SlimMessageBus is a client façade for message brokers for .NET. It comes with implementations for specific brokers (Apache Kafka, Azure EventHub, MQTT/Mosquitto, Redis Pub/Sub) and in-memory message passing (in-process communication). SlimMessageBus additionally provides request-response implementation over message queues.
- Key elements of SlimMessageBus
- Docs
- Packages
- Samples
- Features
- Principles
- License
- Build
- Testing
- Credits
- Consumers:
IConsumer<in TMessage>
- subscriber in pub/sub (or queue consumer)IRequestHandler<in TRequest, TResponse>
- request handler in request-response
- Producers:
IPublishBus
- publisher in pub/sub (or queue producer)IRequestResponseBus
- sender in req/respIMessageBus
- extendsIPublishBus
andIRequestResponseBus
- Misc:
IRequestMessage<TResponse>
- marker for request messagesMessageBus
- static accessor for current contextIMessageBus
- Introduction
- Providers:
- Serialization Plugins
Typically your application components (business logic, domain) only need to depend on SlimMessageBus
which is the facade, and ultimately your application hosting layer (ASP.NET, Windows Service, Console App) will reference and configure the other packages (SlimMessageBus.Host.*
) which are the providers and plugins.
Check out the Samples folder for complete overview.
Some service (or domain layer) sends a message:
IMessageBus bus; // injected
await bus.Publish(new SomeMessage())
Another service (or application layer) handles the message:
public class SomeMessageConsumer : IConsumer<SomeMessage>
{
public async Task OnHandle(SomeMessage message, string path) // path = topic or queue name
{
// handle the message
}
}
Note: It is also possible to avoid having to implement the interface
IConsumer<T>
(see here).
The bus also supports request-response implemented via queues (or topics - depending on what the chosen transport provider supports). The sender side sends a request message:
var messageResponse = await bus.Send(new MessageRequest());
Note: It is possible to configure the bus to timeout a request when the response does not arrive within the allotted time (see here).
The receiving side handles the request and replies back:
public class MessageRequestHandler : IRequestHandler<MessageRequest, MessageResponse>
{
public async Task<MessageResponse> OnHandle(MessageRequest request, string path)
{
// handle the request message and return response
}
}
The bus will ask the chosen DI to provide the consumer instances (SomeMessageConsumer
, MessageRequestHandler
).
To configure SMB in your service:
var mbb = MessageBusBuilder.Create();
mbb
// Pub/Sub example:
.Produce<SomeMessage>(x => x.DefaultTopic("some-topic"))
.Consume<SomeMessage>(x => x
.Topic("some-topic")
.WithConsumer<SomeMessageConsumer>()
//.KafkaGroup("some-kafka-consumer-group") // Kafka provider specific
//.SubscriptionName("some-azure-sb-topic-subscription") // Azure ServiceBus provider specific
)
// Use JSON for message serialization
.WithSerializer(new JsonMessageSerializer())
// Use DI from Ms Dependency Injection (or ASP.NET Core, Autofac, Unity, ServiceLocator)
.WithDependencyResolver(new MsDependencyInjectionDependencyResolver(serviceProvider))
//.WithDependencyResolver(new AspNetCoreMessageBusDependencyResolver(serviceProvider))
// Use Apache Kafka transport provider
.WithProviderKafka(new KafkaMessageBusSettings("localhost:9092"));
// Use Azure Service Bus transport provider
//.WithProviderServiceBus(...)
// Use Azure Azure Event Hub transport provider
//.WithProviderEventHub(...)
// Use Redis transport provider
//.WithProviderRedis(...)
// Use in-memory transport provider
//.WithProviderMemory(...)
// Build the bus from the builder. Message consumers will start consuming messages from the configured topics/queues of the chosen provider.
IMessageBus bus = mbb.Build();
// Register bus in your DI (as a singleton)
When your service uses Microsoft.Extensions.DependencyInjection
, the SMB can be configured in a more compact way
(requires SlimMessageBus.Host.MsDependencyInjection
or SlimMessageBus.Host.AspNetCore
package):
// Startup.cs
IServiceCollection services;
services.AddSlimMessageBus((mbb, svp) =>
{
mbb
.Produce<SomeMessage>(x => x.DefaultTopic("some-topic"))
.Consume<SomeMessage>(x => x
.Topic("some-topic")
.WithConsumer<SomeMessageConsumer>()
//.KafkaGroup("some-kafka-consumer-group") // Kafka provider specific
//.SubscriptionName("some-azure-sb-topic-subscription") // Azure ServiceBus provider specific
)
// ...
