Event Sourcing and CQRS are two of the most important patterns for data management in a Microservices architecture. We talked about the various data management patterns for microservices in this post.

In the next couple of posts, we will implement Event Sourcing and CQRS with Axon and Spring Boot. Here’s the plan we would be working on:

In Part 1 (this post), we will look at the concept behind CQRS. We would look at when CQRS is a viable option. Also, we would look at a classic implementation design for CQRS based systems.

In Part 2, we will actually implement Event Sourcing and CQRS in one of the simpler ways.

What is Event Sourcing?

On a high-level, Event Sourcing deals with storing the state of the application in the form of the domain events. We have already spent a fair amount of time on Event Sourcing in a previous series of posts. We have also implemented a working example for Event Sourcing using Axon and Spring Boot. You can check out the series here.

What is CQRS?

CQRS stands for Command Query Responsibility Segregation. Quite a mouthful, isn’t it?

However, it’s not half as bad as it sounds.

In a nutshell, CQRS works on the principle that the model to update information can be different to the model used to read that information.

But why do you need it?

Well, the need for this pattern arises from the Database-per-service pattern. The Database-per-service pattern says that each microservice should be responsible for its own data. While it sounds neat, this introduces a situation where it can become problematic to implement queries on data scattered across microservices.

Let’s consider an example to understand this. Suppose we are running a food delivery application. When an order is placed by a customer, below services could be triggered:

  • Order Service to create an order in the system.
  • Restaurant Service to contact the restaurant and place the order in the restaurants queue
  • Delivery Service to assign a delivery boy to the order and provide him/her the location to the restaurant and delivery location
  • Payment Service to handle the payment for the order.

Basically, all these services deal with a particular functionality. Each of these services can have their own database. They can have their own set of business rules.

However, if we want to provide a view to the customer about the entire Order lifecycle. This view should show everything that has happened in the life of an order. Arguably, such a view will need data from each of the above services.

How do we tackle this?

One way is to aggregate the data from each of those microservices by calling interfaces exposed by the services. However, this leads to unwanted in-memory joins. It also leads to tight coupling between the aggregator application and the individual microservices.

The other approach is to maintain a separate query store. This query store maintains the query specific view. Basically, this query store is built by listening to domain events from individual microservices. In the case of Order lifecycle example, such a query store could maintain an up-to date view of the order’s lifecycle ready to be served whenever the user requests.

Another reason for CQRS can be found when we look at User Interface. Typically, the user interfaces create a need to look at our data in different way. The user may need to see certain pieces of information together to perform certain actions. However, this view might differ from the underlying domain model.

CQRS starts to make sense when you see a need to split your conceptual or domain model into separate model for updates and reads.

Event Sourcing and CQRS

Often times, Event Sourcing and CQRS are used in conjunction with each other. Together, a combination of these two patterns can become a powerful tool for programmers.

In fact, it is often a critical requirement that an event sourced system also uses CQRS. It is harder to query an event sourced system. And hence, an efficient query store might be an indispensable need.

Some of the advantages of using Event Sourcing and CQRS are:

  • You can scale up the command (or update) side separately than your query (or read) side. This could be a great advantage for a system where reads outnumber writes by a huge margin.
  • You can chose different strategies for event store and query store. For example, event store can be a typical RDBMS. You can handle queries using NoSQL database (like MongoDB).
  • Using Event Sourcing and CQRS together, you can basically get rid of data aggregation pattern.

Event Sourcing and CQRS – A Classic Implementation Approach

Event Sourcing and CQRS are basically two separate patterns serving a common use-case. Ideally, we should implement them as two separate applications. Below is a high-level view of how such an implementation can look like:

event sourcing and cqrs implementation

Let’s understand what’s happening in this diagram

  • We have a command handler. Basically, all command requests are received here.
  • The command processing part takes care of handling all the commands and generating appropriate events. The events are persisted in the event store. Of course, validations and enforcement of business rules care performed before the events are persisted. Also, after the events are persisted, they are published on a message queue.
  • The messaging queue could be a broker like RabbitMQ or Kafka.
  • The Query Processing application listens to the events. Basically, this application takes the event payload and persists the data in the query store based on the required read models.
  • The query handler part handles the incoming read requests. It retrieves the data from the query store and outputs it.

How do we implement Event Sourcing and CQRS?

While the implementation diagram shown is the best way to implement Event Sourcing and CQRS, we will first implement it on a smaller level. Basically, we will perform Event Sourcing and CQRS in the same application.

We will be using Spring Boot for our normal application logic. However, for Event Sourcing and CQRS we will be using Axon Framework. Since we are using the same application for both Event Sourcing and CQRS, we will be using RDBMS (in this case an in-memory H2 database) as both a event store and query store. To connect to the database, we will leverage Spring Data JPA.

If you are not familiar with Spring Boot and H2 database, you can read more about it in this post.

In our case, we already have an application using Event Sourcing. We built it one of our earlier posts. You can read the step-by-step process over here. We will basically enhance the same application to support CQRS as well.

Since this post has already become too long, we will handle the implementation part in the next post.


Saurabh Dashora

Saurabh is a Software Architect with over 12 years of experience. He has worked on large-scale distributed systems across various domains and organizations. He is also a passionate Technical Writer and loves sharing knowledge in the community.

2 Comments

Juan Diego Hernandez · February 3, 2023 at 4:06 pm

Looks like your work has been copied: https://raghuramg.medium.com/implementing-event-sourcing-cqrs-using-axon-and-spring-boot-6eaa9ddde89f

BTW, I found the source code in github but it does not work. Do you reply to comments for that? Thanks.

    Saurabh Dashora · February 5, 2023 at 2:08 am

    Well, it’s hard to chase down people who wish to do a complete copy. Anyways, thanks for bringing it to my attention.

    As far as the source code in Github, I haven’t looked at it for many years now. Are you seeing any errors when you try to use it?

Leave a Reply

Your email address will not be published. Required fields are marked *