我有一些问题,试图包装我的代码在单元测试中使用。问题在于。我有接口IHttpHandler:

public interface IHttpHandler
{
    HttpClient client { get; }
}

使用它的类HttpHandler:

public class HttpHandler : IHttpHandler
{
    public HttpClient client
    {
        get
        {
            return new HttpClient();
        }
    }
}

然后是Connection类,它使用simpleIOC注入客户端实现:

public class Connection
{
    private IHttpHandler _httpClient;

    public Connection(IHttpHandler httpClient)
    {
        _httpClient = httpClient;
    }
}

然后我有一个单元测试项目,它有这个类:

private IHttpHandler _httpClient;

[TestMethod]
public void TestMockConnection()
{
    var client = new Connection(_httpClient);
     
    client.doSomething();  

    // Here I want to somehow create a mock instance of the http client
    // Instead of the real one. How Should I approach this?     

}

现在很明显,我将在Connection类中拥有从后端检索数据(JSON)的方法。但是,我想为这个类编写单元测试,显然我不想针对真正的后端编写测试,而是一个模拟的后端。我试着给这个问题一个好的答案,但没有成功。我以前可以用Moq来模拟,但从来没有在HttpClient这样的东西上使用过。我应该如何处理这个问题?


当前回答

加入这个派对有点晚了,但我喜欢在带有下游REST依赖的dotnet核心微服务的集成测试中尽可能使用wiremocking (https://github.com/WireMock-Net/WireMock.Net)。

通过实现一个扩展IHttpClientFactory的TestHttpClientFactory,我们可以重写这个方法

HttpClient CreateClient(字符串名称)

所以当你在应用中使用命名客户端时,你可以控制返回一个连接到你的wiremock的HttpClient。

这种方法的好处是,您不会更改正在测试的应用程序中的任何内容,并且允许课程集成测试对您的服务执行实际的REST请求,并模拟实际下游请求应该返回的json(或任何东西)。这将导致在应用程序中进行简洁的测试和尽可能少的模拟。

    public class TestHttpClientFactory : IHttpClientFactory 
{
    public HttpClient CreateClient(string name)
    {
        var httpClient = new HttpClient
        {
            BaseAddress = new Uri(G.Config.Get<string>($"App:Endpoints:{name}"))
            // G.Config is our singleton config access, so the endpoint 
            // to the running wiremock is used in the test
        };
        return httpClient;
    }
}

and

// in bootstrap of your Microservice
IHttpClientFactory factory = new TestHttpClientFactory();
container.Register<IHttpClientFactory>(factory);

其他回答

您的接口公开了具体的HttpClient类,因此任何使用该接口的类都被绑定到它,这意味着它不能被模拟。

HttpClient不继承任何接口,所以你必须自己编写。我建议使用类似装饰器的图案:

public interface IHttpHandler
{
    HttpResponseMessage Get(string url);
    HttpResponseMessage Post(string url, HttpContent content);
    Task<HttpResponseMessage> GetAsync(string url);
    Task<HttpResponseMessage> PostAsync(string url, HttpContent content);
}

你的类看起来是这样的:

public class HttpClientHandler : IHttpHandler
{
    private HttpClient _client = new HttpClient();

    public HttpResponseMessage Get(string url)
    {
        return GetAsync(url).Result;
    }

    public HttpResponseMessage Post(string url, HttpContent content)
    {
        return PostAsync(url, content).Result;
    }

    public async Task<HttpResponseMessage> GetAsync(string url)
    {
        return await _client.GetAsync(url);
    }

    public async Task<HttpResponseMessage> PostAsync(string url, HttpContent content)
    {
        return await _client.PostAsync(url, content);
    }
}

所有这些的重点是HttpClientHandler创建了它自己的HttpClient,然后你当然可以创建多个以不同方式实现IHttpHandler的类。

这种方法的主要问题是,你实际上是在编写一个只调用另一个类中的方法的类,然而你可以创建一个从HttpClient继承的类(参见Nkosi的例子,这是一个比我的方法更好的方法)。如果HttpClient有一个可以模拟的接口,事情会简单得多,不幸的是它没有。

This example is not the golden ticket however. IHttpHandler still relies on HttpResponseMessage, which belongs to System.Net.Http namespace, therefore if you do need other implementations other than HttpClient, you will have to perform some kind of mapping to convert their responses into HttpResponseMessage objects. This of course is only a problem if you need to use multiple implementations of IHttpHandler but it doesn't look like you do so it's not the end of the world, but it's something to think about.

无论如何,您可以简单地模拟IHttpHandler,而不必担心具体的HttpClient类,因为它已经被抽象掉了。

我建议测试非异步方法,因为这些方法仍然调用异步方法,但不必担心单元测试异步方法的麻烦,参见这里

正如注释中提到的,您需要抽象HttpClient,这样就不会与它耦合。我以前也做过类似的事情。我会试着把我的方法和你们要做的方法相适应。

首先看一下HttpClient类,并决定需要它提供哪些功能。

以下是一种可能性:

public interface IHttpClient {
    System.Threading.Tasks.Task<T> DeleteAsync<T>(string uri) where T : class;
    System.Threading.Tasks.Task<T> DeleteAsync<T>(Uri uri) where T : class;
    System.Threading.Tasks.Task<T> GetAsync<T>(string uri) where T : class;
    System.Threading.Tasks.Task<T> GetAsync<T>(Uri uri) where T : class;
    System.Threading.Tasks.Task<T> PostAsync<T>(string uri, object package);
    System.Threading.Tasks.Task<T> PostAsync<T>(Uri uri, object package);
    System.Threading.Tasks.Task<T> PutAsync<T>(string uri, object package);
    System.Threading.Tasks.Task<T> PutAsync<T>(Uri uri, object package);
}

如前所述,这是为了特定目的。我完全抽象出了对HttpClient的依赖,并专注于我想要返回的东西。您应该评估如何抽象HttpClient以只提供您想要的必要功能。

这将允许您只模拟需要测试的内容。

我甚至建议完全放弃IHttpHandler,而使用HttpClient抽象IHttpClient。但我只是没有选择,因为你可以用抽象客户端的成员替换你的处理程序接口的主体。

IHttpClient的实现可以用来包装/改编一个真正的/具体的HttpClient或任何其他对象,它可以用来发出HTTP请求,因为你真正想要的是一个服务,它提供了与HttpClient相反的功能。使用抽象是一种干净(我的观点)和可靠的方法,并且可以使您的代码更易于维护,如果您需要在框架更改时将底层客户端切换为其他东西。

下面是如何实现的代码片段。

/// <summary>
/// HTTP Client adaptor wraps a <see cref="System.Net.Http.HttpClient"/> 
/// that contains a reference to <see cref="ConfigurableMessageHandler"/>
/// </summary>
public sealed class HttpClientAdaptor : IHttpClient {
    HttpClient httpClient;

    public HttpClientAdaptor(IHttpClientFactory httpClientFactory) {
        httpClient = httpClientFactory.CreateHttpClient(**Custom configurations**);
    }

    //...other code

     /// <summary>
    ///  Send a GET request to the specified Uri as an asynchronous operation.
    /// </summary>
    /// <typeparam name="T">Response type</typeparam>
    /// <param name="uri">The Uri the request is sent to</param>
    /// <returns></returns>
    public async System.Threading.Tasks.Task<T> GetAsync<T>(Uri uri) where T : class {
        var result = default(T);
        //Try to get content as T
        try {
            //send request and get the response
            var response = await httpClient.GetAsync(uri).ConfigureAwait(false);
            //if there is content in response to deserialize
            if (response.Content.Headers.ContentLength.GetValueOrDefault() > 0) {
                //get the content
                string responseBodyAsText = await response.Content.ReadAsStringAsync().ConfigureAwait(false);
                //desrialize it
                result = deserializeJsonToObject<T>(responseBodyAsText);
            }
        } catch (Exception ex) {
            Log.Error(ex);
        }
        return result;
    }

    //...other code
}

正如你在上面的例子中所看到的,很多通常与使用HttpClient相关的繁重工作都隐藏在抽象后面。

然后可以用抽象客户端注入连接类

public class Connection
{
    private IHttpClient _httpClient;

    public Connection(IHttpClient httpClient)
    {
        _httpClient = httpClient;
    }
}

然后,您的测试可以模拟SUT所需的内容

private IHttpClient _httpClient;

[TestMethod]
public void TestMockConnection()
{
    SomeModelObject model = new SomeModelObject();
    var httpClientMock = new Mock<IHttpClient>();
    httpClientMock.Setup(c => c.GetAsync<SomeModelObject>(It.IsAny<string>()))
        .Returns(() => Task.FromResult(model));

    _httpClient = httpClientMock.Object;

    var client = new Connection(_httpClient);

    // Assuming doSomething uses the client to make
    // a request for a model of type SomeModelObject
    client.doSomething();  
}

因为HttpClient使用SendAsync方法来执行所有的HTTP请求,你可以重写SendAsync方法并模拟HttpClient。

对于将HttpClient创建到接口的封装,如下所示

public interface IServiceHelper
{
    HttpClient GetClient();
}

然后在您的服务中使用上述接口进行依赖注入,示例如下

public class SampleService
{
    private readonly IServiceHelper serviceHelper;

    public SampleService(IServiceHelper serviceHelper)
    {
        this.serviceHelper = serviceHelper;
    }

    public async Task<HttpResponseMessage> Get(int dummyParam)
    {
        try
        {
            var dummyUrl = "http://www.dummyurl.com/api/controller/" + dummyParam;
            var client = serviceHelper.GetClient();
            HttpResponseMessage response = await client.GetAsync(dummyUrl);               

            return response;
        }
        catch (Exception)
        {
            // log.
            throw;
        }
    }
}

Now in unit test project create a helper class for mocking SendAsync. Here it is a FakeHttpResponseHandler class which is inheriting DelegatingHandler which will provide an option to override the SendAsync method. After overriding the SendAsync method need to setup a response for each HTTP Request which is calling SendAsync method, for that create a Dictionary with key as Uri and value as HttpResponseMessage so that whenever there is a HTTP Request and if the Uri matches SendAsync will return the configured HttpResponseMessage.

public class FakeHttpResponseHandler : DelegatingHandler
{
    private readonly IDictionary<Uri, HttpResponseMessage> fakeServiceResponse;
    private readonly JavaScriptSerializer javaScriptSerializer;
    public FakeHttpResponseHandler()
    {
        fakeServiceResponse =  new Dictionary<Uri, HttpResponseMessage>();
        javaScriptSerializer =  new JavaScriptSerializer();
    }

    /// <summary>
    /// Used for adding fake httpResponseMessage for the httpClient operation.
    /// </summary>
    /// <typeparam name="TQueryStringParameter"> query string parameter </typeparam>
    /// <param name="uri">Service end point URL.</param>
    /// <param name="httpResponseMessage"> Response expected when the service called.</param>
    public void AddFakeServiceResponse(Uri uri, HttpResponseMessage httpResponseMessage)
    {
        fakeServiceResponse.Remove(uri);
        fakeServiceResponse.Add(uri, httpResponseMessage);
    }

    /// <summary>
    /// Used for adding fake httpResponseMessage for the httpClient operation having query string parameter.
    /// </summary>
    /// <typeparam name="TQueryStringParameter"> query string parameter </typeparam>
    /// <param name="uri">Service end point URL.</param>
    /// <param name="httpResponseMessage"> Response expected when the service called.</param>
    /// <param name="requestParameter">Query string parameter.</param>
    public void AddFakeServiceResponse<TQueryStringParameter>(Uri uri, HttpResponseMessage httpResponseMessage, TQueryStringParameter requestParameter)
    {
        var serilizedQueryStringParameter = javaScriptSerializer.Serialize(requestParameter);
        var actualUri = new Uri(string.Concat(uri, serilizedQueryStringParameter));
        fakeServiceResponse.Remove(actualUri);
        fakeServiceResponse.Add(actualUri, httpResponseMessage);
    }

    // all method in HttpClient call use SendAsync method internally so we are overriding that method here.
    protected override Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        if(fakeServiceResponse.ContainsKey(request.RequestUri))
        {
            return Task.FromResult(fakeServiceResponse[request.RequestUri]);
        }

        return Task.FromResult(new HttpResponseMessage(HttpStatusCode.NotFound)
        {
            RequestMessage = request,
            Content = new StringContent("Not matching fake found")
        });
    }
}

通过模仿框架或如下所示创建IServiceHelper的新实现。 我们可以使用这个FakeServiceHelper类来注入FakeHttpResponseHandler类,这样每当这个类创建HttpClient时,它都会使用FakeHttpResponseHandler类而不是实际的实现。

public class FakeServiceHelper : IServiceHelper
{
    private readonly DelegatingHandler delegatingHandler;

    public FakeServiceHelper(DelegatingHandler delegatingHandler)
    {
        this.delegatingHandler = delegatingHandler;
    }

    public HttpClient GetClient()
    {
        return new HttpClient(delegatingHandler);
    }
}

And in test configure FakeHttpResponseHandler class by adding the Uri and expected HttpResponseMessage. The Uri should be the actual serviceendpoint Uri so that when the overridden SendAsync method is called from actual service implementation it will match the Uri in Dictionary and respond with the configured HttpResponseMessage. After configuring inject the FakeHttpResponseHandler object to the fake IServiceHelper implementation. Then inject the FakeServiceHelper class to the actual service which will make the actual service to use the override SendAsync method.

[TestClass]
public class SampleServiceTest
{
    private FakeHttpResponseHandler fakeHttpResponseHandler;

    [TestInitialize]
    public void Initialize()
    {
        fakeHttpResponseHandler = new FakeHttpResponseHandler();
    }

    [TestMethod]
    public async Task GetMethodShouldReturnFakeResponse()
    {
        Uri uri = new Uri("http://www.dummyurl.com/api/controller/");
        const int dummyParam = 123456;
        const string expectdBody = "Expected Response";

        var expectedHttpResponseMessage = new HttpResponseMessage(HttpStatusCode.OK)
        {
            Content = new StringContent(expectdBody)
        };

        fakeHttpResponseHandler.AddFakeServiceResponse(uri, expectedHttpResponseMessage, dummyParam);

        var fakeServiceHelper = new FakeServiceHelper(fakeHttpResponseHandler);

        var sut = new SampleService(fakeServiceHelper);

        var response = await sut.Get(dummyParam);

        var responseBody = await response.Content.ReadAsStringAsync();

        Assert.AreEqual(HttpStatusCode.OK, response.StatusCode);
        Assert.AreEqual(expectdBody, responseBody);
    }
}

GitHub链接:有示例实现

HttpClient的可扩展性在于传递给构造函数的HttpMessageHandler。它的目的是允许特定于平台的实现,但您也可以模拟它。不需要为HttpClient创建装饰器包装。

如果你更喜欢DSL而不是使用Moq,我在GitHub/Nuget上有一个库,它可以让事情变得更容易一些:https://github.com/richardszalay/mockhttp

Nuget包RichardSzalay。MockHttp可以在这里找到。

var mockHttp = new MockHttpMessageHandler();

// Setup a respond for the user api (including a wildcard in the URL)
mockHttp.When("http://localhost/api/user/*")
        .Respond("application/json", "{'name' : 'Test McGee'}"); // Respond with JSON

// Inject the handler or client into your application code
var client = new HttpClient(mockHttp);

var response = await client.GetAsync("http://localhost/api/user/1234");
// or without async: var response = client.GetAsync("http://localhost/api/user/1234").Result;

var json = await response.Content.ReadAsStringAsync();

// No network connection required
Console.Write(json); // {'name' : 'Test McGee'}

这是一个常见的问题,我非常希望能够模拟HttpClient,但我想我最终意识到不应该模拟HttpClient。这样做似乎是合乎逻辑的,但我认为我们已经被我们在开源库中看到的东西洗脑了。

We often see "Clients" out there that we mock in our code so that we can test in isolation, so we automatically try to apply the same principle to HttpClient. HttpClient actually does a lot; you can think of it as a manager for HttpMessageHandler, so you don't wanna mock that, and that's why it still doesn't have an interface. The part that you're really interested in for unit testing, or designing your services, even, is the HttpMessageHandler since that is what returns the response, and you can mock that.

同样值得指出的是,您可能应该开始把HttpClient当作一个更大的交易来对待。例如:让你的新HttpClients的实例化最小化。重复使用它们,它们被设计成可重复使用的,如果你这样做,会使用更少的资源。如果您开始把它当作一个更大的事情来对待,那么想要模拟它就会感觉更错误,现在消息处理程序将开始成为您正在注入的东西,而不是客户端。

换句话说,围绕处理程序而不是客户端设计依赖项。更好的是,使用HttpClient的抽象“服务”允许你注入一个处理程序,并将其作为你的可注入依赖项。事实上,HttpClientFactor(您应该使用它)被设计为带有注入消息处理程序的扩展。然后在测试中,可以伪造处理程序来控制设置测试的响应。

包装HttpClient是一种疯狂的时间浪费。

更新: 请看约书亚·杜姆斯的例子。这正是我所推荐的。