In a microservice environment or any other distributed system, you may come upon the requirement to exchange events between services. This article shows how to implement a messaging solution with RabbitMQ.
Event Messaging Requirements
Before jumping into the solution, let’s define some requirements that an eventing mechanism in a distributed system should fulfill. We’ll use the following diagram to derive those requirements.
- The event producing service must not call the event consuming services directly in order to preserve loose coupling.
- The event producing service must be able to send events of different types (e.g. “customer.created” or “customer.deleted”).
- The event consuming services must be able to receive only events of types they are interested in (e.g. “*.deleted,” which means all events concerning a customer).
- In our distributed system we have several service clusters (e.g. a cluster of “order service” instances and a cluster of “archive service” instances). Each event must be processed by at most one instance per service cluster.
The eventing solution presented in this article makes use of some messaging concepts that are described in the following sections.
A producer is simply a piece of software that sends a message to a message broker, for example, a customer service in a system of microservices that wants to tell other services that a new customer was created by sending the event
customer.created that contains the newly created customers’ ID as a payload.
A consumer is a piece of software that receives messages from a message broker and processes those messages. In our example, this might be an order service that needs the address of all customers to create orders for those customers. It would process the
customer.created event by reading the ID from the event and calling the customer service to load the corresponding customers’ address.
A queue is first-in-first-out message store. The messages are put into a queue by a producer and read from it by a consumer. Once a message is read, it is consumed and removed from the queue. A message can thus only be processed exactly once.
An exchange is a concept that is part of the AMQP protocol. Basically, it acts as an intermediary between the producer and a queue. Instead of sending messages directly to a queue, a producer can send them to an exchange instead. The exchange then sends those messages to one or more queues following a specified set of rules. Thus, the producer does not need to know the queues that eventually receive those messages.
A binding connects a queue to an exchange. The exchange forwards all messages it receives to the queues it is bound to. A binding can contain a routing key that specifies which events should be forwarded. For example, a binding might contain the routing key
customer.* meaning that all events whose type starts with
customer. will be routed to the specified queue.
An Event Messaging Concept With AMQP
Using the concepts above, we can create an eventing solution with RabbitMQ. The solution is depicted in the figure below.
Each service cluster gets its own queue. This is necessary since not all events are relevant to each service cluster. An order service may be interested in all customer events (
customer.*) whereas an archiving service may be interested in all events where an object has been deleted (
*.deleted). If we had only one queue for all events that queue would sooner or later overflow since it might contain events that no consumer is interested in.
Each consuming service cluster binds its queue the central exchange with a routing key that specifies which events it is interested in. Only those events are then routed into the queue. The events are then consumed by exactly one of the service instances connected to that queue.
The event producing services only need to know the central exchange and send all events to that exchange. Since the consuming services take care of the binding and routing, we have a real, loosely coupled eventing mechanism.
Implementing Event Messaging With Spring Boot and RabbitMQ
The eventing concept described above can be implemented with Spring Boot and RabbitMQ. The implementation is pretty straightforward. If you don’t feel like reading and more like delving into code, you will find a link to a GitHub repository with a working example at the end of this article.
Including the Spring Boot AMQP Starter
Spring Boot offers a starter for Messaging with AMQP that integrates the Spring AMQP project with Spring Boot. The AMQP Starter currently only supports RabbitMQ as underlying message broker, which is fine for us. To use the starter, include the following dependency into your project (Gradle notation):
The starter contains an auto configuration which is automatically activated.
Connecting to RabbitMQ
In order to connect to a RabbitMQ server, the Spring AMQP starter reads the following properties, which you can specify as environment variables, for example in your
application.properties. The following settings are the default connection settings once you have installed RabbitMQ locally.
Configuring an Event Producer
Creating an event producer is pretty straightforward. We make use of the
RabbitTemplate provided by the AMQP starter and call the method
convertAndSend() to send an event. The event in the code example only contains a String. If the message should contain a complex object, you can make use of message converters.
RabbitTemplate automatically uses the connection settings provided in the
Note that the call to
RabbitTemplate needs the name of the exchange to which the event should be sent. To wire our application against a specific exchange, we simply create a Spring Bean of type
TopicExchange and choose a name for that exchange (in case of the example code below, the exchange is called
eventExchange). The application will automatically connect to RabbitMQ and create an exchange with this name if it doesn’t exist yet. We use a so-called “topic exchange” here, since it allows to specify a routing key (a “topic”) when sending a message to it.
RabbitTemplate passed into the
CustomerService is provided to the Spring application context by the AMQP starter.
Configuring an Event Consumer
First off, the event consumer itself is a simple java class. Again, to process more complex objects than simple strings, you can use Spring AMQPs message converters. We use the
@RabbitListener annotation on a method to mark it as an event receiver.
We now need to declare a queue and bind it to the same exchange used in the event producer.
First, we define the same
Exchange as we did in the event consumer configuration. Then, we define a
Queue with a unique name. This is the queue for our service cluster. To connect the two, we then create a
Binding with the routing key
customer.* specifying that we are only interested in customer events.
As with the exchange before, a Queue and a Binding will be automatically created on the RabbitMQ server if they do not exist yet.
With the concepts of exchanges, bindings, and queues, AMQP provides everything we need to create an event mechanism for a distributed system. Spring AMQP and its integration into Spring Boot via the AMQP Starter provide a very convenient programming model to connect to such an event broker.
A working code example can be found in my GitHub repository. It contains a readme with which you can get it up and running in no time. Note that the code examples in this article are modified for better readability and do not match the code in the repository completely.