Those using (or learning about) Microsoft Orleans, especially the newer 2.0.x releases that target .NET Standard and are cross-platform, might be interested to know that version 2.0.4 has just been released.

This release includes a couple of important bugfixes:

• A number of Orleans users observed grain calls getting really slow after the silo has been running for around 12 hours. The long issue discussion reveals a lot of collective findings and ultimately provides the means to reproduce the problem. The root cause was traced to a bug in BlockingCollection<T> in .NET Core, which can lead to memory leaks and even lost items (Orleans messages in this case). A workaround has been implemented to sort this out.
• Another issue prevented Orleans build-time code generation from being built when targeting .NET Core 2.1. This has also been fixed.

If you’re using Orleans 2.0.x, it’s therefore a good idea to upgrade to 2.0.4, especially if you are running Orleans in production.

Microsoft Orleans is an implementation of the actor model, and many people have leveraged it to build highly scalable distributed systems while completely avoiding the pain of multithreaded programming.

The actor model is still not a very mainstream thing, and people who come across it are often confused about what it is and why it is useful.

In order to address this, .NET contractor Jakub Konecki (Twitter | GitHub) has kindly agreed to share with us how he has been using Microsoft Orleans in his own particular use case. You can also learn more about his project from the Orleans Virtual Meetups in which he presented (Meetup #1: Event Sourced Grains, and  Meetup #12: Deploying Orleans).

DD: What is the problem you’re addressing with Microsoft Orleans?

JK: Currently I’m working for a company in the hospitality domain that manages bookings for a number of luxury resorts in the Caribbean.

I’m responsible for designing and delivering a greenfield system for a multi-tenant system for managing reservations.

The main features of the system are:

• ability to register resort accommodation,
• ability to manage pricing: rate plans, special offers,
• integration with third-party marketplace used by tour operators
• integration with third-party systems for flight searching and ticket purchasing
• integration with property management systems used by resorts
• a bespoke website that resort customers can use to search for and make their reservations.

The non-functional requirements include elastic scaling to allow for easy onboarding of new tenants and allow flexibility for existing tenants – for example the traffic may change drastically when special offers are introduced.

DD: How did Microsoft Orleans help you develop a solution?

The system is designed using DDD principles and benefits from event sourcing and event-driven architecture.

An actor framework is a good fit for this kind of system – mapping between actors and aggregate roots is natural, and implementation of event sourcing is quite straightforward and encapsulated by actors.

The Microsoft Orleans framework was selected as it was the most advanced actor framework implementation at the time that used technologies familier to the team (C#, Azure), was battle-tested, and was implicitly backed by Microsoft. Open-sourcing Orleans (and an active community that emerged shortly after) was another argument for using it.

DD: What benefits did Microsoft Orleans provide, and what challenges did you face?

JK: The most important benefits for using Orleans are scalability and programming model. Scaling an Orleans solution is as easy as moving a slider in the Azure portal. The ability to specify auto-scaling triggers in Azure means that changes in load can be handled with ease. We haven’t run into any problems related to scaling – there is no difference between running a cluster in Azure and a single node on local machine during development.

This brings me nicely to Orleans’ programming model, which makes development of distributed systems straightforward. Orleans handles a lot of complexity allowing developers to concentrate on business logic within essentially single-threaded grains. On the other hand Orleans doesn’t go to the other extreme and pretend the issues inherently related to distributed systems do not exist. That balance allows for rapid development – we’ve seen senior developers being able to pick up Orleans fundamentals and be productive within a day or two.

In this article, we’ll see how to create grains that automatically scale up and down depending on load, in Microsoft Orleans 2.0.

The source code for this article is very similar to that in “Getting Started with Microsoft Orleans 2.0 in .NET Core“, with a few key differences:

• It has been modified to gracefully stop the silo and gracefully close the client.
• It uses the latest packages at the time of writing this article – Orleans 2.0.3 and OrleansDashboard 2.0.7.
• It uses a slightly different example, and the load generation has been adapted accordingly.

