dotnet-messaging-patterns

Durable messaging patterns for .NET event-driven architectures. Covers publish/subscribe, competing consumers, dead-letter queues, saga/process manager orchestration, and delivery guarantee strategies using Azure Service Bus, RabbitMQ, and MassTransit.

Scope

• Publish/subscribe and competing consumer patterns
• Dead-letter queues and poison message handling
• Saga/process manager orchestration
• Delivery guarantee strategies (at-least-once, exactly-once)
• Azure Service Bus, RabbitMQ, and MassTransit integration

Out of scope

• Background service lifecycle and IHostedService registration -- see [skill:dotnet-background-services]
• Resilience pipelines and retry policies -- see [skill:dotnet-resilience]
• JSON/binary serialization configuration -- see [skill:dotnet-serialization]
• In-process producer/consumer queues with Channel<T> -- see [skill:dotnet-channels]

Cross-references: [skill:dotnet-background-services] for hosting message consumers, [skill:dotnet-resilience] for fault tolerance around message handlers, [skill:dotnet-serialization] for message envelope serialization, [skill:dotnet-channels] for in-process queuing patterns.

---

Messaging Fundamentals

Message Types

Type	Purpose	Example
Command	Request an action (one recipient)	PlaceOrder, ShipPackage
Event	Notify something happened (many recipients)	OrderPlaced, PaymentReceived
Document	Transfer data between systems	CustomerProfile, ProductCatalog

Commands are sent to a specific queue; events are published to a topic/exchange and delivered to all subscribers. This distinction drives the choice between point-to-point and pub/sub topologies.

Delivery Guarantees

Guarantee	Behavior	Implementation
At-most-once	Fire and forget; message may be lost	No ack, no retry
At-least-once	Message retried until acknowledged; duplicates possible	Ack after processing + retry on failure
Exactly-once	Each message processed exactly once	At-least-once + idempotent consumer

At-least-once with idempotent consumers is the standard approach for durable messaging. True exactly-once requires distributed transactions (which most brokers do not support) or consumer-side deduplication.

---

Publish/Subscribe

Azure Service Bus Topics

// Publisher -- send event to a topic
await using var client = new ServiceBusClient(connectionString);
await using var sender = client.CreateSender("order-events");

var message = new ServiceBusMessage(
    JsonSerializer.SerializeToUtf8Bytes(new OrderPlaced(orderId, total)))
{
    Subject = nameof(OrderPlaced),
    ContentType = "application/json",
    MessageId = Guid.NewGuid().ToString()
};

await sender.SendMessageAsync(message, cancellationToken);

```text

```csharp

// Subscriber -- process events from a subscription
await using var processor = client.CreateProcessor(
    topicName: "order-events",
    subscriptionName: "billing-service",
    new ServiceBusProcessorOptions
    {
        MaxConcurrentCalls = 10,
        AutoCompleteMessages = false
    });

processor.ProcessMessageAsync += async args =>
{
    var body = args.Message.Body.ToObjectFromJson<OrderPlaced>();
    await HandleOrderPlacedAsync(body);
    await args.CompleteMessageAsync(args.Message);
};

processor.ProcessErrorAsync += args =>
{
    logger.LogError(args.Exception, "Error processing message");
    return Task.CompletedTask;
};

await processor.StartProcessingAsync(cancellationToken);

```text

**Key packages:**

```xml

<PackageReference Include="Azure.Messaging.ServiceBus" Version="7.*" />

```xml

### RabbitMQ Fanout Exchange

```csharp

// Publisher -- declare exchange and publish
var factory = new ConnectionFactory { HostName = "localhost" };
await using var connection = await factory.CreateConnectionAsync();
await using var channel = await connection.CreateChannelAsync();

await channel.ExchangeDeclareAsync(
    exchange: "order-events",
    type: ExchangeType.Fanout,
    durable: true);

var body = JsonSerializer.SerializeToUtf8Bytes(
    new OrderPlaced(orderId, total));

await channel.BasicPublishAsync(
    exchange: "order-events",
    routingKey: string.Empty,
    body: body);

```text

**Key packages:**

```xml

<PackageReference Include="RabbitMQ.Client" Version="7.*" />

```xml

### MassTransit Publish

MassTransit abstracts the broker, providing a unified API for Azure Service Bus, RabbitMQ, Amazon SQS, and in-memory
transport.

