Awesome

EventSourcing .NET

Tutorial, practical samples and other resources about Event Sourcing in .NET. See also my similar repositories for JVM and NodeJS.

EventSourcing .NET

1. Event Sourcing

1.1 What is Event Sourcing?

Event Sourcing is a design pattern in which results of business operations are stored as a series of events.

It is an alternative way to persist data. In contrast with state-oriented persistence that only keeps the latest version of the entity state, Event Sourcing stores each state change as a separate event.

Thanks to that, no business data is lost. Each operation results in the event stored in the database. That enables extended auditing and diagnostics capabilities (both technically and business-wise). What's more, as events contains the business context, it allows wide business analysis and reporting.

In this repository I'm showing different aspects and patterns around Event Sourcing from the basic to advanced practices.

Read more in my articles:

1.2 What is Event?

Events represent facts in the past. They carry information about something accomplished. It should be named in the past tense, e.g. "user added", "order confirmed". Events are not directed to a specific recipient - they're broadcasted information. It's like telling a story at a party. We hope that someone listens to us, but we may quickly realise that no one is paying attention.

Events:

are immutable: "What has been seen, cannot be unseen".
can be ignored but cannot be retracted (as you cannot change the past).
can be interpreted differently. The basketball match result is a fact. Winning team fans will interpret it positively. Losing team fans - not so much.

Read more in my articles:

1.3 What is Stream?

Events are logically grouped into streams. In Event Sourcing, streams are the representation of the entities. All the entity state mutations end up as the persisted events. Entity state is retrieved by reading all the stream events and applying them one by one in the order of appearance.

A stream should have a unique identifier representing the specific object. Each event has its own unique position within a stream. This position is usually represented by a numeric, incremental value. This number can be used to define the order of the events while retrieving the state. It can also be used to detect concurrency issues.

1.4 Event representation

Technically events are messages.

They may be represented, e.g. in JSON, Binary, XML format. Besides the data, they usually contain:

id: unique event identifier.
type: name of the event, e.g. "invoice issued".
stream id: object id for which event was registered (e.g. invoice id).
stream position (also named version, order of occurrence, etc.): the number used to decide the order of the event's occurrence for the specific object (stream).
timestamp: representing a time at which the event happened.
other metadata like correlation id, causation id, etc.

Sample event JSON can look like:

{
  "id": "e44f813c-1a2f-4747-aed5-086805c6450e",
  "type": "invoice-issued",
  "streamId": "INV/2021/11/01",
  "streamPosition": 1,
  "timestamp": "2021-11-01T00:05:32.000Z",

  "data":
  {
    "issuedTo": {
      "name": "Oscar the Grouch",
      "address": "123 Sesame Street"
    },
    "amount": 34.12,
    "number": "INV/2021/11/01",
    "issuedAt": "2021-11-01T00:05:32.000Z"
  },

  "metadata":
  {
    "correlationId": "1fecc92e-3197-4191-b929-bd306e1110a4",
    "causationId": "c3cf07e8-9f2f-4c2d-a8e9-f8a612b4a7f1"
  }
}

Read more in my articles:

1.5 Event Storage

Event Sourcing is not related to any type of storage implementation. As long as it fulfills the assumptions, it can be implemented having any backing database (relational, document, etc.). The state has to be represented by the append-only log of events. The events are stored in chronological order, and new events are appended to the previous event. Event Stores are the databases' category explicitly designed for such purpose.

Read more in my articles:

1.6 Retrieving the current state from events

In Event Sourcing, the state is stored in events. Events are logically grouped into streams. Streams can be thought of as the entities' representation. Traditionally (e.g. in relational or document approach), each entity is stored as a separate record.

Id	IssuerName	IssuerAddress	Amount	Number	IssuedAt
e44f813c	Oscar the Grouch	123 Sesame Street	34.12	INV/2021/11/01	2021-11-01

In Event Sourcing, the entity is stored as the series of events that happened for this specific object, e.g. InvoiceInitiated, InvoiceIssued, InvoiceSent.

