Developer's guide to durable entities in .NET

In this article, we describe the available interfaces for developing durable entities with .NET in detail, including examples and general advice.

Entity functions provide serverless application developers with a convenient way to organize application state as a collection of fine-grained entities. For more detail about the underlying concepts, see the Durable Entities: Concepts article.

We currently offer two APIs for defining entities:

  • The class-based syntax represents entities and operations as classes and methods. This syntax produces easily readable code and allows operations to be invoked in a type-checked manner through interfaces.

  • The function-based syntax is a lower-level interface that represents entities as functions. It provides precise control over how the entity operations are dispatched, and how the entity state is managed.

This article focuses primarily on the class-based syntax, as we expect it to be better suited for most applications. However, the function-based syntax may be appropriate for applications that wish to define or manage their own abstractions for entity state and operations. Also, it may be appropriate for implementing libraries that require genericity not currently supported by the class-based syntax.

Note

The class-based syntax is just a layer on top of the function-based syntax, so both variants can be used interchangeably in the same application.

Defining entity classes

The following example is an implementation of a Counter entity that stores a single value of type integer, and offers four operations Add, Reset, Get, and Delete.

[JsonObject(MemberSerialization.OptIn)]
public class Counter
{
    [JsonProperty("value")]
    public int Value { get; set; }

    public void Add(int amount) 
    {
        this.Value += amount;
    }

    public Task Reset() 
    {
        this.Value = 0;
        return Task.CompletedTask;
    }

    public Task<int> Get() 
    {
        return Task.FromResult(this.Value);
    }

    public void Delete() 
    {
        Entity.Current.DeleteState();
    }

    [FunctionName(nameof(Counter))]
    public static Task Run([EntityTrigger] IDurableEntityContext ctx)
        => ctx.DispatchAsync<Counter>();
}

The Run function contains the boilerplate required for using the class-based syntax. It must be a static Azure Function. It executes once for each operation message that is processed by the entity. When DispatchAsync<T> is called and the entity isn't already in memory, it constructs an object of type T and populates its fields from the last persisted JSON found in storage (if any). Then it invokes the method with the matching name.

Note

The state of a class-based entity is created implicitly before the entity processes an operation, and can be deleted explicitly in an operation by calling Entity.Current.DeleteState().

Class Requirements

Entity classes are POCOs (plain old CLR objects) that require no special superclasses, interfaces, or attributes. However:

Also, any method that is intended to be invoked as an operation must satisfy additional requirements:

  • An operation must have at most one argument, and must not have any overloads or generic type arguments.
  • An operation meant to be called from an orchestration using an interface must return Task or Task<T>.
  • Arguments and return values must be serializable values or objects.

What can operations do?

All entity operations can read and update the entity state, and changes to the state are automatically persisted to storage. Moreover, operations can perform external I/O or other computations, within the general limits common to all Azure Functions.

Operations also have access to functionality provided by the Entity.Current context:

  • EntityName: the name of the currently executing entity.
  • EntityKey: the key of the currently executing entity.
  • EntityId: the ID of the currently executing entity (includes name and key).
  • SignalEntity: sends a one-way message to an entity.
  • CreateNewOrchestration: starts a new orchestration.
  • DeleteState: deletes the state of this entity.

For example, we can modify the counter entity so it starts an orchestration when the counter reaches 100 and passes the entity ID as an input argument:

    public void Add(int amount) 
    {
        if (this.Value < 100 && this.Value + amount > 100)
        {
            Entity.Current.StartNewOrchestration("MilestoneReached", Entity.Current.EntityId)
        }
        this.Value += amount;      
    }

Accessing entities directly

Class-based entities can be accessed directly, using explicit string names for the entity and its operations. We provide some examples below; for a deeper explanation of the underlying concepts (such as signals vs. calls) see the discussion in Access entities.

Note

Where possible, we recommend Accessing entities through interfaces, because it provides more type checking.

Example: client signals entity

The following Azure Http Function implements a DELETE operation using REST conventions. It sends a delete signal to the counter entity whose key is passed in the URL path.

[FunctionName("DeleteCounter")]
public static async Task<HttpResponseMessage> DeleteCounter(
    [HttpTrigger(AuthorizationLevel.Function, "delete", Route = "Counter/{entityKey}")] HttpRequestMessage req,
    [DurableClient] IDurableEntityClient client,
    string entityKey)
{
    var entityId = new EntityId("Counter", entityKey);
    await client.SignalEntityAsync(entityId, "Delete");    
    return req.CreateResponse(HttpStatusCode.Accepted);
}

Example: client reads entity state

The following Azure Http Function implements a GET operation using REST conventions. It reads the current state of the counter entity whose key is passed in the URL path.

[FunctionName("GetCounter")]
public static async Task<HttpResponseMessage> GetCounter(
    [HttpTrigger(AuthorizationLevel.Function, "get", Route = "Counter/{entityKey}")] HttpRequestMessage req,
    [DurableClient] IDurableEntityClient client,
    string entityKey)
{
    var entityId = new EntityId("Counter", entityKey);
    var state = await client.ReadEntityStateAsync<Counter>(entityId); 
    return req.CreateResponse(state);
}

Note

The object returned by ReadEntityStateAsync is just a local copy, that is, a snapshot of the entity state from some earlier point in time. In particular, it may be stale, and modifying this object has no effect on the actual entity.

Example: orchestration first signals, then calls entity

The following orchestration signals a counter entity to increment it, and then calls the same entity to read its latest value.

[FunctionName("IncrementThenGet")]
public static async Task<int> Run(
    [OrchestrationTrigger] IDurableOrchestrationContext context)
{
    var entityId = new EntityId("Counter", "myCounter");

    // One-way signal to the entity - does not await a response
    context.SignalEntity(entityId, "Add", 1);

    // Two-way call to the entity which returns a value - awaits the response
    int currentValue = await context.CallEntityAsync<int>(entityId, "Get");

    return currentValue;
}

Accessing entities through interfaces

Interfaces can be used for accessing entities via generated proxy objects. This approach ensures that the name and argument type of an operation matches what is implemented. We recommend using interfaces for accessing entities whenever possible.

For example, we can modify the counter example as follows:

public interface ICounter
{
    void Add(int amount);
    Task Reset();
    Task<int> Get();
    void Delete();
}

public class Counter : ICounter
{
    ...
}

Entity classes and entity interfaces are similar to the grains and grain interfaces popularized by Orleans. For a more information about similarities and differences between Durable Entities and Orleans, see Comparison with virtual actors.

Besides providing type checking, interfaces are useful for a better separation of concerns within the application. For example, since an entity may implement multiple interfaces, a single entity can serve multiple roles. Also, since an interface may be implemented by multiple entities, general communication patterns can be implemented as reusable libraries.

Example: client signals entity through interface

Client code can use SignalEntityAsync<TEntityInterface> to send signals to entities that implement TEntityInterface. For example:

[FunctionName("DeleteCounter")]
public static async Task<HttpResponseMessage> DeleteCounter(
    [HttpTrigger(AuthorizationLevel.Function, "delete", Route = "Counter/{entityKey}")] HttpRequestMessage req,
    [DurableClient] IDurableEntityClient client,
    string entityKey)
{
    var entityId = new EntityId("Counter", entityKey);
    await client.SignalEntityAsync<ICounter>(entityId, proxy => proxy.Delete());    
    return req.CreateResponse(HttpStatusCode.Accepted);
}

In this example, the proxy parameter is a dynamically generated instance of ICounter, which internally translates the call to Delete into a signal.

Note

The SignalEntityAsync APIs can be used only for one-way operations. Even if an operation returns Task<T>, the value of the T parameter will always be null or default, not the actual result. For example, it doesn't make sense to signal the Get operation, as no value is returned. Instead, clients can use either ReadStateAsync to access the counter state directly, or can start an orchestrator function that calls the Get operation.

Example: orchestration first signals, then calls entity through proxy

To call or signal an entity from within an orchestration, CreateEntityProxy can be used, along with the interface type, to generate a proxy for the entity. This proxy can then be used to call or signal operations:

[FunctionName("IncrementThenGet")]
public static async Task<int> Run(
    [OrchestrationTrigger] IDurableOrchestrationContext context)
{
    var entityId = new EntityId("Counter", "myCounter");
    var proxy = context.CreateEntityProxy<ICounter>(entityId);

    // One-way signal to the entity - does not await a response
    proxy.Add(1);

    // Two-way call to the entity which returns a value - awaits the response
    int currentValue = await proxy.Get();

    return currentValue;
}

Implicitly, any operations that return void are signaled, and any operations that return Task or Task<T> are called. One can change this default behavior, and signal operations even if they return Task, by using the SignalEntity<IInterfaceType> method explicitly.

Shorter option for specifying the target

When calling or signaling an entity using an interface, the first argument must specify the target entity. The target can be specified either by specifying the entity ID, or, in cases where there's just one class that implements the entity, just the entity key:

context.SignalEntity<ICounter>(new EntityId(nameof(Counter), "myCounter"), ...);
context.SignalEntity<ICounter>("myCounter", ...);

If only the entity key is specified and a unique implementation can't be found at runtime, InvalidOperationException is thrown.

Restrictions on entity interfaces

As usual, all parameter and return types must be JSON-serializable. Otherwise, serialization exceptions are thrown at runtime.

We also enforce some additional rules:

  • Entity interfaces must only define methods.
  • Entity interfaces must not contain generic parameters.
  • Entity interface methods must not have more than one parameter.
  • Entity interface methods must return void, Task, or Task<T>

If any of these rules are violated, an InvalidOperationException is thrown at runtime when the interface is used as a type argument to SignalEntity or CreateProxy. The exception message explains which rule was broken.

Note

Interface methods returning void can only be signaled (one-way), not called (two-way). Interface methods returning Task or Task<T> can be either called or signalled. If called, they return the result of the operation, or re-throw exceptions thrown by the operation. However, when signalled, they do not return the actual result or exception from the operation, but just the default value.

Entity serialization

Since the state of an entity is durably persisted, the entity class must be serializable. The Durable Functions runtime uses the Json.NET library for this purpose, which supports a number of policies and attributes to control the serialization and deserialization process. Most commonly used C# data types (including arrays and collection types) are already serializable, and can easily be used for defining the state of durable entities.

For example, Json.NET can easily serialize and deserialize the following class:

[JsonObject(MemberSerialization = MemberSerialization.OptIn)]
public class User
{
    [JsonProperty("name")]
    public string Name { get; set; }

    [JsonProperty("yearOfBirth")]
    public int YearOfBirth { get; set; }

    [JsonProperty("timestamp")]
    public DateTime Timestamp { get; set; }

    [JsonProperty("contacts")]
    public Dictionary<Guid, Contact> Contacts { get; set; } = new Dictionary<Guid, Contact>();

    [JsonObject(MemberSerialization = MemberSerialization.OptOut)]
    public struct Contact
    {
        public string Name;
        public string Number;
    }

    ...
}

Serialization Attributes

In the example above, we chose to include several attributes to make the underlying serialization more visible:

  • We annotate the class with [JsonObject(MemberSerialization.OptIn)] to remind us that the class must be serializable, and to persist only members that are explicitly marked as JSON properties.
  • We annotate the fields to be persisted with [JsonProperty("name")] to remind us that a field is part of the persisted entity state, and to specify the property name to be used in the JSON representation.

However, these attributes aren't required; other conventions or attributes are permitted as long as they work with Json.NET. For example, one may use [DataContract] attributes, or no attributes at all:

[DataContract]
public class Counter
{
    [DataMember]
    public int Value { get; set; }
    ...
}

public class Counter
{
    public int Value;
    ...
}

By default, the name of the class is not stored as part of the JSON representation: that is, we use TypeNameHandling.None as the default setting. This default behavior can be overridden using JsonObject or JsonProperty attributes.

Making changes to class definitions

Some care is required when making changes to a class definition after an application has been run, because the stored JSON object may no longer match the new class definition. Still, it is often possible to deal correctly with changing data formats as long as one understands the deserialization process used by JsonConvert.PopulateObject.

For example, here are some examples of changes and their effect:

  1. If a new property is added, which is not present in the stored JSON, it assumes its default value.
  2. If a property is removed, which is present in the stored JSON, the previous content is lost.
  3. If a property is renamed, the effect is as if removing the old one and adding a new one.
  4. If the type of a property is changed so it can no longer be deserialized from the stored JSON, an exception is thrown.
  5. If the type of a property is changed, but it can still be deserialized from the stored JSON, it will do so.

There are many options available for customizing the behavior of Json.NET. For example, to force an exception if the stored JSON contains a field that is not present in the class, specify the attribute JsonObject(MissingMemberHandling = MissingMemberHandling.Error). It is also possible to write custom code for deserialization that can read JSON stored in arbitrary formats.

Entity construction

Sometimes we want to exert more control over how entity objects are constructed. We now describe several options for changing the default behavior when constructing entity objects.

Custom initialization on first access

Occasionally we need to perform some special initialization before dispatching an operation to an entity that has never been accessed, or that has been deleted. To specify this behavior, one can add a conditional before the DispatchAsync:

[FunctionName(nameof(Counter))]
public static Task Run([EntityTrigger] IDurableEntityContext ctx)
{
    if (!ctx.HasState)
    {
        ctx.SetState(...);
    }
    ctx.DispatchAsync<Counter>();
}

Bindings in entity classes

Unlike regular functions, entity class methods don't have direct access to input and output bindings. Instead, binding data must be captured in the entry-point function declaration and then passed to the DispatchAsync<T> method. Any objects passed to DispatchAsync<T> will be automatically passed into the entity class constructor as an argument.

The following example shows how a CloudBlobContainer reference from the blob input binding can be made available to a class-based entity.

public class BlobBackedEntity
{
    [JsonIgnore]
    private readonly CloudBlobContainer container;

    public BlobBackedEntity(CloudBlobContainer container)
    {
        this.container = container;
    }

    // ... entity methods can use this.container in their implementations ...
    
    [FunctionName(nameof(BlobBackedEntity))]
    public static Task Run(
        [EntityTrigger] IDurableEntityContext context,
        [Blob("my-container", FileAccess.Read)] CloudBlobContainer container)
    {
        // passing the binding object as a parameter makes it available to the
        // entity class constructor
        return context.DispatchAsync<BlobBackedEntity>(container);
    }
}

For more information on bindings in Azure Functions, see the Azure Functions Triggers and Bindings documentation.

Dependency injection in entity classes

Entity classes support Azure Functions Dependency Injection. The following example demonstrates how to register an IHttpClientFactory service into a class-based entity.

[assembly: FunctionsStartup(typeof(MyNamespace.Startup))]

namespace MyNamespace
{
    public class Startup : FunctionsStartup
    {
        public override void Configure(IFunctionsHostBuilder builder)
        {
            builder.Services.AddHttpClient();
        }
    }
}

The following snippet demonstrates how to incorporate the injected service into your entity class.

public class HttpEntity
{
    [JsonIgnore]
    private readonly HttpClient client;

    public class HttpEntity(IHttpClientFactory factory)
    {
        this.client = factory.CreateClient();
    }

    public Task<int> GetAsync(string url)
    {
        using (var response = await this.client.GetAsync(url))
        {
            return (int)response.StatusCode;
        }
    }

    [FunctionName(nameof(HttpEntity))]
    public static Task Run([EntityTrigger] IDurableEntityContext ctx)
        => ctx.DispatchAsync<HttpEntity>();
}

Note

To avoid issues with serialization, make sure to exclude fields meant to store injected values from the serialization.

Note

Unlike when using constructor injection in regular .NET Azure Functions, the functions entry point method for class-based entities must be declared static. Declaring a non-static function entry point may cause conflicts between the normal Azure Functions object initializer and the Durable Entities object initializer.

Function-based syntax

So far we have focused on the class-based syntax, as we expect it to be better suited for most applications. However, the function-based syntax can be appropriate for applications that wish to define or manage their own abstractions for entity state and operations. Also, it may be appropriate when implementing libraries that require genericity not currently supported by the class-based syntax.

With the function-based syntax, the Entity Function explicitly handles the operation dispatch, and explicitly manages the state of the entity. For example, the following code shows the Counter entity implemented using the function-based syntax.

[FunctionName("Counter")]
public static void Counter([EntityTrigger] IDurableEntityContext ctx)
{
    switch (ctx.OperationName.ToLowerInvariant())
    {
        case "add":
            ctx.SetState(ctx.GetState<int>() + ctx.GetInput<int>());
            break;
        case "reset":
            ctx.SetState(0);
            break;
        case "get":
            ctx.Return(ctx.GetState<int>()));
            break;
        case "delete":
            ctx.DeleteState();
            break;
    }
}

The entity context object

Entity-specific functionality can be accessed via a context object of type IDurableEntityContext. This context object is available as a parameter to the entity function, and via the async-local property Entity.Current.

The following members provide information about the current operation, and allow us to specify a return value.

  • EntityName: the name of the currently executing entity.
  • EntityKey: the key of the currently executing entity.
  • EntityId: the ID of the currently executing entity (includes name and key).
  • OperationName: the name of the current operation.
  • GetInput<TInput>(): gets the input for the current operation.
  • Return(arg): returns a value to the orchestration that called the operation.

The following members manage the state of the entity (create, read, update, delete).

  • HasState: whether the entity exists, that is, has some state.
  • GetState<TState>(): gets the current state of the entity. If it does not already exist, it is created.
  • SetState(arg): creates or updates the state of the entity.
  • DeleteState(): deletes the state of the entity, if it exists.

If the state returned by GetState is an object, it can be directly modified by the application code. There is no need to call SetState again at the end (but also no harm). If GetState<TState> is called multiple times, the same type must be used.

Finally, the following members are used to signal other entities, or start new orchestrations:

  • SignalEntity(EntityId, operation, input): sends a one-way message to an entity.
  • CreateNewOrchestration(orchestratorFunctionName, input): starts a new orchestration.

Next steps