.WithSerializer(new JsonMessageSerializer())
.WithProviderKafka(new KafkaMessageBusSettings("localhost:9092"));
},
// Option 1 (optional)
addConsumersFromAssembly: new[] { Assembly.GetExecutingAssembly() }, // auto discover consumers and register into DI (see next section)
addConfiguratorsFromAssembly: new[] { Assembly.GetExecutingAssembly() } // auto discover modular configuration and register into DI (see next section)
);
// Option 2 (optional)
services.AddMessageBusConsumersFromAssembly(Assembly.GetExecutingAssembly());
services.AddMessageBusConfiguratorsFromAssembly(Assembly.GetExecutingAssembly());
This example shows how SlimMessageBus
and SlimMessageBus.Host.Memory
can be used to implement Domain Events pattern. The provider passes messages in the same app domain process (no external message broker is required).
The domain event is a simple POCO:
// domain event
public class OrderSubmittedEvent
{
public Order Order { get; }
public DateTime Timestamp { get; }
public OrderSubmittedEvent(Order order) { ... }
}
The event handler implements the IConsumer<T>
interface:
// domain event handler
public class OrderSubmittedHandler : IConsumer<OrderSubmittedEvent>
{
public Task OnHandle(OrderSubmittedEvent e, string name)
{
Console.WriteLine("Customer {0} {1} just placed an order for:", e.Order.Customer.Firstname, e.Order.Customer.Lastname);
foreach (var orderLine in e.Order.Lines)
{
Console.WriteLine("- {0}x {1}", orderLine.Quantity, orderLine.ProductId);
}
Console.WriteLine("Generating a shipping order...");
return Task.Delay(1000);
}
}
The domain handler (well, really the consumer) is obtained from the dependency resolver at the time of event publication. It can be scoped (per web request, per unit of work) as configured in your favorite DI container.
Somewhere in your domain layer, the domain event gets raised:
// aggregate root
public class Order
{
public Customer Customer { get; }
public OrderState State { get; private set; }
private IList<OrderLine> lines = new List<OrderLine>();
public IEnumerable<OrderLine> Lines => lines.AsEnumerable();
public Order(Customer customer)
{
Customer = customer;
State = OrderState.New;
}
public OrderLine Add(string productId, int quantity) { }
public Task Submit()
{
State = OrderState.Submitted;
var e = new OrderSubmittedEvent(this);
return MessageBus.Current.Publish(e); // raise domain event
}
}
Some sample logic executed in your domain:
var john = new Customer("John", "Whick");
var order = new Order(john);
order.Add("id_machine_gun", 2);
order.Add("id_grenade", 4);
order.Submit(); // events fired here
Notice the static MessageBus.Current
property might actually be configured to resolve a scoped IMessageBus
instance (web request-scoped).
The SlimMessageBus
configuration for in-memory provider looks like this:
// Define the recipie how to create our IMessageBus
var mbb = MessageBusBuilder.Create()
.Produce<OrderSubmittedEvent>(x => x.DefaultTopic(x.MessageType.Name))
.Consume<OrderSubmittedEvent>(x => x.Topic(x.MessageType.Name).WithConsumer<OrderSubmittedHandler>())
.WithDependencyResolver(new MsDependencyInjectionDependencyResolver(services))
.WithProviderMemory(new MemoryMessageBusSettings
{
// Suppress serialization and pass the same event instance to subscribers (domain events contain domain objects we do not want to be serialized, we also gain on speed)
EnableMessageSerialization = false
});
// Create the IMessageBus instance from the builder
IMessageBus bus = mbb.Build();
// Set the provider to resolve our bus - this setup will work as a singleton.
MessageBus.SetProvider(() => bus);
See the complete sample for ASP.NET Core where the handler and bus are web-request scoped.
Use case:
- Some front-end web app needs to display downsized image (thumbnails) of large images to speed up the page load.
- The thumbnails are requested in the WebApi and are generated on demand (and cached to disk) by the Worker (unless they exist already).
- WebApi and Worker exchange messages via Apache Kafka
- Worker can be scaled out (more instances, more Kafka partitions)
The front-end web app makes a call to resize an image DSC0862.jpg
to 120x80
resolution, by using this URL:
https://localhost:56788/api/image/DSC3781.jpg/r/?w=120&h=80&mode=1
This gets handled by the WebApi method of the ImageController
private readonly IRequestResponseBus _bus;
// ...
[Route("{fileId}")]
public async Task<HttpResponseMessage> GetImageThumbnail(string fileId, ThumbnailMode mode, int w, int h)
{
var thumbFileContent = // ... try to load content for the desired thumbnail w/h/mode/fileId
if (thumbFileContent == null)
{
// Task will await until response comes back (or timeout happens). The HTTP request will be queued and IIS processing thread released.
var thumbGenResponse = await _bus.Send(new GenerateThumbnailRequest(fileId, mode, w, h));
thumbFileContent = await _fileStore.GetFile(thumbGenResponse.FileId);
}
return ServeStream(thumbFileContent);
}
The GenerateThumbnailRequest
request is handled by a handler in one of the pool of Worker console apps.