[
    {
        "id": "e44f813c-1a2f-4747-aed5-086805c6450e",
        "type": "invoice-initiated",
        "streamId": "INV/2021/11/01",
        "streamPosition": 1,
        "timestamp": "2021-11-01T00:05:32.000Z",

        "data":
        {
            "issuer": {
                "name": "Oscar the Grouch",
                "address": "123 Sesame Street",
            },
            "amount": 34.12,
            "number": "INV/2021/11/01",
            "initiatedAt": "2021-11-01T00:05:32.000Z"
        }
    },
    {
        "id": "5421d67d-d0fe-4c4c-b232-ff284810fb59",
        "type": "invoice-issued",
        "streamId": "INV/2021/11/01",
        "streamPosition": 2,
        "timestamp": "2021-11-01T00:11:32.000Z",

        "data":
        {
            "issuedTo": "Cookie Monster",
            "issuedAt": "2021-11-01T00:11:32.000Z"
        }
    },
    {
        "id": "637cfe0f-ed38-4595-8b17-2534cc706abf",
        "type": "invoice-sent",
        "streamId": "INV/2021/11/01",
        "streamPosition": 3,
        "timestamp": "2021-11-01T00:12:01.000Z",

        "data":
        {
            "sentVia": "email",
            "sentAt": "2021-11-01T00:12:01.000Z"
        }
    }
]

All of those events share the stream id ("streamId": "INV/2021/11/01"), and have incremented stream positions.

In Event Sourcing each entity is represented by its stream: the sequence of events correlated by the stream id ordered by stream position.

To get the current state of an entity we need to perform the stream aggregation process. We're translating the set of events into a single entity. This can be done with the following steps:

Read all events for the specific stream.
Order them ascending in the order of appearance (by the event's stream position).
Construct the empty object of the entity type (e.g. with default constructor).
Apply each event on the entity.

This process is called also stream aggregation or state rehydration.

We could implement that as:

public record Person(
    string Name,
    string Address
);

public record InvoiceInitiated(
    double Amount,
    string Number,
    Person IssuedTo,
    DateTime InitiatedAt
);

public record InvoiceIssued(
    string IssuedBy,
    DateTime IssuedAt
);

public enum InvoiceSendMethod
{
    Email,
    Post
}

public record InvoiceSent(
    InvoiceSendMethod SentVia,
    DateTime SentAt
);

public enum InvoiceStatus
{
    Initiated = 1,
    Issued = 2,
    Sent = 3
}

public class Invoice
{
    public string Id { get;set; }
    public double Amount { get; private set; }
    public string Number { get; private set; }

    public InvoiceStatus Status { get; private set; }

    public Person IssuedTo { get; private set; }
    public DateTime InitiatedAt { get; private set; }

    public string IssuedBy { get; private set; }
    public DateTime IssuedAt { get; private set; }

    public InvoiceSendMethod SentVia { get; private set; }
    public DateTime SentAt { get; private set; }

    public void Evolve(object @event)
    {
        switch (@event)
        {
            case InvoiceInitiated invoiceInitiated:
                Apply(invoiceInitiated);
                break;
            case InvoiceIssued invoiceIssued:
                Apply(invoiceIssued);
                break;
            case InvoiceSent invoiceSent:
                Apply(invoiceSent);
                break;
        }
    }

    private void Apply(InvoiceInitiated @event)
    {
        Id = @event.Number;
        Amount = @event.Amount;
        Number = @event.Number;
        IssuedTo = @event.IssuedTo;
        InitiatedAt = @event.InitiatedAt;
        Status = InvoiceStatus.Initiated;
    }

    private void Apply(InvoiceIssued @event)
    {
        IssuedBy = @event.IssuedBy;
        IssuedAt = @event.IssuedAt;
        Status = InvoiceStatus.Issued;
    }

    private void Apply(InvoiceSent @event)
    {
        SentVia = @event.SentVia;
        SentAt = @event.SentAt;
        Status = InvoiceStatus.Sent;
    }
}

and use it as:

var invoiceInitiated = new InvoiceInitiated(
    34.12,
    "INV/2021/11/01",
    new Person("Oscar the Grouch", "123 Sesame Street"),
    DateTime.UtcNow
);
var invoiceIssued = new InvoiceIssued(
    "Cookie Monster",
    DateTime.UtcNow
);
var invoiceSent = new InvoiceSent(
    InvoiceSendMethod.Email,
    DateTime.UtcNow
);