```csharp

// Registration
builder.Services.AddMassTransit(x =>
{
    x.AddConsumer<OrderPlacedConsumer>();

    x.UsingRabbitMq((context, cfg) =>
    {
        cfg.Host("localhost", "/", h =>
        {
            h.Username("guest");
            h.Password("guest");
        });
        cfg.ConfigureEndpoints(context);
    });
});

// Publisher
public sealed class OrderService(IPublishEndpoint publishEndpoint)
{
    public async Task PlaceOrderAsync(
        Guid orderId, decimal total, CancellationToken ct)
    {
        // Process order...
        await publishEndpoint.Publish(
            new OrderPlaced(orderId, total), ct);
    }
}

// Consumer
public sealed class OrderPlacedConsumer(
    ILogger<OrderPlacedConsumer> logger)
    : IConsumer<OrderPlaced>
{
    public async Task Consume(ConsumeContext<OrderPlaced> context)
    {
        logger.LogInformation(
            "Processing order {OrderId}", context.Message.OrderId);
        await ProcessAsync(context.Message);
    }
}

// Message contract (use records in a shared contracts assembly)
public record OrderPlaced(Guid OrderId, decimal Total);

```text

**Key packages:**

```xml

<PackageReference Include="MassTransit" Version="8.*" />
<!-- Pick ONE transport: -->
<PackageReference Include="MassTransit.RabbitMQ" Version="8.*" />
<!-- OR -->
<PackageReference Include="MassTransit.Azure.ServiceBus.Core" Version="8.*" />

```text

---

## Competing Consumers

Multiple consumer instances process messages from the same queue in parallel. The broker delivers each message to
exactly one consumer, distributing load across instances.

### Pattern

```text

Queue: order-processing
  ├── Consumer Instance A  (picks message 1)
  ├── Consumer Instance B  (picks message 2)
  └── Consumer Instance C  (picks message 3)

```text

### Azure Service Bus -- Scaling Consumers

```csharp

// Multiple instances reading from the same queue automatically compete.
// MaxConcurrentCalls controls per-instance parallelism.
var processor = client.CreateProcessor("order-processing",
    new ServiceBusProcessorOptions
    {
        MaxConcurrentCalls = 20,
        PrefetchCount = 50,
        AutoCompleteMessages = false
    });

```text

### MassTransit -- Concurrency Limits

```csharp

x.AddConsumer<OrderProcessor>(cfg =>
{
    cfg.UseConcurrentMessageLimit(10);
});

```text

### Ordering Considerations

Competing consumers sacrifice strict ordering for throughput. When order matters:

- **Azure Service Bus**: Use sessions (`RequiresSession = true`) to guarantee FIFO within a session ID (e.g., per
  customer)
- **RabbitMQ**: Use a single consumer per queue, or consistent-hash exchange to partition by key
- **MassTransit**: Configure `UseMessagePartitioner` for key-based ordering

---

## Dead-Letter Queues

Dead-letter queues (DLQs) capture messages that cannot be processed after exhausting retries. They prevent poison
messages from blocking the main queue.

### Why Messages Are Dead-Lettered

| Reason                         | Trigger                                      |
| ------------------------------ | -------------------------------------------- |
| Max delivery attempts exceeded | Message failed processing N times            |
| TTL expired                    | Message sat in queue past its time-to-live   |
| Consumer rejection             | Consumer explicitly dead-letters the message |
| Queue length exceeded          | Queue overflow policy routes to DLQ          |

### Azure Service Bus DLQ

```csharp

// Dead-letter a message with reason
await args.DeadLetterMessageAsync(
    args.Message,
    deadLetterReason: "ValidationFailed",
    deadLetterErrorDescription: "Missing required field: CustomerId");

// Read from the dead-letter sub-queue
await using var dlqReceiver = client.CreateReceiver(
    "order-processing",
    new ServiceBusReceiverOptions
    {
        SubQueue = SubQueue.DeadLetter
    });

while (true)
{
    var message = await dlqReceiver.ReceiveMessageAsync(
        TimeSpan.FromSeconds(5), cancellationToken);
    if (message is null) break;

    logger.LogWarning(
        "DLQ message: {Reason} - {Description}",
        message.DeadLetterReason,
        message.DeadLetterErrorDescription);

    // Inspect, fix, and re-submit or discard
    await dlqReceiver.CompleteMessageAsync(message);
}

```text

### MassTransit Error/Fault Queues

MassTransit automatically creates `_error` and `_skipped` queues. Failed messages after retry exhaustion move to the
error queue with fault metadata.

