A Friendly Introduction to NATS: Real-Time Messaging for .NET Developers

This issue is made possible thanks to JetBrains, who help keep this newsletter free for everyone. A huge shout-out to them for their support of our community. Let's thank them by entering the link below. Struggling with slow builds, tricky bugs, or hard-to-understand performance issues? dotUltimate fixes all of that. It’s the all-in-one toolbox for serious .NET developers. 👉 Upgrade your .NET workflow.

Did you hear about NATS?

Seriously - have you?

Most .NET developers haven’t, and that’s a shame because NATS solves a completely different category of messaging problems than the tools we usually reach for.

When we think “messaging,” the same names pop up:

• RabbitMQ
• Azure Service Bus
• Kafka
• Redis Pub/Sub

They’re all great, and they all have their place...

But they’re not the whole story.

There’s an entire class of systems where those brokers either become too slow, too heavy, or too complex.

And that’s exactly where NATS appears as a surprisingly simple answer.

So What Makes NATS Different?

While traditional brokers focus on features like durability, transactions, partitions, advanced routing, and so on…

NATS focuses on something else entirely:

• 👉 Tiny latency
• 👉 Huge throughput
• 👉 Simple operation
• 👉 Lightweight footprint
• 👉 Scales like crazy

• 👉 Works beautifully with distributed edge devices

  If your system needs:

• thousands of small messages per second
• real-time reactions
• low-latency communication
• services loosely connected with zero overhead
• resilient connections over shaky networks
• tiny agents that can run anywhere

…then NATS suddenly becomes very interesting.

A Different Kind of Problem: Real-Time Smart Agriculture

Let me give you a scenario where NATS actually makes perfect sense.

Imagine a modern agriculture system spread across a huge field.

You have:

• soil moisture sensors
• small climate sensors
• drone stations
• irrigation controllers
• a dashboard
• and some lightweight AI logic that helps make decisions

These sensors send data every few hundred milliseconds.

Some devices drop offline. Some reconnect. Some flood the system with bursts of readings. Some need instructions back immediately (“turn on irrigation now”).

It’s not a big “enterprise event bus” job.

It’s a high-frequency, real-time, lightweight messaging challenge.

And this is exactly the kind of system where RabbitMQ or Kafka start to feel too heavy and too slow.

This Is Where NATS Fits Perfectly

NATS is built for situations like this:

• Where messages are extremely frequent
• Where low latency really matters
• Where devices or services come and go
• Where you need both Pub/Sub and request-reply
• Where you want to scale easily without complexity
• Where you want something simple and fast, not a big distributed monster

If your message volume is small and you want durability or transactions, use RabbitMQ or Service Bus.

If you want big data analytics or huge event streams, Kafka is the king.

But if you need:

• speed
• simplicity
• easy fan-out
• real-time messaging
• millions of events per second

…that’s NATS territory.

Let’s Build Something in .NET

Here’s a realistic example using NATS for that smart agriculture system.

We’ll build:

• sensor simulators
• real-time analytics
• irrigation controllers (queue groups)
• device actuators
• and a request–reply endpoint

1. Install the client

First, install the official .NET client:

1
dotnet add package NATS.Client.Core

What’s happening here?

• NATS.Client.Core is the modern, high-performance NATS client for .NET.

• You’ll use NatsConnection from this package to publish and subscribe to messages.

  2. Creating a Reusable NATS Connection

  You generally want one connection per service, reused everywhere (like HttpClient)

1234567891011121314
public static class NatsConnectionFactory{  public static NatsConnection Create()  {    return new NatsConnection(new NatsOpts    {      Url = "nats://localhost:4222",      Name = "smart-agriculture",      Reconnect = true,      MaxReconnect = 10,      ReconnectWait = TimeSpan.FromSeconds(1),    });  }}

Explanation

• Url - where your NATS server lives (locally, Docker, Kubernetes, cloud, etc.).
• Name - just a human-friendly identifier, helpful in monitoring tools.
• Reconnect = true - NATS will try to reconnect if the connection drops (important for sensors on a weak network).
• MaxReconnect + ReconnectWait - limits and spreads out reconnection attempts so your app doesn’t go crazy if the server is temporarily down. In a real app, you’d typically wrap this in DI as a singleton.

