neon_connection (psycopg2)

neon_engine (SQLAlchemy)

Using Your Own SQLAlchemy Engine

Migrations

How It Works

Setup

Important Notes

Configuration

Environment Variables

Command Line Options

pyproject.toml / pytest.ini

CI/CD Integration

GitHub Actions

How It Works

What Reset Does

Branch Naming

Parallel Test Execution (pytest-xdist)

Troubleshooting

"psycopg not installed" or "psycopg2 not installed"

"Neon API key not configured"

"Neon project ID not configured"

"SSL connection has been closed unexpectedly" (SQLAlchemy)

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 16

Packages 0

Contributors 2

Uh oh!

Languages

`neon_branch_readonly` (session-scoped, enforced read-only)

`neon_branch_dirty` (session-scoped, shared state)

`neon_branch_isolated` (function-scoped, full isolation)

`NeonBranch` dataclass

`neon_branch_readwrite` (deprecated)

`neon_branch` (deprecated)

`neon_branch_shared` (module-scoped, no isolation)

`neon_connection_psycopg` (psycopg v3)

`neon_connection` (psycopg2)

`neon_engine` (SQLAlchemy)

Packages