Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
168 changes: 168 additions & 0 deletions src/content/docs/dotnet-next/infra/azure-blob-storage.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,168 @@
---
title: "Azure Blob Storage"
description: "Projections for Azure Blob Storage"
sidebar:
order: 8
---

[Azure Blob Storage](https://azure.microsoft.com/en-us/products/storage/blobs/) is a fully managed object storage service in the cloud. Eventuous supports Blob Storage as a [projection](../../read-models/rm-concept) target using the `Eventuous.Azure.Storage.Blobs` package.
It allows you to project event store events to Azure Blob Storage as state objects, maintaining a separate state document for each event stream.

## Using projections

Create your own projection class that inherits from `BlobStorageProjector<T>` where `T` is your state type. The state type must be a class with a parameterless constructor.

Register event handlers using the `On<TEvent>` methods. When an event is received, the projector retrieves the current state blob (or creates a new state instance if the blob doesn't exist), applies the event to the state using the registered event handler, and uploads the updated state back to Blob Storage.

The class provides two constructors:

* `BlobStorageProjector(BlobContainerClient container, ...)` where the container client is passed directly
* `BlobStorageProjector(BlobServiceClient serviceClient, string containerName, ...)` where the service client comes from DI and the container name is set by the projection

```csharp
public class BookingProjection : BlobStorageProjector<BookingState> {
public BookingProjection(BlobServiceClient client)
: base(client, "bookings-container") {

// Uses default blob ID from stream
On<BookingImported>((state, evt) => {
state.RoomId = evt.RoomId;
state.CheckInDate = evt.CheckIn;
return state;
});

// Custom blob ID using event data
On<BookingPaymentRegistered>(
(state, evt) => {
state.PaidAmount += evt.AmountPaid;
return state;
},
context => new ValueTask<string>($"custom-{context.Message.BookingId}")
);
}
}
```

:::caution
The blob container must exist before the projector handles events. The projector doesn't create it.
:::

### Registration

The projector needs an Azure `BlobServiceClient` registered in the DI container:

```csharp
builder.Services.AddSingleton(new BlobServiceClient(connectionString));
```

Then add the projection to a [subscription](../../subscriptions/subs-concept) as an event handler:

```csharp
builder.Services.AddSubscription<AllStreamSubscription, AllStreamSubscriptionOptions>(
"BookingsBlobProjection",
b => b.AddEventHandler<BookingProjection>()
);
```

Give the projection its own subscription and [checkpoint](../../subscriptions/checkpoint) when you add it to a system that already has data, so it replays from the beginning of the log and backfills the blobs instead of resuming from another projection's position.

### JSON serialization

JSON serialization is configured via `BlobStorageProjectorOptions.JsonOptions`. When it isn't set, the projector uses `JsonSerializerOptions.Web`. If you keep serializer options in DI, pass them on. The example below uses `Microsoft.AspNetCore.Http.Json.JsonOptions`, which is what minimal APIs configure:

```csharp
public BookingProjection(BlobServiceClient client, IOptions<JsonOptions> options)
: base(
client,
"bookings-container",
new BlobStorageProjectorOptions { JsonOptions = options.Value.SerializerOptions }
) { }
```

## Projector options

The `BlobStorageProjectorOptions` class provides several configuration options for fine-tuning the projector behavior.

| Option | Type | Default | Description |
|--------|------|---------|-------------|
| `JsonOptions` | `JsonSerializerOptions?` | `null` (uses `JsonSerializerOptions.Web`) | JSON serializer options for state serialization/deserialization. Controls formatting, naming policies, etc. |
| `RaceRetries` | `int` | `0` | Number of retry attempts for optimistic concurrency conflicts. Increase when concurrent updates are likely. |
| `IdempotencyMode` | `IdempotencyMode` | `IdempotencyMode.None` | Controls duplicate message detection behavior. |

When the projector loses the optimistic concurrency race and `RaceRetries` is exhausted, it returns `EventHandlingStatus.Failure` for that event.