3. Defining the Messages (Contracts)

Let’s define simple message types for:

• soil moisture readings, and
• irrigation commands.

12345678910
public record SoilMoistureReading(  string SensorId,  string FieldId,  double MoisturePercent,  DateTime TimestampUtc);public record IrrigationCommand(  string FieldId,  bool EnableIrrigation,  double TargetMoisturePercent,  DateTime TimestampUtc);

Explanation

• SoilMoistureReading - a single reading from one sensor.
• FieldId - groups sensors by field/zone.
• IrrigationCommand - a decision: “turn irrigation on/off for this field, targeting X% moisture”.

• Records are convenient because:

  1. they’re immutable,
  2. serialize nicely,
  3. and clearly express intent.

4. Simulating a Sensor (Publisher)

Now, let’s simulate a sensor sending readings every 500ms.

123456789101112131415161718192021222324252627282930
public sealed class SoilSensorSimulator{  private NatsConnection _connection;  private readonly string _sensorId;  private readonly string _fieldId;  private readonly Random _random = new();    public SoilSensorSimulator(NatsConnection connection, string sensorId, string fieldId)  {    _connection = connection;    _sensorId = sensorId;    _fieldId = fieldId;  }   public async Task RunAsync(CancellationToken ct)  {    while (!ct.IsCancellationRequested)    {      var reading = new SoilMoistureReading(      SensorId: _sensorId,      FieldId: _fieldId,      MoisturePercent: 15 + _random.NextDouble() * 20, // 15–35%      TimestampUtc: DateTime.UtcNow);      var subject = $"sensors.soil.moisture.{_fieldId}";      await _connection.PublishAsync(subject, reading, cancellationToken: ct);      Console.WriteLine($"[Sensor {_sensorId}] {reading.MoisturePercent:F1}% in field {reading.FieldId}");      await Task.Delay(500, ct);    }  }}

Explanation

• subject is like a topic. Here we use a structured pattern: sensors.soil.moisture.{fieldId}. This makes it easy to subscribe to:

  1. all soil sensors,
  2. or a specific field,
  3. or even a subset using wildcards.
• We send a new reading every 500ms to simulate a real sensor.
• In a real system:

1. this code might run on a device,

2. or you’d have multiple instances for multiple sensors.

   5. Real-Time Analytics (Subscriber with Wildcard)

   Now let’s build a consumer that listens to all soil moisture sensors and logs them.

123456789101112131415161718192021222324252627
using Microsoft.Extensions.Hosting;using NATS.Client.Core;public sealed class SoilAnalyticsWorker : BackgroundService{  private NatsConnection _connection;  public SoilAnalyticsWorker(NatsConnection connection)  {    _connection = connection;  }    protected override async Task ExecuteAsync(CancellationToken stoppingToken)  {    // Subscribe to all soil moisture readings from all fields    await foreach (var msg in _connection.SubscribeAsync<SoilMoistureReading>(    subject: "sensors.soil.moisture.*",    cancellationToken: stoppingToken))    {      var reading = msg.Data;      Console.WriteLine(      $"[Analytics] Field={reading.FieldId}, Sensor={reading.SensorId}, Moisture={reading.MoisturePercent:F1}%");      // Here you could:      // - update an in-memory average      // - push to a time-series DB      // - feed a dashboard, etc.    }  }}

Explanation

• BackgroundService integrates nicely with ASP.NET Core / generic host.

• SubscribeAsync tells NATS to:

  1. subscribe to this subject,
  2. and deserialize payloads into SoilMoistureReading.
• sensors.soil.moisture.* uses a wildcard to listen to all fields.
• We process messages in a simple await foreach loop. When the app stops, the loop ends via the stoppingToken.

