unitofwork 0.3.0

No project description provided

unitofwork

A lightweight, database-agnostic implementation of the Unit of Work pattern for Python applications. Designed for clean architecture, type safety, and atomic transactions across mixed repository types.

Features

• Atomic Transactions: Ensure all operations succeed or fail together
• Mixed Repository Support: Works with SQL, in-memory, file-based, and custom repositories
• Type Safety: Full mypy support with generics and protocols
• Simple API: Intuitive context manager interface
• No Dependencies: Pure Python implementation
• Comprehensive Testing: 100% test coverage with extensive test suite
• Rollback Support: Automatic rollback for in-memory repositories
• Protocol-Based: Uses structural typing with SupportsRollback interface

Installation

$ pip install unitofwork

Quick Start

import copy
from dataclasses import dataclass
from uuid import UUID, uuid4

from unitofwork import UnitOfWork

# Your repositories must implement the SupportsRollback interface
class InMemoryUserRepository:
    def __init__(self):
        self._users: dict[UUID, User] = {}
        self._snapshots: list[dict[UUID, User]] = []

    def checkpoint(self) -> dict[UUID, User]:
        snapshot = copy.deepcopy(self._users)
        self._snapshots.append(snapshot)
        return snapshot

    def restore(self, snapshot: dict[UUID, User]) -> None:
        self._users = snapshot

    def commit(self) -> None:
        self._snapshots.clear()

    def add(self, user: User) -> None:
        self._users[user.id] = user

@dataclass
class User:
    id: UUID
    name: str
    email: str

# Create repository that follows SupportsRollback interface
user_repo = InMemoryUserRepository()

# Atomic transaction
with UnitOfWork(user_repo) as uow:
    user = User(uuid4(), "Alice", "[e-mail]")
    uow.register_operation(lambda: user_repo.add(user))

# Operation is committed automatically on successful exit

SupportsRollback interface

All repositories must implement three essential methods:

from typing import Any, Protocol

class SupportsRollback(Protocol):
    """Protocol that all repositories must follow to work with UnitOfWork"""
    
    def checkpoint(self) -> Any:
        """Return a snapshot of the current state for potential rollback"""
        ...
    
    def restore(self, snapshot: Any) -> None:
        """Restore state from a previously taken snapshot"""
        ...
    
    def commit(self) -> None:
        """Finalize the transaction after successful operations"""
        ...

Implementing the interface

In-memory repository example

For in-memory repo we simply make a deep copy of the repo content before the transaction (checkpoint). On failure we can restore using this deep copy of the repo content as shown below:

class InMemoryRepository:
    def __init__(self):
        self._data = {}
        self._snapshots = []
    
    def checkpoint(self) -> dict:
        return copy.deepcopy(self._data)
    
    def restore(self, snapshot: dict) -> None:
        self._data = snapshot
    
    def commit(self) -> None:
        self._snapshots.clear()

SQL repository example

In case of an SQL repository, we can use SAVEPOINT to save the transaction checkpoint. It allows us to ROLLBACK to SAVEPOINT if the transaction fails.

from sqlalchemy.engine import Connection

# Example: SQLAlchemy repository
class SQLUserRepository:
    def __init__(self, connection: Connection):
        self.conn = connection
        self._savepoint = None
    
    def checkpoint(self) -> str:
        """Create a database savepoint"""
        if not self.conn.in_transaction():
            self.conn.begin()
        self._savepoint = f'savepoint_{id(self)}'
        statement = sqlalchemy.text(f'SAVEPOINT {self._savepoint}')
        self.conn.execute(statement)
        return self._savepoint

    def restore(self, savepoint: str) -> None:
        """Rollback to savepoint"""
        if not self._savepoint or savepoint != self._savepoint:
            return

        try:
            statement = sqlalchemy.text(f'ROLLBACK TO SAVEPOINT {savepoint}')
            self.conn.execute(statement)
        except sqlalchemy.exc.OperationalError as e:
            if 'no such savepoint' in str(e).lower():
                # Savepoint was already released (maybe automatically by SQLite)
                # This is OK - means work was already committed
                pass
            else:
                raise

    def commit(self) -> None:
        """Release savepoint"""
        if self._savepoint:
            try:
                self.conn.execute(
                    sqlalchemy.text(f'RELEASE SAVEPOINT {self._savepoint}')
                )
            except Exception as err:
                print(f'Error releasing savepoint: {err}')
            self._savepoint = None

Why Unit of Work?

The Unit of Work pattern maintains a list of objects affected by a business transaction and coordinates the writing out of changes and the resolution of concurrency problems.

Without Unit of Work

# Risk: Partial failures
user_repo.add(user)        # Success
product_repo.add(product)  # Failure - database error
# Now user exists but product doesn't - inconsistent state, not OK!

With Unit of Work

# Safe: Atomic operations
with UnitOfWork(user_repo, product_repo) as uow:
    uow.register_operation(lambda: user_repo.add(user))
    uow.register_operation(lambda: product_repo.add(product))
# Both succeed or both fail - guaranteed consistency, now it's OK!

Usage Guide

Basic Usage

from unitofwork import UnitOfWork

# Create repository with custom ID field
class Product:
    def __init__(self, sku: str, name: str, price: float):
        self.sku = sku
        self.name = name
        self.price = price

product_repo = InMemoryRepository[str, Product](id_field="sku")

# Simple transaction
with UnitOfWork(product_repo) as uow:
    uow.register_operation(
        lambda: product_repo.add(
            Product("laptop-123", "Premium Laptop", 1299.99),
        ),
    )

Mixed Repository Types

Provided you have multiple repositories implementing SupportsRollback interface, the same UnitOfWork pattern can be applied to all of them.

from unitofwork import SqlUnitOfWork, UnitOfWork
from your_app.repositories import SQLUserRepository, FileLogRepository

# Mix different repository types
sql_user_repo = SQLUserRepository(connection)
in_memory_cache = InMemoryRepository[str, CachedData](id_field="key")
file_log_repo = FileLogRepository("/path/to/logs")

with SqlUnitOfWork(  # use for SQL-involved operations
    UnitOfWork(sql_user_repo, in_memory_cache, file_log_repo),
    connection,
) as uow:
    uow.register_operation(lambda: sql_user_repo.add_user(new_user))
    uow.register_operation(lambda: in_memory_cache.add(cached_data))
    uow.register_operation(lambda: file_log_repo.log_operation("user_created"))

Custom Repositories

from typing import Dict, Any
from unitofwork import SupportsRollback

class CustomRepository(SupportsRollback[str, str]):
    def __init__(self):
        self._data: Dict[str, str] = {}
    
    def checkpoint(self) -> Dict[str, str]:
        return self._data.copy()
    
    def restore(self, snapshot: Dict[str, str]) -> None:
        self._data = snapshot
    
    def add(self, key: str, value: str) -> None:
        self._data[key] = value
    
    def get(self, key: str) -> str:
        return self._data[key]

# Use your custom repository
custom_repo = CustomRepository()
with UnitOfWork(custom_repo) as uow:
    uow.register_operation(lambda: custom_repo.add("test_key", "test_value"))

Error handling

try:
    with UnitOfWork(user_repo, order_repo) as uow:
        uow.register_operation(lambda: user_repo.add(user))
        uow.register_operation(lambda: order_repo.add(order))
        
        # Simulate business rule violation
        if not user.can_purchase():
            raise ValueError("User cannot make purchase")
            
except ValueError as e:
    print(f"Transaction failed: {e}")
    # Both user_repo and order_repo are automatically rolled back!

Architecture

Core Components

• UnitOfWork: Main coordinator class managing transactions
• SupportsRollback: Protocol defining repository interface
• UnitOfWorkError: Base exception for UnitOfWork errors
• RollbackError: Exception raised when rollback fails partially

Design Principles

• Database Agnostic: Works with any persistence mechanism
• Type Safe: Full static type checking support
• Minimal API: Simple, intuitive interface
• Extensible: Easy to adapt existing repositories
• Thread Safe: Designed for concurrent usage

Key Principle: Structural Typing

Your repositories do not need to inherit from SupportsRollback --- they just need to implement:

checkpoint() -> Any
restore(snapshot: Any) -> None
commit() -> None

This enables seamless integration with any persistence mechanism, provided it allows for creating a checkpoint and makes it technically possible to revert to this checkpoint.

Advanced Usage

Manual Rollback

with UnitOfWork(user_repo, product_repo) as uow:
    # These will commit if no exception
    uow.register_operation(lambda: user_repo.add(user))
    uow.register_operation(lambda: product_repo.add(product))
    
    # Manual rollback if needed
    if some_condition:
        uow.rollback()
        # Additional cleanup...

Acknowledgements

• Inspired by Domain-Driven Design patterns
• Based on concepts from "Architecture Patterns with Python"
• Built with type safety and reliability as first-class citizens