```csharp

// Configure retry before dead-lettering
x.AddConsumer<OrderProcessor>(cfg =>
{
    cfg.UseMessageRetry(r => r.Intervals(
        TimeSpan.FromSeconds(1),
        TimeSpan.FromSeconds(5),
        TimeSpan.FromSeconds(15)));
});

```text

### DLQ Monitoring

Always monitor DLQ depth with alerts. Unmonitored DLQs accumulate silently until data is lost or stale.

---

## Saga / Process Manager

Sagas coordinate multi-step business processes across services. Each step publishes events that trigger the next step,
with compensation logic for failures.

### Choreography vs Orchestration

| Style             | How it works                                                   | Use when                                                |
| ----------------- | -------------------------------------------------------------- | ------------------------------------------------------- |
| **Choreography**  | Services react to events independently; no central coordinator | Simple flows, few steps, loosely coupled                |
| **Orchestration** | A saga/process manager directs each step                       | Complex flows, compensation needed, visibility required |

### MassTransit State Machine Saga

```csharp

// Saga state
public class OrderState : SagaStateMachineInstance
{
    public Guid CorrelationId { get; set; }
    public string CurrentState { get; set; } = default!;
    public Guid OrderId { get; set; }
    public decimal Total { get; set; }
    public DateTime? PaymentReceivedAt { get; set; }
}

// State machine definition
public sealed class OrderStateMachine : MassTransitStateMachine<OrderState>
{
    public State Submitted { get; private set; } = default!;
    public State PaymentPending { get; private set; } = default!;
    public State Completed { get; private set; } = default!;
    public State Faulted { get; private set; } = default!;

    public Event<OrderSubmitted> OrderSubmitted { get; private set; } = default!;
    public Event<PaymentReceived> PaymentReceived { get; private set; } = default!;
    public Event<PaymentFailed> PaymentFailed { get; private set; } = default!;

    public OrderStateMachine()
    {
        InstanceState(x => x.CurrentState);

        Event(() => OrderSubmitted,
            x => x.CorrelateById(ctx => ctx.Message.OrderId));
        Event(() => PaymentReceived,
            x => x.CorrelateById(ctx => ctx.Message.OrderId));
        Event(() => PaymentFailed,
            x => x.CorrelateById(ctx => ctx.Message.OrderId));

        Initially(
            When(OrderSubmitted)
                .Then(ctx =>
                {
                    ctx.Saga.OrderId = ctx.Message.OrderId;
                    ctx.Saga.Total = ctx.Message.Total;
                })
                .Publish(ctx => new RequestPayment(
                    ctx.Saga.OrderId, ctx.Saga.Total))
                .TransitionTo(PaymentPending));

        During(PaymentPending,
            When(PaymentReceived)
                .Then(ctx =>
                    ctx.Saga.PaymentReceivedAt = DateTime.UtcNow)
                .Publish(ctx => new FulfillOrder(ctx.Saga.OrderId))
                .TransitionTo(Completed),
            When(PaymentFailed)
                .Publish(ctx => new CancelOrder(ctx.Saga.OrderId))
                .TransitionTo(Faulted));
    }
}

// Registration -- requires MassTransit.EntityFrameworkCore package for EF persistence
// NuGet: MassTransit.EntityFrameworkCore Version="8.*"
builder.Services.AddMassTransit(x =>
{
    x.AddSagaStateMachine<OrderStateMachine, OrderState>()
        .EntityFrameworkRepository(r =>
        {
            r.ExistingDbContext<SagaDbContext>();
            r.UsePostgres();
        });

    x.UsingRabbitMq((context, cfg) =>
    {
        cfg.ConfigureEndpoints(context);
    });
});

```text

### Saga Persistence

| Store                 | Package                           | Use when                                 |
| --------------------- | --------------------------------- | ---------------------------------------- |
| Entity Framework Core | `MassTransit.EntityFrameworkCore` | Already using EF Core; need transactions |
| MongoDB               | `MassTransit.MongoDb`             | Document-oriented state; high throughput |
| Redis                 | `MassTransit.Redis`               | Ephemeral sagas; low latency             |
| In-Memory             | Built-in                          | Testing only -- state lost on restart    |

### Compensation Pattern

When a saga step fails, publish compensating commands to undo prior steps:

```bash

OrderSubmitted -> RequestPayment -> PaymentReceived -> ReserveInventory
                                                          |
                                                     InventoryFailed
                                                          |
                                                    RefundPayment (compensation)
                                                          |
                                                    CancelOrder (compensation)

```text

---

## Idempotent Consumers

At-least-once delivery means consumers may receive the same message multiple times. Idempotent consumers ensure repeated
processing produces the same result.

### Database-Based Deduplication