6. Irrigation Controller with Queue Groups (Load Balancing)

What if we have multiple instances of an irrigation controller service?

We want only one instance to handle each reading and decide on irrigation, not all of them.

That’s exactly what NATS queue groups do.

12345678910111213141516171819202122232425262728293031323334353637383940
public sealed class IrrigationControllerWorker : BackgroundService{  private NatsConnection _connection;  private const string QueueGroupName = "irrigation-controllers";    public IrrigationControllerWorker(NatsConnection connection)  {    _connection = connection;  }    protected override async Task ExecuteAsync(CancellationToken stoppingToken)  {    await foreach (var msg in    _connection.SubscribeAsync<SoilMoistureReading>(      subject: "sensors.soil.moisture.*",      queueGroup: QueueGroupName,      cancellationToken: stoppingToken))    {      var reading = msg.Data;      Console.WriteLine(      $"[Controller] Field={reading.FieldId}, Moisture={reading.MoisturePercent:F1}%");            if (reading.MoisturePercent < 20)      {        var command = new IrrigationCommand(          FieldId: reading.FieldId,          EnableIrrigation: true,          TargetMoisturePercent: 25,          TimestampUtc: DateTime.UtcNow);         await _connection.PublishAsync(          "irrigation.commands",          command,          stoppingToken);          Console.WriteLine(          $"[Controller] Issued irrigation command for field {reading.FieldId}");      }    }  }}

Explanation

• queueGroup: "irrigation-controllers" makes this subscription part of a queue group.

• If you run 3 instances of this worker:

  1. NATS will load-balance messages across them.
  2. Each message goes to exactly one instance in the group.

• This is perfect for scaling “decision” services horizontally.

  7. Irrigation Device (Command Subscriber)

  Now we simulate a device that receives irrigation commands and acts on them.

123456789101112131415161718192021222324252627
public sealed class IrrigationDeviceWorker : BackgroundService{  private NatsConnection _connection;  private readonly string _deviceId;    public IrrigationDeviceWorker(NatsConnection connection, string deviceId)  {    _connection = connection;    _deviceId = deviceId;  }    protected override async Task ExecuteAsync(CancellationToken stoppingToken)  {    await foreach (var msg in _connection.SubscribeAsync<IrrigationCommand>(      subject: "irrigation.commands",      cancellationToken: stoppingToken))    {      var command = msg.Data;      Console.WriteLine(      $"[Device {_deviceId}] Apply irrigation → Field={command.FieldId}, Enable={command.EnableIrrigation}, Target={command.TargetMoisturePercent}%");      // Here you'd talk to actual hardware:      // - open valve      // - configure timer      // - log to local storage, etc.    }  }}

Explanation

• This worker subscribes to a simple subject: irrigation.commands.
• Commands are broadcast - multiple devices can listen if needed.

• In reality, you might:

  1. use per-device subjects,
  2. or add addressing fields inside the command.

• The core idea: the controller and device are loosely coupled via NATS.

8. Request–Reply: Ask for a Field’s Status

Sometimes you don’t want just a stream - you want to ask the system something:

“What’s the current moisture in Field-42?”

NATS supports this pattern natively with request–reply.

8.1. Define request/response models and cache

1234567891011
public record FieldStatusRequest(string FieldId); public record FieldStatusResponse(    string FieldId,    double AverageMoisturePercent,    DateTime LastUpdatedUtc);     public interface IFieldStatusCache    {      (double AverageMoisture, DateTime LastUpdatedUtc) GetStatusForField(string fieldId);    }

Explanation

• FieldStatusRequest - what the caller sends (just a FieldId).
• FieldStatusResponse - what the responder returns.
• IFieldStatusCache - simple abstraction; your analytics worker could update this cache as readings arrive. 8.2. The responder (service answering status requests)