Since there’s nothing really new in the client and silo setup, we’ll be focusing mainly on the grain and load generation parts. However, you may find the full source code for this article in the Orleans2StatelessWorkers folder in the Gigi Labs BitBucket repository.

Example Grain

For the sake of example, we’ll imagine that the job of our Orleans cluster is to provide hashing as a service. A client provides an input string, and we’ll have a grain that computes a hash of the string (it doesn’t really matter what hash function it is – we’ll use MD5 in the example) and returns it.

Based on this requirement, we can easily write a grain and its corresponding interface to perform the hash calculation:

    public interface IHashGeneratorGrain : IGrainWithIntegerKey
    {
        Task<string> GenerateHashAsync(string input);
    }

    public class HashGeneratorGrain : Grain, IHashGeneratorGrain
    {
        private HashAlgorithm hashAlgorithm;

        public HashGeneratorGrain()
        {
            this.hashAlgorithm = MD5.Create();
        }

        public Task<string> GenerateHashAsync(string input)
        {
            var inputBytes = Encoding.UTF8.GetBytes(input);
            var hashBytes = hashAlgorithm.ComputeHash(inputBytes);
            var hashBase64Str = Convert.ToBase64String(hashBytes);

            return Task.FromResult(hashBase64Str);
        }
    }

Load Generation

Typically, when we talk about actor models, the whole point is to have an instance of an actor (grain in Orleans) per entity ID. For instance, you’d have a grain instance for each Device, Vehicle, BlogPost, Game, User, or whatever other domain object you’re dealing with. In this case, however, our grain is completely stateless, and there is no difference in behaviour between one activation and another. In fact, since the grain ID doesn’t matter, we can just pass in 0 as a sort of convention when requesting a grain of this kind:

var hashGenerator = client.GetGrain<IHashGeneratorGrain>(0);

Once we have an instance of the grain, we can generate some load by creating random strings and invoking the relevant method on the grain repeatedly:

            while (true)
            {
                var randomString = GenerateRandomString();
                var hash = await hashGenerator.GenerateHashAsync(randomString);
                Console.WriteLine(hash);
            }

You can monitor the grain’s activity from the Orleans Dashboard (localhost:8080 by default), and as you’d expect, there is only one activation of the grain:

Stateless Worker Grains

This situation is a very good fit for Stateless Worker Grains.

Normally, when you request a grain with a particular ID, you get a single activation – and it is a singleton throughout the cluster, so you would never (bar edge cases involving failover scenarios) get more than one instance of that grain in the cluster. However, if you just add a [StatelessWorker] attribute on the grain…

    [StatelessWorker]
    public class HashGeneratorGrain : Grain, IHashGeneratorGrain

…you’ll see very different behaviour:

Notice how there are now two activations of the HashGeneratorGrain, even though we’re still requesting an instance with ID 0.

When Orleans sees the [StatelessWorker] attribute, it will create a pool of grains behind the ID you specify. This is similar to a load balancer. Those grains are hidden behind that same ID, so you can’t access individual grains in the pool directly (it wouldn’t make any sense to do that). The number of grains will grow up to as many CPU cores are available on the machine, unless you pass an argument to the attribute specifying otherwise.

Aside from autoscaling, another important benefit of stateless worker grains is that they are always local. Orleans will always execute a request to a stateless worker on the same silo where the request was generated, spawning a new activation if necessary. This saves the overhead of potentially passing the request to an instance in a different silo (i.e. remote call), which makes a lot of sense for stateless workers that are pure logic and there’s no difference between activations running in different places.

Although stateless worker grains are best used for stateless logic (as one would expect), there is nothing preventing their use with state. However, coordination of state between multiple grain activations with the same ID can be complicated. The Stateless Worker Grains documentation describes some patterns where stateless worker grains with state make sense (although calling them that way is bizarre).

Summary

• Use the [StatelessWorker] attribute to treat a grain as a stateless worker grain.
• This creates a load-balanced autoscaling pool of grains with the same ID.
• Requests to stateless worker grains are always local and never incur a remote call.
• Stateless worker grains may have state, although this is unusual.