```csharp

public sealed class IdempotentOrderConsumer(
    AppDbContext db,
    ILogger<IdempotentOrderConsumer> logger)
    : IConsumer<OrderPlaced>
{
    public async Task Consume(ConsumeContext<OrderPlaced> context)
    {
        var messageId = context.MessageId
            ?? throw new InvalidOperationException("Missing MessageId");

        // Check if already processed
        var exists = await db.ProcessedMessages
            .AnyAsync(m => m.MessageId == messageId);

        if (exists)
        {
            logger.LogInformation(
                "Duplicate message {MessageId}, skipping", messageId);
            return;
        }

        // Process the message
        await ProcessOrderAsync(context.Message);

        // Record as processed
        db.ProcessedMessages.Add(new ProcessedMessage
        {
            MessageId = messageId,
            ProcessedAt = DateTime.UtcNow,
            ConsumerType = nameof(IdempotentOrderConsumer)
        });

        await db.SaveChangesAsync();
    }
}

```text

### Natural Idempotency

Prefer operations that are naturally idempotent:

- **Upserts** (`INSERT ... ON CONFLICT UPDATE`) instead of blind inserts
- **Conditional updates** (`UPDATE ... WHERE Status = 'Pending'`) instead of unconditional
- **Deterministic IDs** derived from message content instead of auto-generated

---

## Message Envelope Pattern

Wrap message payloads in a standard envelope with metadata for tracing, versioning, and routing.

```csharp

public sealed record MessageEnvelope<T>(
    string MessageId,
    string MessageType,
    DateTimeOffset Timestamp,
    string CorrelationId,
    string Source,
    int Version, // Schema version for backward-compatible deserialization
    T Payload);

```text

MassTransit provides this automatically via `ConsumeContext` (MessageId, CorrelationId, Headers). When using raw broker
clients, implement envelopes explicitly.

---

## Agent Gotchas

1. **Do not use auto-complete with Azure Service Bus** -- set `AutoCompleteMessages = false` and call
   `CompleteMessageAsync` after successful processing. Auto-complete acknowledges before processing finishes, risking
   data loss on failure.
2. **Do not forget to handle poison messages** -- always configure max delivery count and DLQ monitoring. Without these,
   a single bad message blocks the entire queue indefinitely.
3. **Do not use in-memory saga persistence in production** -- saga state is lost on restart, leaving business processes
   in unknown states. Use Entity Framework, MongoDB, or Redis persistence.
4. **Do not assume message ordering across partitions** -- competing consumers and topic subscriptions deliver messages
   out of order by default. Use sessions or partitioning when order matters.
5. **Do not skip idempotency for at-least-once consumers** -- brokers may redeliver on timeout, network glitch, or
   consumer restart. Every consumer must handle duplicate messages safely.
6. **Do not hardcode connection strings** -- use environment variables or Azure Key Vault references. For local
   development, use user secrets or `.env` files excluded from source control.

---

## References

- [Azure Service Bus documentation](https://learn.microsoft.com/en-us/azure/service-bus-messaging/)
- [Azure Service Bus client library for .NET](https://learn.microsoft.com/en-us/dotnet/api/overview/azure/messaging.servicebus-readme)
- [RabbitMQ .NET client documentation](https://www.rabbitmq.com/client-libraries/dotnet-api-guide)
- [MassTransit documentation](https://masstransit.io/documentation/concepts)
- [MassTransit sagas](https://masstransit.io/documentation/patterns/saga)
- [Enterprise Integration Patterns](https://www.enterpriseintegrationpatterns.com/)

Code Navigation (Serena MCP)

Primary approach: Use Serena symbol operations for efficient code navigation:

1. Find definitions: serena_find_symbol instead of text search
2. Understand structure: serena_get_symbols_overview for file organization
3. Track references: serena_find_referencing_symbols for impact analysis
4. Precise edits: serena_replace_symbol_body for clean modifications

When to use Serena vs traditional tools:

• Use Serena: Navigation, refactoring, dependency analysis, precise edits
• Use Read/Grep: Reading full files, pattern matching, simple text operations
• Fallback: If Serena unavailable, traditional tools work fine

Example workflow:

# Instead of:
Read: src/Services/OrderService.cs
Grep: "public void ProcessOrder"

# Use:
serena_find_symbol: "OrderService/ProcessOrder"
serena_get_symbols_overview: "src/Services/OrderService.cs"