// 1,2. Get all events and sort them in the order of appearance
var events = new object[] {invoiceInitiated, invoiceIssued, invoiceSent};

// 3. Construct empty Invoice object
var invoice = new Invoice();

// 4. Apply each event on the entity.
foreach (var @event in events)
{
    invoice.Evolve(@event);
}

and generalise this into Aggregate base class:

public abstract class Aggregate<T>
{
    public T Id { get; protected set; }

    protected Aggregate() { }

    public virtual void Evolve(object @event) { }
}

The biggest advantage of "online" stream aggregation is that it always uses the most recent business logic. So after the change in the apply method, it's automatically reflected on the next run. If events data is fine, then it's not needed to do any migration or updates.

In Marten Evolve method is not needed. Marten uses naming convention and call the Apply method internally. It has to:

have single parameter with event object,
have void type as the result.

See samples:

1.7 Strongly-Typed ids with Marten

Strongly typed ids (or, in general, a proper type system) can make your code more predictable. It reduces the chance of trivial mistakes, like accidentally changing parameters order of the same primitive type.

So for such code:

var reservationId = "RES/01";
var seatId = "SEAT/22";
var customerId = "CUS/291";

var reservation = new Reservation(
    reservationId,
    seatId,
    customerId
);

the compiler won't catch if you switch reservationId with seatId.

If you use strongly typed ids, then compile will catch that issue:

var reservationId = new ReservationId("RES/01");
var seatId = new SeatId("SEAT/22");
var customerId = new CustomerId("CUS/291");

var reservation = new Reservation(
    reservationId,
    seatId,
    customerId
);

They're not ideal, as they're usually not playing well with the storage engines. Typical issues are: serialisation, Linq queries, etc. For some cases they may be just overkill. You need to pick your poison.

To reduce tedious, copy/paste code, it's worth defining a strongly-typed id base class, like:

public class StronglyTypedValue<T>: IEquatable<StronglyTypedValue<T>> where T: IComparable<T>
{
    public T Value { get; }

    public StronglyTypedValue(T value)
    {
        Value = value;
    }

    public bool Equals(StronglyTypedValue<T>? other)
    {
        if (ReferenceEquals(null, other)) return false;
        if (ReferenceEquals(this, other)) return true;
        return EqualityComparer<T>.Default.Equals(Value, other.Value);
    }

    public override bool Equals(object? obj)
    {
        if (ReferenceEquals(null, obj)) return false;
        if (ReferenceEquals(this, obj)) return true;
        if (obj.GetType() != this.GetType()) return false;
        return Equals((StronglyTypedValue<T>)obj);
    }

    public override int GetHashCode()
    {
        return EqualityComparer<T>.Default.GetHashCode(Value);
    }

    public static bool operator ==(StronglyTypedValue<T>? left, StronglyTypedValue<T>? right)
    {
        return Equals(left, right);
    }

    public static bool operator !=(StronglyTypedValue<T>? left, StronglyTypedValue<T>? right)
    {
        return !Equals(left, right);
    }
}

Then you can define specific id class as:

public class ReservationId: StronglyTypedValue<Guid>
{
    public ReservationId(Guid value) : base(value)
    {
    }
}

You can even add additional rules:

public class ReservationNumber: StronglyTypedValue<string>
{
    public ReservationNumber(string value) : base(value)
    {
        if (string.IsNullOrEmpty(value) || !value.StartsWith("RES/") || value.Length <= 4)
            throw new ArgumentOutOfRangeException(nameof(value));
    }
}

The base class working with Marten, can be defined as:

public abstract class Aggregate<TKey, T>
    where TKey: StronglyTypedValue<T>
    where T : IComparable<T>
{
    public TKey Id { get; set; } = default!;

    [Identity]
    public T AggregateId    {
        get => Id.Value;
        set {}
    }

    public int Version { get; protected set; }

    [JsonIgnore] private readonly Queue<object> uncommittedEvents = new();

    public object[] DequeueUncommittedEvents()
    {
        var dequeuedEvents = uncommittedEvents.ToArray();

        uncommittedEvents.Clear();

        return dequeuedEvents;
    }

    protected void Enqueue(object @event)
    {
        uncommittedEvents.Enqueue(@event);
    }
}

Marten requires the id with public setter and getter of string or Guid. We used the trick and added AggregateId with a strongly-typed backing field. We also informed Marten of the Identity attribute to use this field in its internals.

