# Messages

In MassTransit, a message is defined using the .NET type system. Messages can be defined using both classes and interfaces, however, it is suggested that types use read-only properties and no behavior.

TIP

It is strongly suggested to use interfaces for message contracts, based on experience over several years with varying levels of developer experience. MassTransit will create dynamic interface implementations for the messages, ensuring a clean separation of the message contract from the consumer.

An example message to update a customer address is shown below.

	namespace Company.Application.Contracts
	{
		using System;

		public interface UpdateCustomerAddress
		{
			Guid CommandId { get; }
			DateTime Timestamp { get; }
			string CustomerId { get; }
			string HouseNumber { get; }
			string Street { get; }
			string City { get; }
			string State { get; }
			string PostalCode { get; }
		}
	}

A common mistake when engineers are new to messaging is to create a base class for messages, and try to dispatch that base class in the consumer – including the behavior of the subclass. Ouch. This always leads to pain and suffering, so just say no to base classes.

# Message Names

There are two main message types, events and commands. When choosing a name for a message, the type of message should dictate the tense of the message.

# Commands

A command tells a service to do something. Commands are sent (using Send) to an endpoint, as it is expected that a single service instance performs the command action. A command should never be published.

Commands should be expressed in a verb-noun sequence, following the tell style.

Example Commands:

  • UpdateCustomerAddress
  • UpgradeCustomerAccount
  • SubmitOrder

# Events

An event signifies that something has happened. Events are published (using Publish) via either IBus (standalone) or ConsumeContext (within a message consumer). An event should not be sent directly to an endpoint.

Events should be expressed in a noun-verb (past tense) sequence, indicating that something happened.

Example Events:

  • CustomerAddressUpdated
  • CustomerAccountUpgraded
  • OrderSubmitted, OrderAccepted, OrderRejected, OrderShipped

# Message Headers

MassTransit encapsulates every sent or published message in a message envelope (described by the Envelope Wrapper pattern). The envelope includes the message along with additional message headers, including:

Property Type Description
MessageId Auto Generated for each message using NewId.NextGuid.
CorrelationId User Assigned by the application, or automatically by convention, and should uniquely identify the operation, event, etc.
RequestId Request Assigned by the request client, and automatically copied by the Respond methods to correlate responses to the original request.
InitiatorId Auto Assigned when publishing or sending from a consumer, saga, or activity to the value of the CorrelationId on the consumed message.
ConversationId Auto Assigned when the first message sent or published and no consumed message is available, ensuring that a set of messages within the same conversation have the save identifier.
SourceAddress Auto Where the message originated (may be a temporary address for messages published or sent from IBus).
DestinationAddress Auto Where the message was sent
ResponseAddress Request Where responses to the request should be sent. If not present, responses are published.
FaultAddress User Where consumer faults should be sent. If not present, faults are published.
ExpirationTime User When the message should expire, which may be used by the transport to remove the message if it isn't consumed by the expiration time.
SentTime Auto When the message was sent, in UTC.
MessageType Auto An array of message types, in a MessageUrn format, which can be deserialized.
Host Auto The host information of the machine that sent or published the message.
Headers User Additional headers, which can be added by the user, middleware, or diagnostic trace filters.

# Message Correlation

Since messages usually do not live in isolation, publishing one message usually leads to publishing another message, and then another, and so on. It is useful to trace such sequences, however, to find them these messages need to have some information detailing how they relate to each other.

Correlation is the principle of connecting messages together, usually by using a unique identifier that is included in every message that is part of a logical sequence. In MassTransit, the unique identifier is referred to as the CorrelationId, which is included in the message envelope and available via the ConsumeContext or the SendContext. MassTransit also includes a ConversationId which is the same across an entire set of related messages.

In a distributed message-based system, message correlation is very important. Since operations are potentially executing across hundreds of nodes, the ability to correlate different messages to build a path through the system is absolutely necessary for engineers to troubleshoot problems.

The headers on the message envelope provided by MassTransit already make it easy to specify correlation values. In fact, most are setup by default if not specified by the developer.

MassTransit provides the interface CorrelatedBy<T>, which can be used to setup a default correlationId. This is used by sagas as well, since all sagas have a unique CorrelationId for each instance of the saga. If a message implements CorrelatedBy<Guid>, it will automatically be directed to the saga instance with the matching identifier. If a new saga instance is created by the event, it will be assigned the CorrelationId from the initiating message.

For message types that have a correlation identifier, but are not using the CorrelatedBy interface, it is possible to declare the identifier for the message type and MassTransit will use that identifier by default for correlation.

MessageCorrelation.UseCorrelationId<YourMessageClass>(x => x.SomeGuidValue);

TIP

This should be called before you start the bus. We currently recommend that you put all of these in a static method for easy grouping and then call it at the beginning of the MassTransit configuration block.

Most transactions in a system will end up being logged and wide scale correlation is likely. Therefore, the use of consistent correlation identifiers is recommended. In fact, using a Guid type is highly recommended. MassTransit uses the NewId library to generate identifiers that are unique and sequential that are represented as a Guid. The identifiers are clustered-index friendly, being ordered in a way that SQL Server can efficiently insert them into a database with the uniqueidentifier as the primary key. Just use NewId.NextGuid() to generate an identifier -- it's fast, fun, and all your friends are doing it.

TIP

So, what does correlated actually mean? In short it means that this message is a part of a larger conversation. For instance, you may have a message that says New Order (Item:Hammers; Qty:22; OrderNumber:45) and there may be another message that is a response to that message that says Order Allocated(OrderNumber:45). In this case, the order number is acting as your correlation identifier, it ties the messages together.

# Correlation Conventions

In addition to the explicit CorrelateBy<T> interface, a convention-based correlation is supported. If the message contract has a property named CorrelationId, CommandId, or EventId, the correlationId header is automatically populated on Send or Publish. It can also be manually specified using the SendContext.

Bear in mind that sagas default CorrelateById() only support messages where the explicit CorrelateBy<Guid> interface is implemented. However, the header is still useful if you do not use sagas, for example for message flow analysis and debugging.

# Guidelines

Given everything that was stated above, here are a few guidelines.

  1. Interface-based inheritance is OK, don't be afraid, but don't go nuts.
  2. Class-based inheritance is to be approached with caution.
  3. Composing messages together ends up pushing us into content-based routing which is something we don't recommend.
  4. Message Design is not OO Design (a message is just state, no behavior). There is a greater focus on interop and contract design.
  5. As messages are more about contracts, we suggest subscribing to interfaces that way you can easily evolve the message definition.
  6. A big base class may cause pain down the road as each change will have a larger ripple. This can be especially bad when you need to support multiple versions.