Skip to content

Latest commit

 

History

History
291 lines (197 loc) · 13.1 KB

README.md

File metadata and controls

291 lines (197 loc) · 13.1 KB
Raw

Logo NCafe

Minimal .NET microservices implementation in the context of a cafe.

Heavily inspired on the microcafe project by Richard Banks.

More information about the application domain, please check the links below:

https://www.slideshare.net/rbanks54/architecting-microservices-in-net https://www.enterpriseintegrationpatterns.com/ramblings/18_starbucks.html

Tech stack:

  • C# 12/.NET 8
  • ASP.NET Core Minimal APIs
  • .NET Aspire
  • Docker
  • Kubernetes
  • EventStore: Database for Event Sourcing where we store events as the source of truth instead of current state
  • RabbitMQ: Message broker used for asynchronous messaging

Warning

This code should be treated as sample code. It's a sandbox for practicing microservice ideas, CQRS, Event Sourcing, Kubernetes/helm deployment and related things.

I wrote some documentation below for those who are starting on this journey. I hope it helps! 😄

TODO

  • Use SignalR to update the Barista page in real time as new orders are placed
  • Kubernetes/helm/helmfile deployment to my internal K3S cluster
  • Add observability with OpenTelemetry (Thanks to .NET Aspire)
  • Cashier should be able to add the customer name to the order
  • Cashier should be able to add multiple products to the same order
  • Barista should be able to see the name of the customer and products/quantities
  • Add timestamp to orders so barista sees in correct order
  • Cashier should be able to remove products from the order
  • Admin should be able to delete products
  • Cashier should be able to cancel order
  • Make sure we always publish OrderPlaceMessage when Cashier places an order (Outbox?)
  • Use a database for the read models (projections)
  • Move projection services to their own microservice so reads and writes are separate
  • Admin should be able to edit products

Content

NCafe Solution

The NCafe Solution is based on Clean Architecture, so dependencies only point torwards the center.

Clean Architecture

Core (Shared Abstractions)

All common interfaces and abstractions are here. For example:

  • AggregateRoot: Abstract base class for our domain entities
  • IEvent and Event: Base for events that represent domain entity state changes
  • IRepository: Contract used to fetch and save domain entities
  • IBusPublisher: Contract used for publishing integration events
  • MessageBus Events: Integration events used to communicate between microservices
  • IProjectionService: Contract used for building projections based on domain entity events
  • Read Model: Contract used to fetch and save read models (projections)
  • Basic exceptions

So, basically, this project is the core of our solution and will only contain the abstractions used by other layers. This project doesn't have any microservice specific code.

Shared

There is a shared project (NCafe.Shared), that contains code that don't require any core abstraction or logic. For example, types that define SignalR objects for real-time functionality.

The Web UI project doesn't really need a reference to NCafe.Core (NCafe.Shared is enough).

Application Domain

Depends only on the Core project in order to define domain entities (aggregates), events, commands, queries, their respective handlers and business logic. So it might have:

  • Entities: Their current state is defined by a stream of events stored in EventStore. Provide methods to make it do something (ex.: CompletePreparation). These methods will raise events (after doing some validation if necessary), that will be appended to its event stream in EventStore
  • Events: Represent something that has happened in the domain (ex.: OrderPlaced). These events will be raised by the entities as described above
  • Commands/Handlers: Commands and handlers will perform some kind of action. Most probably changing the state of an entity. Must use abstractions/interfaces (ex.: IRepository) instead of implementations
  • Queries/Handlers: Used to return data from read models. Must use abstractions/interfaces (ex.: IReadModelRepository) instead of implementations
  • Read Models: To be used by projection services and query handlers

Web API

The entry points (runners) of our microservices. They simply register the required dependencies using methods from the Infrastructure project and map endpoints, which use MediatR to invoke a handler (see Program.cs).

These projects have a reference to its Domain, which should actually be called Application, to conform with Clean Architecture since it contains use cases. The APIs also have a reference to the Infrastructure project.

Projection services can also be in the Web API project (Find more about projections below).

In case the microservice needs to consume integration events, a Consumer service can be created (see OrdersConsumerService in Barista.Api). Basically, this service implements .NET's IHostedService, subscribes to a RabbitMQ stream specifying a queue, a topic and a callback, which in case will use MediatR to invoke a domain command.

Web UI

The Web UI project is a Blazor WebAssembly project that interacts with the Admin, Barista and Cashier microservices. It stores the base address of each microservice in wwwroot/appsettings.json.

It should work out of the box locally when running with .NET Aspire and docker compose, due to the port configuration in services-compose.yaml and applicationUrl (inside each microservice's launchSettings.json).

