Routing topology

Target Version: NServiceBus 6.x
Standard support for version 6.x of NServiceBus has expired. For more information see our Support Policy.

The RabbitMQ transport has the concept of a routing topology, which controls how it creates exchanges, queues, and the bindings between them in the RabbitMQ broker. The routing topology also controls how the transport uses the exchanges it creates to send and publish messages. All endpoints in a system must use the same topology to be able to communicate with each other. For new systems, the conventional routing topology should be used. The direct routing topology is recommended only when adding an endpoint to an existing system that already uses that topology. A custom topology can be useful when integrating with a legacy system.

Conventional routing topology

The conventional routing topology relies on fanout exchanges to route messages.

The default and recommended routing topology is the conventional routing topology.

Sending using the conventional routing topology

Each endpoint creates a pair of a fanout exchange and a queue named after the endpoint's name. It also creates a binding between them. Messages are sent to the endpoint by publishing them to the endpoint's exchange. The binding then routes the message to the endpoint's queue.

Publishing using the conventional routing topology

For each type being published, a series of fanout exchanges are created to model the inheritance hierarchy of the type. For each type involved, an exchange is created, named in the following format: Namespace:TypeName. Bindings are created between the types, going from child to parent, until the entire hierarchy has been modeled. Exchanges are also created for each interface the type implements.

When an endpoint subscribes to an event, it first ensures that the above infrastructure exists. It then adds a binding from the exchange corresponding to the subscribed type to its own exchange.

When an endpoint publishes an event, it first ensures that the above infrastructure exists. It then publishes the message to the exchange corresponding to the message type being published.

Direct routing topology

The direct routing topology routes all events through a single exchange, amq.topic by default. Events are published using a routing key based on the event type, and subscribers will use that key to filter their subscriptions.

Sending using the direct routing topology

Every endpoint creates a queue named after the endpoint's name. When an endpoint sends a message it publishes it to a default exchange with a routing key equal to the destination endpoint name. This makes use of RabbitMQ default exchanges to move the message to a queue with the same name.

Publishing using the direct routing topology

Every endpoint publishes an event using the amq.topic exchange with a routing key of the form Namespace.TypeName, corresponding to the type of the event. The event is moved to all queues that have a binding for that event type.

An endpoint that subscribes to a given event creates a binding to the default exchange with the appropriate routing key.

In accordance with the AMQP 0.9.1 standard the routing key has a length limit of 255 characters.

Enabling the direct routing topology

To enable the direct routing topology, use the following configuration:

var transport = endpointConfiguration.UseTransport<RabbitMQTransport>();

Overriding the default conventions

The default conventions for exchange names and routing keys can be overridden by using the following overload:

var transport = endpointConfiguration.UseTransport<RabbitMQTransport>();
    routingKeyConvention: MyRoutingKeyConvention,
    exchangeNameConvention: (address, eventType) => "MyTopic");
In some cases, the direct routing topology may not deliver message types with "non-system" interfaces in their inheritance hierarchy. A "non-system" interface is any interface which is not contained in a .NET Framework assembly (any assembly signed with the same public key as mscorlib), and is not one of the interfaces. When using the direct routing topology, message types must not inherit from "non-system" interfaces. To guarantee delivery of message types which inherit from non-system interfaces, the conventional routing topology must be used.

Controlling exchange and queue durability

The routing topologies provided by the transport will create durable exchanges and queues unless durable messages have been disabled globally for the endpoint. Transient exchanges and queues will be created when DisableDurableMessages is called.

Custom routing topology

If the built-in routing topologies do not satisfy the requirements of the system, a custom routing topology may be used. To do this:

  1. Define the routing topology by creating a class implementing IRoutingTopology.
  2. Register it with the transport calling UseRoutingTopology as shown below.
var transport = endpointConfiguration.UseTransport<RabbitMQTransport>();
    topologyFactory: createDurableExchangesAndQueues =>
        return new MyRoutingTopology(createDurableExchangesAndQueues);

The boolean argument supplied to the factory delegate indicates whether the custom routing topology should create durable exchanges and queues on the broker. Read more about durable exchanges and queues in the AMQP Concepts Guide.

The generic overload of UseRoutingTopology shown below is still supported in NServiceBus.RabbitMQ version 4.x, but it is marked as [Obsolete] with a warning. It is not available in version 5.0 and above.
var transport = endpointConfiguration.UseTransport<RabbitMQTransport>();

For each queue required by the endpoint, the transport will first declare that queue and will then call the Initialize method of the routing topology. The routing topology should then perform all initialization related to that specific queue such as the creation of appropriate exchanges and bindings.

Taking control of queue declaration

In NServiceBus.RabbitMQ versions 4.2 and above, custom routing topologies can take control of queue declaration by implementing IDeclareQueues in addition to IRoutingTopology in the routing topology class.

When the routing topology implements IDeclareQueues, the transport will not declare queues, and it will call the DeclareAndInitialize method of the routing topology instead of calling the Initialize method. The routing topology is then responsible for creating the queue and any exchanges, bindings and extra queues which are required for that queue to operate. The Initialize method of the routing topology will not be called.

IRoutingTopology and IDeclareQueues have been combined into a single interface in NServiceBus.RabbitMQ versions 5 and above.

Delayed delivery

In Versions 4.3 and above, custom routing topologies can support delayed delivery without requiring the timeout manager by implementing ISupportDelayedDelivery in addition to IRoutingTopology in the routing topology class.

When the routing topology implements ISupportDelayedDelivery, the transport will call the BindToDelayInfrastructure method to allow the routing topology to create the appropriate bindings to the delivery exchange that is part of the native delayed delivery infrastructure.

IRoutingTopology and ISupportDelayedDelivery will be combined into a single interface in Version 5.0.

Related Articles

Last modified