public class GenerateThumbnailRequestHandler : IRequestHandler<GenerateThumbnailRequest, GenerateThumbnailResponse>
{
public Task<GenerateThumbnailResponse> OnHandle(GenerateThumbnailRequest request, string name)
{
// some processing
return new GenerateThumbnailResponse { FileId = thumbnailFileId };
}
}
The response gets replied onto the originating WebApi instance and the Task resolves causing the queued HTTP request to serve the resized image thumbnail.
var thumbGenResponse = await _bus.Send(new GenerateThumbnailRequest(fileId, mode, w, h));
The message bus configuration for the WebApi:
private IMessageBus BuildMessageBus()
{
// unique id across instances of this application (e.g. 1, 2, 3)
var instanceId = Configuration["InstanceId"];
var kafkaBrokers = Configuration["Kafka:Brokers"];
var instanceGroup = $"webapi-{instanceId}";
var instanceReplyTo = $"webapi-{instanceId}-response";
var mbb = MessageBusBuilder.Create()
.Produce<GenerateThumbnailRequest>(x =>
{
//x.DefaultTimeout(TimeSpan.FromSeconds(10)); // Default response timeout for this request type
x.DefaultTopic("thumbnail-generation"); // Use this topic as default when topic is not specified in IMessageBus.Publish() for that message type
})
.ExpectRequestResponses(x =>
{
x.ReplyToTopic(instanceReplyTo); // Expect all responses to my reqests replied to this topic
x.KafkaGroup(instanceGroup); // Kafka consumer group
x.DefaultTimeout(TimeSpan.FromSeconds(30)); // Default global response timeout
})
.WithDependencyResolver(new MsDependencyInjectionDependencyResolver(serviceProvider))
.WithSerializer(new JsonMessageSerializer())
.WithProviderKafka(new KafkaMessageBusSettings(kafkaBrokers));
var messageBus = mbb.Build();
return messageBus;
}
The message bus configuration for the Worker:
private static IMessageBus BuildMessageBus(IServiceProvider serices)
{
// unique id across instances of this application (e.g. 1, 2, 3)
var instanceId = Configuration["InstanceId"];
var kafkaBrokers = Configuration["Kafka:Brokers"];
var instanceGroup = $"worker-{instanceId}";
var sharedGroup = "workers";
var mbb = MessageBusBuilder.Create()
.Handle<GenerateThumbnailRequest, GenerateThumbnailResponse>(s =>
{
s.Topic("thumbnail-generation", t =>
{
t.WithHandler<GenerateThumbnailRequestHandler>()
.KafkaGroup(sharedGroup) // kafka consumer group
.Instances(3);
});
})
.WithDependencyResolver(new MsDependencyInjectionDependencyResolver(services))
.WithSerializer(new JsonMessageSerializer())
.WithProviderKafka(new KafkaMessageBusSettings(kafkaBrokers));
var messageBus = mbb.Build();
return messageBus;
}
Because topics are partitioned in Kafka, requests originating from WebApi instances will be distributed across all Worker instances. However, to fine tune this, message key providers should be configured (see Kafka provider wiki and samples).
Check out the complete sample for image resizing.
- Types of messaging patterns supported:
- Publish-subscribe
- Request-response
- Queues
- Hybrid of the above (e.g. Kafka with multiple topic consumers in one group)
- Modern async/await syntax and TPL
- Fluent configuration
- Because SlimMessageBus is a facade, you have the ability to swap broker implementations
- Using nuget pull another broker provider
- Reconfigure SlimMessageBus and retest your app
- Try out the messaging middleware that works best for your app (Kafka vs. Redis) without having to rewrite your app.
- The core of
SlimMessageBus
is "slim"- Simple, common and friendly API to work with messaging systems
- No external dependencies.
- The core interface can be used in domain model (e.g. Domain Events)
- Plugin architecture:
- DI integration (Microsoft.Extensions.DependencyInjection, Autofac, CommonServiceLocator, Unity)
- Message serialization (JSON, XML)
- Use your favorite messaging broker as provider by simply pulling a NuGet package
- No threads created (pure TPL)
- Async/Await support
- Fluent configuration
- Logging is done via
Microsoft.Extensions.Logging.Abstractions
, so that you can connect your favorite logger provider.
cd src
dotnet build
dotnet pack --output ../dist
NuGet packaged end up in dist
folder
To run tests you need to update the respective appsettings.json
to match your own cloud infrastructure or local infrastructure.
SMB has some message brokers setup on Azure for integration tests (secrets not shared).
Run all tests:
dotnet test
Run all tests except integration tests which require local/cloud infrastructure:
dotnet test --filter Category!=Integration
Thanks to the following service cloud providers for providing free instances for our integration tests:
- Redis https://redislabs.com/
- Kafka https://www.cloudkarafka.com/
Other test instances are hosted in Azure and paid for by the project maintainer. If you want to help and sponsor, please write to me.