12345678910111213141516171819202122232425262728
public sealed class FieldStatusResponder : BackgroundService{  private NatsConnection _connection;  private IFieldStatusCache _cache;  public FieldStatusResponder(NatsConnection connection, IFieldStatusCache cache)  {    _connection = connection;    _cache = cache;  }    protected override async Task ExecuteAsync(CancellationToken stoppingToken)  {    await foreach (var msg in      _connection.SubscribeAsync<FieldStatusRequest>(      subject: "fields.status.get",      cancellationToken: stoppingToken))    {      var (average, lastUpdated) = _cache.GetStatusForField(msg.Data.FieldId);            var response = new FieldStatusResponse(        FieldId: msg.Data.FieldId,        AverageMoisturePercent: average,        LastUpdatedUtc: lastUpdated);       await msg.ReplyAsync(response, stoppingToken);    }  }}

Explanation

• We subscribe to fields.status.get.

• For each incoming request:

  1. we look up the field in the cache,
  2. create a response,
  3. and send it back using msg.ReplyAsync.

• NATS handles the internal reply subject and correlation.

8.3. The client (e.g., your ASP.NET Core endpoint)

1234567891011121314151617181920
public sealed class FieldStatusClient{  private NatsConnection _connection;   public FieldStatusClient(NatsConnection connection)  {   _connection = connection;  }   public async Task<FieldStatusResponse?> GetStatusAsync(string fieldId, CancellationToken ct = default)  {    var request = new FieldStatusRequest(fieldId);    var msg = await _connection.RequestAsync<FieldStatusRequest, FieldStatusResponse>(      subject: "fields.status.get",      data: request,      cancellationToken: ct);     return msg.Data;  }}

Explanation

• RequestAsync<TRequest, TResponse> sends a request and waits for the reply.

• Under the hood:

  1. NATS creates a unique reply subject,
  2. the responder replies to it,
  3. the client gets the response mapped to FieldStatusResponse

• This pattern is very handy for internal RPC between services without needing gRPC/HTTP. You could easily expose this in ASP.NET Core:

12345
app.MapGet("/fields/{fieldId}/status", async (string fieldId, FieldStatusClient client, CancellationToken ct) =>{  var status = await client.GetStatusAsync(fieldId, ct);  return status is null ? Results.NotFound() : Results.Ok(status);});

When NATS Makes Sense (And When It Doesn’t)

You don’t have to replace RabbitMQ or Kafka everywhere. Use the right tool for the right job.

NATS is a good fit when:

• You have lots of small, frequent messages
• Latency and responsiveness really matter
• You’re dealing with IoT, sensors, or edge devices
• You want simple, fast, real-time messaging
• You prefer minimal operational overhead
• You like the idea of Pub/Sub, queue groups, and request-reply in one system

NATS is not ideal when:

• You need heavy analytics over huge event streams → Kafka is better
• You rely on advanced broker features (transactions, sessions, dead-lettering, etc.) → RabbitMQ / Azure Service Bus
• You want exactly-once semantics (in practice: Kafka territory) ## Wrapping up

Most developers never think about NATS simply because it doesn’t dominate headlines the way Kafka or RabbitMQ do.

But once you see what kinds of problems it solves, it becomes one of those tools you’re genuinely happy to have discovered.

NATS doesn’t try to be everything. It doesn’t come with a huge learning curve. It doesn’t require a cluster of heavyweight brokers.

It just gives you:

• a ridiculously fast messaging system,
• a tiny operational footprint,
• extremely simple APIs,
• and the flexibility to build real-time systems without complexity.

If your next project involves devices, sensors, real-time updates, automation, or anything where speed and simplicity matter more than heavy enterprise features, give NATS a try.

It might surprise you how far you can go with something this small and this fast.

For a production-ready example of consuming NATS at high speed and broadcasting via SignalR with batch flushing, see High-Throughput Real-Time Data with BoundedChannel and SignalR. For other messaging approaches, check out RabbitMQ in .NET from Scratch and Messaging in .NET with Redis.

That's all from me for today.