Skip to content

ftoumHub/event-sourcing-cqrs-examples

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build Status Coverage Status

Event Sourcing and CQRS Examples

This project aims to provide examples of how to use Event Sourcing and CQRS applied to a minimalistic bank context.

We assume the reader has basic knowledge of Event Sourcing and CQRS concepts.
If you want to brush up on the subject we suggest reading:

Domain overview

In this minimalistic bank, a client can open one or more accounts.
On each account, the client can deposit or withdraw money.
The history of an account's transactions is available to the client as well as a summary of the client's accounts.

Design choices

Architecture overview

  Event Store   Projections
    +----+        +----+
    |    |        |    |
    | DB |        | DB |
    +--+-+        +-+--+
      ^             ^
      |             |
+------------+------------+
|     |      |      |     |
|     |    Events   |     |
|     +------+----+ |     |
|     |      |    | |     |
|     +      |    v +     |
|   Domain   |   Read     |
|   Model    |   Model    |
|            |            |
+------------+------------+
|                         |
|           API           |
|                         |
+-------------------------+ 

Ports and Adapters

For the Domain Model, we chose the Ports and Adapters structure because we wanted to protect the domain logic from all the technical concerns.

For more information about it read here.

Package by Feature

For the Read Models, we chose the Package by Feature structure because we would not benefit from isolating the layers and instead we put all feature related parts close together.

For more information about it read here.

DDD and REST

There has been a myth of DDD and REST being incompatible due to DDD being all about behaviour whereas REST is all about state.
In this project we followed both techniques quite strictly and hope that the result shows that they can be well combined.
Note: We did not include REST hypermedia controls as we believe it is a big subject in itself and didn't want to shift focus from Event Sourcing and CQRS.

Event Sourcing and CQRS (finally!)

We have taken a pragmatic approach when combining Event Sourcing and CQRS. By the book, CQRS proposes a complete separation between the read/query and write/command sides, but that's not what we have here. The approach we've taken instead:

  • The writes/commands are all on the domain model side and processed by aggregates;
  • The reads/queries are both in the domain model side and in the read model side.
    • The queries in the domain model side are only allowed when the data we need is a single aggregate itself. The reason being that we can only query the event store by aggregate id and we can actually fulfill those queries by replaying that single aggregate events.
    • For any other kind of query, we don't want to compromise the domain model. Therefore, we create read models to fulfill those queries. They are basically projections, potentially built from different events and aggregates that can be queried by more appropriate fields.

Events

Events are a thing from the past. It communicates a significant change that happened.

Idempotency when replaying events

When replaying events, we don't want to execute any business logic because we can't change history. We only want to do assignments.
A simple example is with a deposit event: instead of adding the deposited amount to the balance when replaying (business logic), we want the updated balance already available so that we can just assign it. This makes it possible to replay the event multiple times with the same outcome.

Ordering of events

In a distributed world, event timestamps are unreliable for ordering - machines have their own clocks.
Instead we can make the ordering explicit with an event version. In this project we use event versioning in two ways:

  • In the write/command side, we use it for protecting ourselves from race conditions via optimistic locking;
  • In the read/query side, we use it for commutative reasons, meaning events can come out of order and we can still handle them properly.

If you are interested in this topic, we also recommend reading about Lamport timestamps as an alternative.

Trying it out

Requirements

  • Java 11
  • Maven

Building the application

mvn clean verify

Starting the application

java -jar target/bank-service-1.0-SNAPSHOT.jar server src/environments/development.yml

Examples of use

Create a client

curl -vvv -X POST -H "Content-Type: application/json" -d '{"name":"Jane Doe", "email":"[email protected]"}' http://localhost:8080/clients

Check the created client from the response 'Location' header

Create an account for the client

curl -vvv -X POST -H "Content-Type: application/json" -d '{"clientId":"{CLIENT_ID}"}' http://localhost:8080/accounts

Check the created account from the response 'Location' header

Make a deposit to the account

curl -vvv -X POST -H "Content-Type: application/json" -d '{"amount":1000000}' http://localhost:8080/accounts/{ACCOUNT_ID}/deposits

Check that you created a millionare!

curl -vvv http://localhost:8080/accounts/{ACCOUNT_ID}

More operations

Go ahead and check the code! :)

Resources

About

No description or website provided.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published