Port and adapters architecture. Python + Nameko microexample.

Introduction

Port and adapters (or hexagonal) architecture is a software design concept introduced by Alistair Cockburn in 2005. The main goal of it is to provide a clear seperation between application logic and external dependencies like database, user interface, framework providing HTTP requests, etc.

Application core, that is agonostic about external services and dependencies, should provide orchestration of whole business process exploiting existing ports. There are two types of ports:

primary port, that is an entry exposing application core to outside world. It is usually a fascade called by a primary adapter (e.g. REST API, CLI, etc.),

secondary port, enables application core to communicate with external world (e.g. database, mail sender, etc). It is an interface that is imlemented by a secondary adapter.

Implementation

We define a secondary port as an interface that enables publishing domain events from application core. Depending on implementation it could publish messages to local event bus or genuine event broker like RabbitMQ (but implementation is a resposibility of an adapter).

from abc import ABC, abstractmethod

class EventPublisher(ABC):
    @abstractmethod
    def publish(self, events):
        pass

To implement secondary adapter for EventPublisher we make a wrapper for Nameko EventDispatcher class.

# domain/events.py
from datetime import datetime
from typing import Dict
from dataclasses import dataclass

@dataclass(frozen=True)
class DomainEvent:
    @property
    def as_dict(self) -> Dict:
        serialized = asdict(self)
        serialized["name"] = self.name
        return serialized

@dataclass(frozen=True)
class RegisteredEntryEvent(DomainEvent):
      email: str 
      name: str = 'RegisteredEntryEvent'
      timestamp: datetime = datetime.now()

# infrastructure/event_publisher.py
from typing import List
from nameko.events import EventDispatcher
from application.event_publisher import EventPublisher
from domain.event import DomainEvent

class NamekoEventPublisher(EventPublisher):
    def __init__(self, dispatcher: EventDispatcher):
        self._dispatcher = dispatcher

    def publish(self, events: List[DomainEvent]) -> None:
        for event in events:
            self._dispatcher(event.name, event.as_dict)

Primary port in our example is a simple application service defining registration use case. EventPublisher interface (not implementation) is injected into the service.

# application/service.py
from application.event_publisher import EventPublisher
from domain.event import RegisteredEntryEvent

class RegistrationService:
    def __init__(self, event_publisher: EventPublisher):
        self._event_publisher = event_publisher

    def register_entry(self, email: str) -> None:
        event = RegisteredEntryEvent(email=email)
        self._event_publisher.publish(event)

Nameko HTTP endpoint is a primary adapter for our RegistrationEntryService port. It handles HTTP request data and execute the service method.

import json
from nameko.events import EventDispatcher
from nameko.web.handlers import http
from werkzeug.wrappers import Request, Response

class NamekoRegistrationService:
    name = "nameko_registration_service"
    dispatcher = EventDispatcher()

    @http("POST", "/register")
    def register_entry(self, request: Request) -> Response:
        request_params = json.loads(request.data)
        email = request_params["email"]
        service = RegistrationService(
       event_dispatcher=NamekoEventPublisher(self.dispatcher),
        )
        service.register_entry(email)
        return Response(f"Registered entry for {email=}")

The crucial thing here is that our application core does not have any knowledge about infrastructure and API, it operates only on interfaces. If we would like to have a database access, we defined a repository interface and injected it into our service. Or, if there is a need to send a notification, we probably make a notification sender interface that can be implemented by SMS or Email sender adapter in the infrastructure layer. But inside the application core we know nothing about infrastructure implementations (thanks to it the application logic can be easily unit tested). And this is the main gain of using this architecture.

Summary

It was a microexample of hexagonal architecture base concepts using Python and Nameko. If you find it interesting, I recommend you to visit my github repository for extended implementation (including also a domain layer that was omitted here, for the sake of simplicity) of similar project (also Python, Nameko and Port and Adapters): https://github.com/jorzel/opentable. For more theoretical background about hexagonal architecture, go here.