Example aggregate can look like:

public class Reservation : Aggregate<ReservationId, Guid>
{
    public CustomerId CustomerId { get; private set; } = default!;

    public SeatId SeatId { get; private set; } = default!;

    public ReservationNumber Number { get; private set; } = default!;

    public ReservationStatus Status { get; private set; }

    public static Reservation CreateTentative(
        SeatId seatId,
        CustomerId customerId)
    {
        return new Reservation(
            new ReservationId(Guid.NewGuid()),
            seatId,
            customerId,
            new ReservationNumber(Guid.NewGuid().ToString())
        );
    }

    // (...)
}

See the full sample here.

2. Videos

2.1. Practical Event Sourcing with Marten

2.2. Keep your streams short! Or how to model event-sourced systems efficiently

2.3. Let's build event store in one hour!

2.4. CQRS is Simpler than you think with C#11 & NET7

2.5. Practical Introduction to Event Sourcing with EventStoreDB

2.6. How to deal with privacy and GDPR in Event-Sourced systems

2.7 Let's build the worst Event Sourcing system!

2.8 The Light and The Dark Side of the Event-Driven Design

2.9 Implementing Distributed Processes

2.10 Conversation with Yves Lorphelin about CQRS

2.11. Never Lose Data Again - Event Sourcing to the Rescue!

3. Support

Feel free to create an issue if you have any questions or request for more explanation or samples. I also take Pull Requests!

💖 If this repository helped you - I'd be more than happy if you join the group of my official supporters at:

👉 Github Sponsors

⭐ Star on GitHub or sharing with your friends will also help!

4. Prerequisites

For running the Event Store examples you need to have:

.NET 6 installed - https://dotnet.microsoft.com/download/dotnet/6.0
Docker installed. Then going to the docker folder and running:

docker compose --profile all up

More information about using .NET, WebApi and Docker you can find in my other tutorials: WebApi with .NET

5. Tools used

Marten - Event Store and Read Models
EventStoreDB - Event Store
Kafka - External Durable Message Bus to integrate services
ElasticSearch - Read Models

6. Samples

See also fully working, real-world samples of Event Sourcing and CQRS applications in Samples folder.

Samples are using CQRS architecture. They're sliced based on the business modules and operations. Read more about the assumptions in "How to slice the codebase effectively?".

6.1 Pragmatic Event Sourcing With Marten

Simplest CQRS and Event Sourcing flow using Minimal API,
Cutting the number of layers and boilerplate complex code to bare minimum,
Using all Marten helpers like WriteToAggregate, AggregateStream to simplify the processing,
Examples of all the typical Marten's projections,
Example of how and where to use C# Records, Nullable Reference Types, etc,
No Aggregates. Commands are handled in the domain service as pure functions.

6.2 ECommerce with Marten

typical Event Sourcing and CQRS flow,
DDD using Aggregates,
microservices example,
stores events to Marten,
distributed processes coordinated by Saga (Order Saga),
Kafka as a messaging platform to integrate microservices,
example of the case when some services are event-sourced (Carts, Orders, Payments) and some are not (Shipments using EntityFramework as ORM)

6.3 Simple EventSourcing with EventStoreDB

typical Event Sourcing and CQRS flow,
functional composition, no aggregates, just data and functions,
stores events to EventStoreDB,
Builds read models using Subscription to $all,
Read models are stored as Postgres tables using EntityFramework.

6.4 Implementing Distributed Processes

orchestrate and coordinate business workflow spanning across multiple aggregates using Saga pattern,
handle distributed processing both for asynchronous commands scheduling and events publishing,
getting at-least-once delivery guarantee,
implementing command store and outbox pattern on top of Marten and EventStoreDB,
unit testing aggregates and Saga with a little help from Ogooreck,
testing asynchronous code.