Dependency Injection (DI) has become a cornerstone of any well-designed and testable application nowadays, and Microsoft Orleans applications are no exception. In the 2.0 release, Microsoft Orleans has replaced some of its old internal frameworks (such as logging and dependency injection) with the corresponding Microsoft packages; thus these will be familiar for those who already worked with ASP .NET Core.

In this article we’ll focus on setting up dependency injection in the silo so that we can pass dependencies into our grains. However, if you read the dependency injection documentation page for Orleans 2.0, you’ll see that you can also have DI on the client side.

The source code for this article is the Orleans2DependencyInjection folder at the Gigi Labs BitBucket repository. Be careful not to confuse it with the OrleansDependencyInjection folder which targets Orleans 1.4.x.

TL;DR if you just want to quickly see how to do DI without going through the whole example, jump to the Registering Dependencies section.

Update 30th June 2018: The source code for this article needs a little adjusting, in order to gracefully stop the silo and gracefully close the client. Counterintuitively, directly disposing a silo or client is non-graceful and is generally discouraged.

Grain

We’ll start off with a project structure based on the Getting Organised article. Once that is in place, we can build an example representing a blog’s comment system. In the Grains project, we’ll add a grain representing a blog post, and that will be responsible for saving and retrieving all comments for that blog post.

    public class BlogPostGrain : Grain, IBlogPostGrain
    {
        private ICommentRepository repo;
        private ITimeService time;

        public BlogPostGrain(ICommentRepository repo, ITimeService time)
        {
            this.repo = repo;
            this.time = time;
        }

        public Task SaveCommentAsync(int blogPostId, InputComment comment)
        {
            var storedComment = new StoredComment()
            {
                Name = comment.Name,
                EmailAddress = comment.EmailAddress,
                Body = comment.Body,
                Timestamp = this.time.UtcNow
            };

            return this.repo.SaveCommentAsync(blogPostId, storedComment);
        }

        public Task<List<StoredComment>> GetCommentsAsync(int blogPostId)
            => this.repo.GetCommentsAsync(blogPostId);
    }

There are a few classes and interfaces in here that we haven’t created yet, but let’s understand what we’re doing here. We have a dependency on a repository where the comments will be held (whatever that is – we don’t care about the implementation at this stage). The grain acts mostly as a pass-through to this repository for storage and retrieval of comments, but when saving, we transform it by adding a timestamp. We use different DTOs for input comments and stored comments so that it is not possible to supply a timestamp with the input data.

We also have a second dependency on something called a time service. While you could just use DateTime.UtcNow in your code, time is typically one of the dependencies you want to factor out of your unit tests because it can affect the results. So we wrap DateTime.UtcNow in something we can mock, just for the sake of unit tests later.

Contracts

In the Contracts project, we’ll add all our interfaces and DTOs. Let’s start with our dependencies:

    public interface ITimeService
    {
        DateTime UtcNow { get; }
    }

    public interface ICommentRepository
    {
        Task SaveCommentAsync(int blogPostId, StoredComment comment);
        Task<List<StoredComment>> GetCommentsAsync(int blogPostId);
    }

Then we have our grain interface:

    public interface IBlogPostGrain : IGrainWithIntegerKey
    {
        Task SaveCommentAsync(int blogPostId, InputComment comment);
        Task<List<StoredComment>> GetCommentsAsync(int blogPostId);
    }

And finally our DTOs:

    public class InputComment
    {
        public string Name { get; set; }
        public string EmailAddress { get; set; }
        public string Body { get; set; }
    }

    public class StoredComment : InputComment
    {
        public DateTime Timestamp { get; set; }
    }

Dependency Implementations

In the Silo, we can create implementations for our dependencies.

To keep it simple, we’ll implement our repository using a ConcurrentDictionary. This is a volatile, in-memory implementation that is for demonstration only, but it allows us to focus on what we’re doing with Orleans, rather than distracting us with store-specific details.

Note: We could also use Orleans storage providers, but that’s out of scope here.

    public class MemoryCommentRepository : ICommentRepository
    {
        private ConcurrentDictionary<int, List<StoredComment>> dict;

        public MemoryCommentRepository()
        {
            this.dict = new ConcurrentDictionary<int, List<StoredComment>>();
        }

        public Task<List<StoredComment>> GetCommentsAsync(int blogPostId)
        {
            this.dict.TryGetValue(blogPostId, out var comments);
            return Task.FromResult(comments);
        }

        public Task SaveCommentAsync(int blogPostId, StoredComment comment)
        {
            this.dict.AddOrUpdate(blogPostId,
                addValue: new List<StoredComment>() { comment },
                updateValueFactory: (postId, commentsList) => {
                    commentsList.Add(comment);
                    return commentsList;
                });

            return Task.CompletedTask;
        }
    }

The time service is really simple: it just wraps DateTime.UtcNow.

    public class TimeService : ITimeService
    {
        public DateTime UtcNow => DateTime.UtcNow;
    }

Registering Dependencies

All the above was setting up the example, and now we get to the part we’ve all been waiting for.

We’ll set up our silo’s code similarly to what we’ve done in the past two articles, but this time, we’ll add a call to ConfigureServices() in order to register our dependencies:

            var siloBuilder = new SiloHostBuilder()
                .UseLocalhostClustering()
                .UseDashboard(options => { })
                .Configure<ClusterOptions>(options =>
                {
                    options.ClusterId = "dev";
                    options.ServiceId = "Orleans2DependencyInjection";
                })
                .Configure<EndpointOptions>(options =>
                    options.AdvertisedIPAddress = IPAddress.Loopback)
                .ConfigureServices(services =>
                {
                    services.AddSingleton<ITimeService, TimeService>();
                    services.AddSingleton<ICommentRepository, MemoryCommentRepository>();
                })
                .ConfigureLogging(logging => logging.AddConsole());

Note: as per the previous articles, C# 7.1 or above is needed in order to allow async/await in Main().

Since AddSingleton() is an extension method coming from Mirosoft.Extensions.DependencyInjection (already included as a dependency of Microsoft.Orleans.Core), you’ll need to add the following for this to work:

using Microsoft.Extensions.DependencyInjection;

The API

We can complete this example by exposing the grain’s functionality via our Web API. For this, we’ll add the following controller:

    [Produces("application/json")]
    [Route("api/BlogPosts")]
    public class BlogPostsController : Controller
    {
        private IClusterClient orleansClient;

        public BlogPostsController(IClusterClient orleansClient)
        {
            this.orleansClient = orleansClient;
        }

        [HttpGet]
        public Task<List<StoredComment>> Get(int blogPostId)
        {
            var grain = this.orleansClient.GetGrain<IBlogPostGrain>(blogPostId);
            return grain.GetCommentsAsync(blogPostId);
        }

        [HttpPut]
        public async Task Put(int blogPostId, InputComment comment)
        {
            var grain = this.orleansClient.GetGrain<IBlogPostGrain>(blogPostId);
            await grain.SaveCommentAsync(blogPostId, comment);
        }
    }

 

Note: as I write this, I am noticing a quirk in this implementation. If you get a grain with a blogPostId, then why do you have to pass it again to call the method on the grain? The grain should know its ID already. Fair enough – that was an oversight on my part. But since grain IDs are retrieved using extension methods, and thus their retrieval would also need to be mocked, I’d rather not overcomplicate things in this example.

We can then add Swagger to the Web API and wire up the Orleans client as we did in the Getting Organised article (complete with retries):

       private IClusterClient CreateOrleansClient()
        {
            var clientBuilder = new ClientBuilder()
                .UseLocalhostClustering()
                .Configure<ClusterOptions>(options =>
                {
                    options.ClusterId = "dev";
                    options.ServiceId = "Orleans2DependencyInjection";
                })
                .ConfigureLogging(logging => logging.AddConsole());

            var client = clientBuilder.Build();

            client.Connect(async ex =>
            {
                Console.WriteLine("Retrying...");
                await Task.Delay(3000);
                return true;
            }).Wait();

            return client;
        }

        // This method gets called by the runtime. Use this method to add services to the container.
        public void ConfigureServices(IServiceCollection services)
        {
            var orleansClient = CreateOrleansClient();
            services.AddSingleton<IClusterClient>(orleansClient);

            services.AddSwaggerGen(c =>
            {
                c.SwaggerDoc("v1", new Info { Title = "My API", Version = "v1" });
            });

            services.AddMvc();
        }

Manual Testing with Swagger

We can quickly add a couple of comments on a blog post and retrieve them to see that all this is working:

Note: seems like Swagger recently changed their UI. I liked it a lot better before.

Unit Testing

Dependency injection makes it easy for us to write unit tests. Let’s add a Grains.Tests project (.NET Core Console App), add a reference to the Grains project, and install the following packages:

Install-Package Microsoft.NET.Test.Sdk
Install-Package NUnit
Install-Package NUnit3TestAdapter
Install-Package Moq

Remove the auto-generated Program.cs file and add the following test class instead:

    public class BlogPostGrainTests
    {
        [Test]
        public async Task SaveCommentTest()
        {
            // arrange

            const int blogPostId = 1;

            var fixedDateTime = new DateTime(2018, 4, 29, 18, 28, 33, DateTimeKind.Utc);
            var mockRepo = new Mock<ICommentRepository>(MockBehavior.Strict);
            var mockTimeService = new Mock<ITimeService>(MockBehavior.Strict);

            mockRepo.Setup(x => x.SaveCommentAsync(blogPostId, It.IsAny<StoredComment>()))
                    .Returns(Task.CompletedTask);
            mockTimeService.Setup(x => x.UtcNow)
                           .Returns(fixedDateTime);

            var grain = new BlogPostGrain(mockRepo.Object, mockTimeService.Object);

            const string name = "George";
            const string emailAddress = "george@food.com";
            const string body = "I'm hungry!";

            var comment = new InputComment()
            {
                Name = name,
                EmailAddress = emailAddress,
                Body = body
            };

            // act

            await grain.SaveCommentAsync(blogPostId, comment);

            // assert

            mockRepo.Verify(x => x.SaveCommentAsync(blogPostId, It.Is<StoredComment>(
                c => c.Name == name
                  && c.EmailAddress == emailAddress
                  && c.Body == body
                  && c.Timestamp == fixedDateTime
            )));
        }
    }

This test verifies that the submitted comment was passed on to the store with the generated timestamp. It should pass:

Exercises

We’ve seen a complete example featuring dependency injection. Registering dependencies is easy; most of the effort in this article was around building the example to demonstrate that.

As you can see, you can write unit tests for grains just as you would for any other class, without having to resort to the Orleans TestCluster.

There are a number of ways you can take this further:

1. Have the grain perform a validation against the email address, and write unit tests for that.
2. Have the grain retrieve its own ID (removing the need to pass it as a parameter to its methods), and find a way to mock the grain retrieval.
3. Try dependency injection in the Orleans client.

In the previous article, “Getting Started with Microsoft Orleans 2.0 in .NET Core“, we saw how to quickly set up a minimal Orleans 2.0 silo and client (in the same application) and run it on Linux thanks to .NET Core.

However, if you’re serious about using Microsoft Orleans in a production environment, your setup won’t be this simple. You’ll need to create an appropriate project structure, introduce reliability, and add certain optimisations. We’ll be covering these in this article. You’ll also want to look into things like clustering providers which are out of scope here.

The source code for this article is the Orleans2GettingOrganised folder in the Gigi Labs BitBucket repository.

Update 30th June 2018: The source code for this article needs a little adjusting, in order to gracefully stop the silo and gracefully close the client. Counterintuitively, directly disposing a silo or client is non-graceful and is generally discouraged.

General Architecture

Before we go on, it is important to understand what the typical components in an Orleans solution look like.

An Orleans cluster consists of a number of silos, which in turn host a number of grains. Part of the Orleans abstraction is that you don’t know where your grains are physically running; this forces you to think distributed first, and also allows Orleans to migrate grains from faulted silos onto healthy ones.

Note: You can run a single-silo cluster, but that would be a single point of failure. You need multiple silos to achieve high availability. A single-silo cluster is typically only used for development and testing.

An Orleans client is used as a gateway between the Orleans cluster and the outside world. The name is actually misleading, because while it is a client to the Orleans cluster, it is typically also a server to external requests. For example, the Orleans client could be a REST API that accepts HTTP requests and interacts with grains in the Orleans cluster accordingly. Or it could be a Console App running as a Windows service with Topshelf. The project type is arbitrary.

Project Structure

The projects in an Orleans 2.0 solution should look something like this:

Purpose	Type	NuGet package	References
Client	ASP .NET Core Web API or Console App etc.	Microsoft.Orleans.Client	Contracts
Silo	Console App	Microsoft.Orleans.Server	Contracts, Grains
Grains	Class Library	Microsoft.Orleans.Core.Abstractions Microsoft.Orleans.OrleansCodeGenerator.Build	Contracts
Contracts (i.e. Interfaces)	Class Library	Microsoft.Orleans.Core.Abstractions Microsoft.Orleans.OrleansCodeGenerator.Build

Instead of clicking through Visual Studio to set this all up every time, we can use the dotnet command to automate this setup. This not only allows us to build this project structure quickly next time, but allows us to set this up on other platforms (e.g. Linux) in an IDE-agnostic manner.

We’ll use the --no-restore switch to prevent restoring packages with every command, which would take ages. We can do a separate dotnet restore at the end once everything is set up.

First, let’s make a folder for the solution:

mkdir Orleans2
cd Orleans2

Set up the Contracts project, which will hold our grain interfaces:

dotnet new classlib --name Contracts --no-restore
dotnet add Contracts/Contracts.csproj package Microsoft.Orleans.Core.Abstractions --no-restore
dotnet add Contracts/Contracts.csproj package Microsoft.Orleans.OrleansCodeGenerator.Build --no-restore

Set up the Grains project:

dotnet new classlib --name Grains --no-restore
dotnet add Grains/Grains.csproj package Microsoft.Orleans.Core.Abstractions --no-restore
dotnet add Grains/Grains.csproj package Microsoft.Orleans.OrleansCodeGenerator.Build --no-restore
dotnet add Grains/Grains.csproj reference Contracts/Contracts.csproj

Set up the Silo project:

dotnet new console --name Silo --no-restore
dotnet add Silo/Silo.csproj package Microsoft.Orleans.Server --no-restore
dotnet add Silo/Silo.csproj package Microsoft.Extensions.Logging.Console --no-restore
dotnet add Silo/Silo.csproj package OrleansDashboard --no-restore
dotnet add Silo/Silo.csproj reference Contracts/Contracts.csproj
dotnet add Silo/Silo.csproj reference Grains/Grains.csproj

Set up the Client project:

dotnet new webapi --name Client --no-restore
dotnet add Client/Client.csproj package Microsoft.Orleans.Client --no-restore
dotnet add Client/Client.csproj package Microsoft.Extensions.Logging.Console --no-restore
dotnet add Client/Client.csproj reference Contracts/Contracts.csproj

Finally, create a solution that includes all the above projects:

dotnet new sln --name Orleans2
dotnet sln Orleans2.sln add Contracts/Contracts.csproj
dotnet sln Orleans2.sln add Grains/Grains.csproj
dotnet sln Orleans2.sln add Silo/Silo.csproj
dotnet sln Orleans2.sln add Client/Client.csproj

Before we proceed, let’s build this solution to make sure it actually works. dotnet build restores packages as part of the build so there’s no need to do a dotnet restore separately.

dotnet build

It will take a little while to go through the restore, build and codegen steps, but it should work:

And there’s no reason why it shouldn’t work on Linux as well:

Setting Up an Example

Before proceeding with other things we need in a proper Orleans 2.0 solution, let’s set up a little example we can work with. This time, we’ll have a GameGrain that keeps track of players in a game. It will support three operations: Join, Leave, and List Players. To keep things simple, the grain will maintain the list of players in memory. This means that the player list won’t survive any failures or grain reactivations.

In the Contracts project, add a grain interface:

    public interface IGameGrain : IGrainWithIntegerKey
    {
        Task JoinAsync(string playerName);
        Task LeaveAsync(string playerName);
        Task<List<string>> ListPlayersAsync();
    }

In the Grains project, add the grain itself:

    public class GameGrain : Grain, IGameGrain
    {
        private HashSet<string> players;

        public GameGrain() => this.players = new HashSet<string>();

        public Task JoinAsync(string playerName)
        {
            this.players.Add(playerName);
            return Task.CompletedTask;
        }

        public Task LeaveAsync(string playerName)
        {
            this.players.Remove(playerName);
            return Task.CompletedTask;
        }

        public Task<List<string>> ListPlayersAsync()
            => Task.FromResult(this.players.ToList());
    }

In the Silo project, our silo startup code will be pretty much the same as in the previous article:

        public static async Task Main(string[] args)
        {
            var siloBuilder = new SiloHostBuilder()
                .UseLocalhostClustering()
                .UseDashboard(options => { })
                .Configure<ClusterOptions>(options =>
                {
                    options.ClusterId = "dev";
                    options.ServiceId = "Orleans2GettingOrganised";
                })
                .Configure<EndpointOptions>(options =>
                    options.AdvertisedIPAddress = IPAddress.Loopback)
                .ConfigureLogging(logging => logging.AddConsole());

            using (var host = siloBuilder.Build())
            {
                await host.StartAsync();

                Console.ReadLine();
            }
        }

Remember that we need at least C# 7.1 to allow async/await in Main().

If you’re targeting Windows, you may want to add Topshelf to make a Windows service out of your silo. However, since this is application-specific, we won’t be covering it here.

The way we set up our Orleans client in the Client project is going to be a bit different from what we did in our previous article, because now we’re dealing with an ASP .NET Core Web API.

We can put the basic client connection code in a new helper method within the Startup class:

        private IClusterClient CreateOrleansClient()
        {
            var clientBuilder = new ClientBuilder()
                .UseLocalhostClustering()
                .Configure<ClusterOptions>(options =>
                {
                    options.ClusterId = "dev";
                    options.ServiceId = "Orleans2GettingOrganised";
                })
                .ConfigureLogging(logging => logging.AddConsole());

            var client = clientBuilder.Build();
            client.Connect().Wait();

            return client;
        }

Note how we’re calling the blocking Wait() instead of doing the usually recommended await when connecting. This is because we’re going to be calling this from the methods in the Startup class, which are synchronous. Not only is there no way to do async in there, but it actually makes sense not to. You want to wait until your services are fully configured before beginning to accept requests.

We can then register the client in the ASP .NET Core IoC container, by adding the following code to the ConfigureServices() method:

        // This method gets called by the runtime. Use this method to add services to the container.
        public void ConfigureServices(IServiceCollection services)
        {
            var orleansClient = CreateOrleansClient();
            services.AddSingleton<IClusterClient>(orleansClient);

            services.AddMvc();
        }

We now need to add a controller that can accept requests and use the Orleans client to interact with the cluster:

    [Produces("application/json")]
    [Route("api/Games")]
    public class GamesController : Controller
    {
        private IClusterClient orleansClient;

        public GamesController(IClusterClient orleansClient)
        {
            this.orleansClient = orleansClient;
        }

        [HttpGet]
        public Task<List<string>> Get(int gameId)
        {
            var grain = this.orleansClient.GetGrain<IGameGrain>(gameId);
            return grain.ListPlayersAsync();
        }

        [HttpPut]
        public async Task Put(int gameId, string playerName)
        {
            var grain = this.orleansClient.GetGrain<IGameGrain>(gameId);
            await grain.JoinAsync(playerName);
        }

        [HttpDelete]
        public async Task Delete(int gameId, string playerName)
        {
            var grain = this.orleansClient.GetGrain<IGameGrain>(gameId);
            await grain.LeaveAsync(playerName);
        }
    }

In order to test this, we need to make sure that the silo has fully started before we start the client. We also need a way to interact with the API. We can add Swagger to the Web API, or use some other tool such as Postman, Fiddler or curl.

It should work nicely:

Client Retries

This is all well and good, but having to wait for the silo to be up before starting the client is silly. This can be tedious and brittle when debugging locally or during deployments. Ideally the client should keep trying to connect to the silo until it is available.

We can do that by putting the client creation and connection code within a loop:

        private IClusterClient CreateOrleansClient()
        {
            while (true) // keep trying to connect until silo is available
            {
                try
                {
                    var clientBuilder = new ClientBuilder()
                        .UseLocalhostClustering()
                        .Configure<ClusterOptions>(options =>
                        {
                            options.ClusterId = "dev";
                            options.ServiceId = "Orleans2GettingOrganised";
                        })
                        .ConfigureLogging(logging => logging.AddConsole());

                    var client = clientBuilder.Build();
                    client.Connect().Wait();

                    return client;
                }
                catch (Exception)
                {
                    Thread.Sleep(3000);
                    // log a warning or something
                }
            }
        }

Now it might seem super weird that we’re going through the hassle of recreating the ClientBuilder, building that into a client, and doing the reconnect, every time. And it is. By some strange design decision, these APIs don’t let you call ClientBuilder.Build() more than once, nor do they let you call Connect() on a client that has already failed. This means that you have to recreate everything with each connection attempt, which is tedious and inefficient.

Also, connection failures result in an OrleansException, which doesn’t really distinguish between different kinds of failures. If you want to distinguish between an intermittent connection failure and some catastrophic event… good luck with that.

Update 23rd April 2018: As a couple of people pointed out on the Orleans gitter chat, an easier way to achieve client retries is to pass a retry delegate to the Connect() method. The following is a simple example of how a fixed-interval retry could be implemented, but such a delegate makes it easy to implement more advanced mechanisms such as exponential backoff.

                    client.Connect(async ex =>
                    {  // replace Console with actual logging
                        Console.WriteLine(ex);
                        Console.WriteLine("Retrying...");
                        await Task.Delay(3000);
                        return true;
                    }).Wait();

Server Garbage Collection

The Orleans documentation recommends configuring .NET garbage collection as an optimisation to get better performance from your silos. In a .NET Core project, this means adding the following two settings to the .csproj file (in the full .NET Framework it’s different):

  <ServerGarbageCollection>true</ServerGarbageCollection>
  <ConcurrentGarbageCollection>true</ConcurrentGarbageCollection>

This should in theory fix the following warnings:

Unfortunately, this doesn’t work at the time of writing this article. Hopefully they’ll fix it sometime soon.

Application Parts

In Orleans 1.x, complaints about silo start times were common. Orleans would scan all the assemblies in the executable’s folder looking for grains, leading to long start times for larger projects. It still does this in Orleans 2.0 by default, but now you can be more explicit and tell it where to look if you want.

Orleans 2.0 introduces something called application parts (based on ASP .NET Core naming, apparently), which is just a really bad way of saying “places from where to load grains”. I’ve already expressed concerns over how unintuitive this part of the API is to work with.

Thankfully, it’s not something you’ll need all the time. You can usually ignore the existence of this feature, and use it only when you notice slow startup times and want to optimise them.

Summary

In this article, we’ve seen a number of things that take us closer towards having a production-ready Orleans setup. These include:

1. A better project structure.
2. A Web API serving as a client to the Orleans cluster.
3. Client retries.
4. Server garbage collection.
5. Application parts (grain sources).

As part of all this, we’ve also seen how to automate creation of our Orleans 2.0 solution and projects, and how to interact with an Orleans cluster via a REST API.

We haven’t, however, covered everything you’d typically have in a full solution. Some enhancements you may also need (which are beyond the scope of this article) include:

• Using Topshelf to install the Client/API as a Windows service (if deploying on Windows). This can also be done for the Silo, if it’s not going to be run under IIS.
• Configuring actual endpoints rather than using localhost.
• Adding Swagger to the Client/API (the source code for this article does include it, but we haven’t covered it since I have a separate article on that).
• Setting up dependency injection.
• Setting up Orleans clustering (and running multiple silos).