For production, the base address should be set in wwwroot/appsettings.json. Since the Web UI project is a static site, I decided to deploy it inside a container running nginx. In the Dockerfile, I added a step to copy prepare-appsettings.sh to the /docker-entrypoint.d/ folder. When nginx starts, it will run scripts inside this folder. The prepare-appsettings.sh script will replace the base addresses in wwwroot/appsettings.json with values from environment variables (ADMIN_BASE_ADDRESS, CASHIER_BASE_ADDRESS and BARISTA_BASE_ADDRESS).

For my deployment, since I'm using a local Kubernetes cluster and Nginx Proxy Manager, I hardcoded the base addresses in the appspec.yaml.

Infrastructure

Implementations for all external dependencies are defined in this project. Like:

  • EventStore Repository and Projection Service
  • RabbitMQ publisher

There are some other implementations in here as well, like a Logging decorator and read model repositories (only in-memory for now).

This project also contains methods for registering all the implementations for interfaces defined in the Core project.

All the implementation types should be marked as internal, because our Application/Domain should only depend on abstractions. The Application/Domain has no idea about the actual implementations. This is the Dependency Inversion Principle in action.

NCafe's CQRS + Event Sourcing implementation

CQRS and Event Sourcing in NCafe

Inspired by https://codeopinion.com/cqrs-event-sourcing-code-walk-through/.

Command

Depending on the command, the handler will instantiate the domain model (entity/aggregate), for example in PlaceOrderHandler. Then the entity will be saved using IRepository.

For other commands, the handler will first fetch the entity by id from the repository, then tell it to do something (ex.: CompletePreparation) and then save it using the repository.

When fetching an entity, the repository will actually get all events for the specific aggregate and apply them in order. This will re-build the current state of the entity based on the events.

The Save method in the repository sends pending events to EventStore, for example, the OrderPrepared event after CompletePreparation is called (see the Barista.Domain project).

Query

This is the simplest part of the system. The handlers will use a read model repository to return one or more items from the query database (in-memory for now). The data from this database comes from a projection service, described below.

Projections

All Projection services subscribed to EventStore event streams (ex.: baristaOrder), will receive new events from EventStore and use it to update the query database (read model) if necessary.

In NCafe, Projection services are implemented using .NET's BackgroundService running in the API projects, as the read models are being stored using an in-memory repository for the time being. But when switch to a proper database, we can easily move the projection specific code to it's own microservice. In fact, I think we will have to move the code, otherwise we need a way to run only one projection service in case there's a need to scale out (create more instances of) the API microservice. In case you're wondering why, it's because I'm imagining multiple projection services subscribed to the same streams and also using the same database. So they would all try to do the same inserts/updates. It looks like trouble.

Also, when we move to a database, we'll need to save checkpoints, so when we restart the projection service we can tell EventStore that we only care about events from a specific position in the stream. Saving the version should be enough.

Run with .NET Aspire

.NET Aspire is an opinionated, cloud ready stack for building observable, production ready, distributed applications.

Make sure to set NCafe.AppHost as Startup Project in the IDE and run. If you prefer CLI, just run the NCafe.AppHost project. If you run through the CLI, you will see a link to the dashboard.

image

Run with docker

In order to run the solution, you need the following:

  • .NET 8 SDK
  • Docker

Starting the infrastructure containers

Run the following command (repo's root folder):

docker compose -f infrastructure-compose.yaml up -d

Starting the microservices

You can start the microservices via the dotnet CLI or your favorite IDE/code editor. If you're using Visual Studio you can also set multiple startup projects by going to solution properties.

If you prefer docker, run the build script in the root folder:

chmod +x build.sh
./build.sh

If you're on Windows:

.\Build.ps1

Run the following command to build and start all microservices containers:

docker compose -f services-compose.yaml up -d

Starting the Web UI in Visual Studio

Sometimes, when working on the Web UI project, it might be worth to run the infrastructure and other services using docker compose and then run the Web UI project in Visual Studio. This way can be easier to debug.

Run this command to start all services except web-ui:

docker compose -f services-compose.yaml up -d --scale web-ui=0

Swagger

NCafe in action

  1. Start the NCafe.Web project (Blazor WebAssembly)
  2. Create a new product in the Admin page
  3. Place an order in the Cashier page
  4. Complete the order in the Barista page

ncafe

Or, you can use Swagger for example:

  1. Create a product via the POST /products endpoint in Admin
  2. Get a product Id using the GET /products endpoint
  3. Create an order via the POST /orders endpoint in Cashier, it will return the order Id
  4. Add item(s) to the order via the POST /orders/add-item in Cashier
  5. Place the order via the POST /orders/place endpoint in Cashier
  6. Complete the order via the POST /orders/prepared endpoint in Barista

EventStore

You can check all the events in EventStore by going to the Stream Browser in http://localhost:2113/.

EventStore Screenshot

RabbitMQ

You can check the message queue in http://localhost:15672/.

Stopping everything

Run the following command:

docker compose -f services-compose.yaml -f infrastructure-compose.yaml down