6.5 ECommerce with EventStoreDB

typical Event Sourcing and CQRS flow,
DDD using Aggregates,
stores events to EventStoreDB,
Builds read models using Subscription to $all.
Read models are stored as Marten documents.

6.6 Warehouse

simplest CQRS flow using .NET Endpoints,
example of how and where to use C# Records, Nullable Reference Types, etc,
No Event Sourcing! Using Entity Framework to show that CQRS is not bounded to Event Sourcing or any type of storage,
No Aggregates! CQRS do not need DDD. Business logic can be handled in handlers.

6.7 Warehouse Minimal API

Variation of the previous example, but:

using Minimal API,
example how to inject handlers in MediatR like style to decouple API from handlers.
📝 Read more CQRS is simpler than you think with .NET 6 and C# 10

6.8 Event Versioning

Shows how to handle basic event schema versioning scenarios using event and stream transformations (e.g. upcasting):

6.9 Event Pipelines

Shows how to compose event handlers in the processing pipelines to:

filter events,
transform them,
NOT requiring marker interfaces for events,
NOT requiring marker interfaces for handlers,
enables composition through regular functions,
allows using interfaces and classes if you want to,
can be used with Dependency Injection, but also without through builder,
integrates with MediatR if you want to.
📝 Read more How to build a simple event pipeline

6.10 Meetings Management with Marten

typical Event Sourcing and CQRS flow,
DDD using Aggregates,
microservices example,
stores events to Marten,
Kafka as a messaging platform to integrate microservices,
read models handled in separate microservice and stored to other database (ElasticSearch)

6.11 Cinema Tickets Reservations with Marten

typical Event Sourcing and CQRS flow,
DDD using Aggregates,
stores events to Marten.

6.12 SmartHome IoT with Marten

typical Event Sourcing and CQRS flow,
DDD using Aggregates,
stores events to Marten,
asynchronous projections rebuild using AsyncDaemon feature.

7. Self-paced training Kits

I prepared the self-paced training Kits for the Event Sourcing. See more in the Workshop description.

7.1 Introduction to Event Sourcing

Event Sourcing is perceived as a complex pattern. Some believe that it's like Nessie, everyone's heard about it, but rarely seen it. In fact, Event Sourcing is a pretty practical and straightforward concept. It helps build predictable applications closer to business. Nowadays, storage is cheap, and information is priceless. In Event Sourcing, no data is lost.

The workshop aims to build the knowledge of the general concept and its related patterns for the participants. The acquired knowledge will allow for the conscious design of architectural solutions and the analysis of associated risks.

The emphasis will be on a pragmatic understanding of architectures and applying it in practice using Marten and EventStoreDB.

You can do the workshop as a self-paced kit. That should give you a good foundation for starting your journey with Event Sourcing and learning tools like Marten and EventStoreDB. If you'd like to get full coverage with all nuances of the private workshop, feel free to contact me via email.

Events definition.
Getting State from events.
Appending Events:
- Marten
- EventStoreDB
Getting State from events
- Marten
- EventStoreDB
Business logic:
Optimistic Concurrency:
- Marten
- EventStoreDB
Projections:

7.2 Build your own Event Store

It teaches the event store basics by showing how to build your Event Store on top of Relational Database. It starts with the tables setup, goes through appending events, aggregations, projections, snapshots, and finishes with the Marten basics.

8. Articles

Read also more on the Event Sourcing and CQRS topics in my blog posts:

9. Event Store - Marten

Creating event store
Event Stream - is a representation of the entity in event sourcing. It's a set of events that happened for the entity with the exact id. Stream id should be unique, can have different types but usually is a Guid.
- Stream starting - stream should be always started with a unique id. Marten provides three ways of starting the stream:
  - calling StartStream method with a stream id
```
var streamId = Guid.NewGuid();
documentSession.Events.StartStream<IssuesList>(streamId);
```
  - calling StartStream method with a set of events
```
var @event = new IssueCreated { IssueId = Guid.NewGuid(), Description = "Description" };
var streamId = documentSession.Events.StartStream<IssuesList>(@event);
```
  - just appending events with a stream id
```
var @event = new IssueCreated { IssueId = Guid.NewGuid(), Description = "Description" };
var streamId = Guid.NewGuid();
documentSession.Events.Append(streamId, @event);
```
- Stream loading - all events that were placed on the event store should be possible to load them back. Marten allows to:
  - get list of event by calling FetchStream method with a stream id
```
var eventsList = documentSession.Events.FetchStream(streamId);
```
  - geting one event by its id
```
var @event = documentSession.Events.Load<IssueCreated>(eventId);
```
- Stream loading from exact state - all events that were placed on the event store should be possible to load them back. Marten allows to get stream from exact state by:
  - timestamp (has to be in UTC)
```
var dateTime = new DateTime(2017, 1, 11);
var events = documentSession.Events.FetchStream(streamId, timestamp: dateTime);
```
  - version number
```
var versionNumber = 3;
var events = documentSession.Events.FetchStream(streamId, version: versionNumber);
```
Event stream aggregation - events that were stored can be aggregated to form the entity once again. During the aggregation, process events are taken by the stream id and then replayed event by event (so eg. NewTaskAdded, DescriptionOfTaskChanged, TaskRemoved). At first, an empty entity instance is being created (by calling default constructor). Then events based on the order of appearance are being applied on the entity instance by calling proper Apply methods.
- Online Aggregation - online aggregation is a process when entity instance is being constructed on the fly from events. Events are taken from the database and then aggregation is being done. The biggest advantage of online aggregation is that it always gets the most recent business logic. So after the change, it's automatically reflected and it's not needed to do any migration or updates.
- Inline Aggregation (Snapshot) - inline aggregation happens when we take the snapshot of the entity from the DB. In that case, it's not needed to get all events. Marten stores the snapshot as a document. This is good for performance reasons because only one record is being materialized. The con of using inline aggregation is that after business logic has changed records need to be reaggregated.
- Reaggregation - one of the biggest advantages of the event sourcing is flexibility to business logic updates. It's not needed to perform complex migration. For online aggregation it's not needed to perform reaggregation - it's being made always automatically. The inline aggregation needs to be reaggregated. It can be done by performing online aggregation on all stream events and storing the result as a snapshot.
  - reaggregation of inline snapshot with Marten
```
var onlineAggregation = documentSession.Events.AggregateStream<TEntity>(streamId);
documentSession.Store<TEntity>(onlineAggregation);
documentSession.SaveChanges();
```
Event transformations
- One event to one object transformations
- Inline Transformation storage
Events projection
- Projection of single stream
Multitenancy per schema

10. CQRS (Command Query Responsibility Separation)

Command handling
Query handling

11. NuGet packages to help you get started.

I gathered and generalized all of the practices used in this tutorial/samples in Nuget Packages maintained by me GoldenEye Framework. See more in:

GoldenEye DDD package - it provides a set of base and bootstrap classes that helps you to reduce boilerplate code and help you focus on writing business code. You can find all classes like Commands/Queries/Event handlers and many more. To use it run:

dotnet add package GoldenEye
GoldenEye Marten package - contains helpers, and abstractions to use Marten as document/event store. Gives you abstractions like repositories etc. To use it run:

dotnet add package GoldenEye.Marten

12. Other resources

12.1 Introduction

12.2 Event Sourcing on production

12.3 Projections

12.4 Snapshots

12.5 Versioning

12.6 Storage

12.7 Design & Modeling

12.8 GDPR

12.9 Conflict Detection

12.10 Functional programming

📝 Jérémie Chassaing - Functional Programming and Event Sourcing

12.12 Testing

🎞 N. Rauch & A. Bailly - From Front to Back: Homomorphic Event Sourcing

12.13 CQRS

12.14 Tools

12.15 Event processing

12.16 Distributed processes

12.17 Domain Driven Design

12.18 Whitepapers

12.19 Event Sourcing Concerns

12.20 This is NOT Event Sourcing (but Event Streaming)

12.21 Architecture Weekly

If you're interested in Architecture resources, check my other repository: https://github.com/oskardudycz/ArchitectureWeekly/.

It contains a weekly updated list of materials I found valuable and educational.

License

This blog is licensed under License Creative Commons BY-SA 4.0.