### Idempotency modes

The `IdempotencyMode` enum controls how the projector handles duplicate messages:

- **`None`** - No idempotency checks. Will process messages and update the blob, without checking for duplicates.
- **`ByGlobalPosition`** - Skips processing if the existing blob has a global position set in its metadata that indicates it has already been processed. The event global position must be greater than that stored in the blob.
- **`ByMessageId`** - Use this when building projections directly from integration events. Skips processing if the message ID in the blob metadata matches that in the event.
Note, this means the idempotency is weaker as only the last message ID is checked. Older messages that are replayed will be processed as normal.

:::caution
`ByGlobalPosition` requires a subscription that provides real global positions, such as an all-stream subscription.
Don't use it with message broker subscriptions where the global position is always zero — the first event would store position `0` and every subsequent event would be treated as a duplicate and silently ignored. Use `ByMessageId` instead.
:::

### Custom blob naming

By default, blob names are generated using `GetBlobName(string id)` which creates names in the format `{id}/{StateType}.json`, where `id` defaults to the stream ID from `context.Stream.GetId()`.

You can customize blob naming in two ways:

**1. Override the virtual methods globally for all events:**

```csharp
public class BookingProjection : BlobStorageProjector<BookingState> {
// ...

protected override string GetBlobName(string id) => $"bookings/{id}.json";
}
```

When the blob name depends on the event, override the overload that takes the consume context instead:

```csharp
protected override string GetBlobName(string id, IMessageConsumeContext context)
=> $"projections/{context.Stream}/{id}.json";
```

The default implementation of the two-argument overload calls the one-argument overload, so overriding the two-argument version replaces the naming completely — a one-argument override is then never called.

**2. Override blob ID per event handler using `getBlobId`:**

```csharp
On<BookingPaymentRegistered>(
(state, evt) => {
state.PaidAmount += evt.AmountPaid;
return state;
},
// Custom blob ID for this specific event only
context => new ValueTask<string>($"payments-{context.Message.BookingId}")
);
```

Note that `getBlobId` returns a blob _ID_, not a full blob name: the result is still passed to `GetBlobName`, so with the default naming the example above produces `payments-{id}/BookingState.json`. To change the full blob path, override `GetBlobName` as well.

Use per-event blob ID overrides when you need different events to target different blobs within the same projector, such as when the business identifier differs from the stream identifier.

## Features

- **Automatic state management** - Creates new state instances when blobs don't exist
- **Optimistic concurrency control** - Uses ETags for safe concurrent updates
- **Idempotency** - Prevents duplicate processing with configurable modes
- **Retry handling** - Automatic retries for race conditions
- **Flexible blob naming** - Customizable blob ID and naming conventions
- **Metadata storage** - Automatically stores stream info, positions, and message IDs

## Background

The projector stores each state as a separate blob in Azure Blob Storage. Each blob contains:

- The serialized state object (JSON by default)
- Metadata including stream name, message ID, stream position, and global position;
because Azure requires metadata values to be ASCII, the stream name and message ID are stored percent-encoded
- Content type set to `application/json`

This approach provides natural partitioning by stream and enables efficient state retrieval for individual streams.
Original file line number Diff line number Diff line change
Expand Up @@ -9,5 +9,6 @@ Eventuous supports the following projection targets:
- [PostgreSQL projections](../../infra/postgres#projections)
- [Microsoft SQL Server projections](../../infra/mssql#projections)
- [SQLite projections](../../infra/sqlite#projections)
- [Azure Blob Storage projections](../../infra/azure-blob-storage)

You can project to any other database using a custom projector, which can be built as a [custom event handler](../../subscriptions/eventhandler#custom-handlers).
Comment on lines +12 to 14

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Informational

1. Missing trailing newline 🐞 Bug ⚙ Maintainability

The modified Markdown file still has no trailing newline, which can cause noisy diffs and formatting
issues in some tooling. Since this PR touched the file, it should be fixed as part of this change.
Agent Prompt
## Issue description
`src/content/docs/dotnet-next/read-models/supported-projectors.md` is missing a trailing newline (`\ No newline at end of file` in the diff). This commonly causes unnecessary future diffs and can trip formatting/lint tooling.

## Issue Context
The PR updates this file to add the Azure Blob Storage projector link, and the patch indicates the file still does not end with a newline.

## Fix Focus Areas
- src/content/docs/dotnet-next/read-models/supported-projectors.md[12-14]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools