Build/Test verification for missing implementations of query/commands in MediatR - asp.net-core

We're using MediatR heavily in our LoB application, where we use the command & query pattern.
Often, to continue in development, we make the commands and the queries first, since they are simple POCOs.
This sometimes can lead to forgetting to create an actual command handler/query handler. Since there's no compile-time validation if there is actually an implementation for the query/command, I was wondering what would be the best approach to see if there's an implementation and throw an error if not, before being able to merge into master.
My idea so far:
Create a two tests, one for queries and one for commands, that scan all the assemblies for an implementation of IRequest<TResponse>, and then scan the assemblies for an associated implementation of IRequestHandler<TRequest, TResponse>
But this would make it still required to first execute the tests (which is happening in the build pipeline), which still depends on the developer manually executing the tests (or configuring VS to do so after compile).
I don't know if there's a compile-time solution for this, and even if that would be a good idea?

We've gone with a test (and thus build-time) verification;
Sharing the code here for the actual test, which we have once per domain project.
The mediator modules contain our query/command(handler) registrations, the infrastructure modules contain our handlers of queries;
public class MissingHandlersTests
{
[Fact]
public void Missing_Handlers()
{
List<Assembly> assemblies = new List<Assembly>();
assemblies.Add(typeof(MediatorModules).Assembly);
assemblies.Add(typeof(InfrastructureModule).Assembly);
var missingTypes = MissingHandlersHelpers.FindUnmatchedRequests(assemblies);
Assert.Empty(missingTypes);
}
}
The helper class;
public class MissingHandlersHelpers
{
public static IEnumerable<Type> FindUnmatchedRequests(List<Assembly> assemblies)
{
var requests = assemblies.SelectMany(x => x.GetTypes())
.Where(t => t.IsClass && t.IsClosedTypeOf(typeof(IRequest<>)))
.ToList();
var handlerInterfaces = assemblies.SelectMany(x => x.GetTypes())
.Where(t => t.IsClass && (t.IsClosedTypeOf(typeof(IRequestHandler<>)) || t.IsClosedTypeOf(typeof(IRequestHandler<,>))))
.SelectMany(t => t.GetInterfaces())
.ToList();
List<Type> missingRegistrations = new List<Type>();
foreach(var request in requests)
{
var args = request.GetInterfaces().Single(i => i.IsClosedTypeOf(typeof(IRequest<>)) && i.GetGenericArguments().Any() && !i.IsClosedTypeOf(typeof(ICacheableRequest<>))).GetGenericArguments().First();
var handler = typeof(IRequestHandler<,>).MakeGenericType(request, args);
if (handler == null || !handlerInterfaces.Any(x => x == handler))
missingRegistrations.Add(handler);
}
return missingRegistrations;
}
}

If you are using .Net Core you could the Microsoft.AspNetCore.TestHost to create an endpoint your tests could hit. Sort of works like this:
var builder = WebHost.CreateDefaultBuilder()
.UseStartup<TStartup>()
.UseEnvironment(EnvironmentName.Development)
.ConfigureTestServices(
services =>
{
services.AddTransient((a) => this.SomeMockService.Object);
});
this.Server = new TestServer(builder);
this.Services = this.Server.Host.Services;
this.Client = this.Server.CreateClient();
this.Client.BaseAddress = new Uri("http://localhost");
So we mock any http calls (or any other stuff we want) but the real startup gets called.
And our tests would be like this:
public SomeControllerTests(TestServerFixture<Startup> testServerFixture)
: base(testServerFixture)
{
}
[Fact]
public async Task SomeController_Returns_Titles_OK()
{
var response = await this.GetAsync("/somedata/titles");
response.StatusCode.Should().Be(HttpStatusCode.OK);
var responseAsString = await response.Content.ReadAsStringAsync();
var actualResponse = Newtonsoft.Json.JsonConvert.DeserializeObject<IEnumerable<string>>(responseAsString);
actualResponse.Should().NotBeNullOrEmpty();
actualResponse.Should().HaveCount(20);
}
So when this test runs, if you have not registered your handler(s) it will fail! We use this to assert what we need (db records added, response what we expect etc) but it is a nice side effect that forgetting to register your handler gets caught at the test stage!
https://fullstackmark.com/post/20/painless-integration-testing-with-aspnet-core-web-api

Related

How to unit test azure function using Xunit and Moq

I am very new to unit tests and recently started learning it from various online resources.
But still it confuses me when I need to implement it in my code.
For the given image which I have attached here, could anyone of you suggest me how should I start or where to start?
This is Azure function which I will be creating unit test for, framework/library I would prefer is Xunit and moq.
As mentioned in a comment, a good place to start when unit testing is looking at your code and identifying the different "paths" it can take and what the result of that path will be.
if (inventoryRequest != null)
{
// path 1
await _inventoryService.ProcessRequest(inventoryRequest);
_logger.LogInformation("HBSI Inventory Queue trigger function processed.");
}
else
{
// path 2
_logger.LogInformation("Unable to process HBSI Rate plan Queue.");
}
In your code, because of your if statement, there are 2 possible paths which will end in 2 different results = 2 unit tests.
Now you can start creating your unit tests but first you need to find out what you need to set up to be able to trigger your code.
private readonly ILogger _logger;
private readonly IInventoryService _inventoryService;
public InventoryServiceBusFunction(ILogger logger, IInventoryService inventoryService)
{
_logger = logger;
_inventoryService = inventoryService;
}
You have some dependencies being passed into your constructor with interfaces - great, this means we can mock them. We want to mock dependencies in unit tests because we want to control their behaviour for the tests. Also, mocking the dependencies negates any "real" behaviour the dependency might be performing i.e. database operations, API calls etc.
Using Moq we can mock the objects like so:
public class InventoryServiceBusFunctionTests
{
private readonly Mock<ILogger> _mockLogger = new Mock<ILogger>();
private readonly Mock<IInventoryService> _mockInventoryService = new Mock<IInventoryService>();
...
We will use these mocks later to make verifications on behaviour we expect to happen.
Next, we need to create an instance of the actual class we want to test.
// using a constructor in the test class will run this code before each test
public InventoryServiceBusFunctionTests()
{
// pass the mocked objects to initialize class
_inventoryServiceBusFunction = new InventoryServiceBusFunction(_mockLogger.Object, _mockInventoryService.Object);
}
Now that we have an instance of the InventoryServiceBusFunction class, we can use any of the public properties/methods in our tests.
[Fact]
public async Task GivenInventoryRequest_WhenFunctionRuns_ThenInventoryServiceProcessesRequest()
{
Now, remembering the paths from earlier, we can start to create the test cases. We can take the first path and create a [Fact] for it. You want to give your test case a meaningful name. I usually use the style of Given_When_Then to describe what is expected to happen.
Next, I usually add 3 comment sections to my test case:
// arrange
// act
// assert
This allows me to clearly see which parts of the test are doing what.
// act
await _inventoryServiceBusFunction.Run(inventoryRequest);
Next, I would fill in the \\ act section because this will tell me (via Intellisense) what I need to arrange. e.g. above, when hovering my mouse over the Run method, I can see that I need to pass an instance of InventoryRequest.
// arrange
var inventoryRequest = new InventoryRequest
{
Name = "abc123",
Quantity = 2,
Tags = new List<string>
{
"foo"
}
};
In the \\ arrange section, initialize an instance of the InventoryRequest class and set the properties. This can be any data as we aren't really interested in the data itself but more what happens when the code runs.
if (inventoryRequest != null)
{
// path 1
await _inventoryService.ProcessRequest(inventoryRequest);
_logger.LogInformation("HBSI Inventory Queue trigger function processed.");
}
Lastly, the \\ assert section. Here, we want to make assertions on what we expect to happen given the set up of the test. So given the InventoryRequest is not null, we expect the if to evaluate to true and we expect the _inventoryService.ProcessRequest(inventoryRequest) method to be executed.
// assert
_mockInventoryService
.Verify(x => x.ProcessRequest(It.Is<InventoryRequest>(ir => ir.Name == inventoryRequest.Name
&& ir.Quantity == inventoryRequest.Quantity
&& ir.Tags.Contains(inventoryRequest.Tags[0]))));
In Moq, we can use the .Verify() method on the mock object to assert that the method was called. We can use the It.Is<T> syntax to make assertions on the data that is passed to the method.
Here is the full test case for path 1:
[Fact]
public async Task GivenInventoryRequest_WhenFunctionRuns_ThenInventoryServiceProcessesRequest()
{
// arrange
var inventoryRequest = new InventoryRequest
{
Name = "abc123",
Quantity = 2,
Tags = new List<string>
{
"foo"
}
};
// act
await _inventoryServiceBusFunction.Run(inventoryRequest);
// assert
_mockInventoryService
.Verify(x => x.ProcessRequest(It.Is<InventoryRequest>(ir => ir.Name == inventoryRequest.Name
&& ir.Quantity == inventoryRequest.Quantity
&& ir.Tags.Contains(inventoryRequest.Tags[0]))));
}
Then for path 2, you are setting up the test so that the else condition is executed.
[Fact]
public async Task GivenInventoryRequestIsNull_WhenFunctionRuns_ThenInventoryServiceDoesNotProcessRequest()
{
// arrange
InventoryRequest inventoryRequest = null;
// act
await _inventoryServiceBusFunction.Run(inventoryRequest);
// assert
_mockInventoryService
.Verify(x => x.ProcessRequest(It.IsAny<InventoryRequest>()), Times.Never);
}
Note - in the \\ assert here, I am asserting that the await _inventoryService.ProcessRequest(inventoryRequest) method is never called. This is because you want the test to fail in this scenario as the method should only be executed in the if condition. You may also choose to verify that the logger method is called with the correct message.

Checking exceptions with TestCaseData parameters

I'm using NUnit 3 TestCaseData objects to feed test data to tests and Fluent Assertions library to check exceptions thrown.
Typically my TestCaseData object contains two parameters param1 and param2 used to create an instance of some object within the test and upon which I then invoke methods that should/should not throw exceptions, like this:
var subject = new Subject(param1, param2);
subject.Invoking(s => s.Add()).Should().NotThrow();
or
var subject = new Subject(param1, param2);
subject.Invoking(s => s.Add()).Should().Throw<ApplicationException>();
Is there a way to pass NotThrow() and Throw<ApplicationException>() parts as specific conditions in a third parameter in TestCaseData object to be used in the test? Basically I want to parameterize the test's expected result (it may be an exception of some type or no exception at all).
[TestCaseData] is meant for Test Case Data, not for assertions methods.
I would keep the NotThrow and Throw in separate tests to maintain readability.
If they share a lot of setup-logic, I would extract that into shared methods to reduce the size of the test method bodies.
TestCaseData accepts compile time values, whereas TestCaseSource generates them on runtime, which would be necessary to use Throw and NotThrow.
Here's a way to do it by misusing TestCaseSource.
The result is an unreadable test method, so please don't use this anywhere.
Anyway here goes:
[TestFixture]
public class ActionTests
{
private static IEnumerable<TestCaseData> ActionTestCaseData
{
get
{
yield return new TestCaseData((Action)(() => throw new Exception()), (Action<Action>)(act => act.Should().Throw<Exception>()));
yield return new TestCaseData((Action)(() => {}), (Action<Action>)(act => act.Should().NotThrow()));
}
}
[Test]
[TestCaseSource(typeof(ActionTests), nameof(ActionTestCaseData))]
public void Calculate_Success(Action act, Action<Action> assert)
{
assert(act);
}
}
I ended up using this:
using ExceptionResult = Action<System.Func<UserDetail>>;
[Test]
[TestCaseSource(typeof(UserEndpointTests), nameof(AddUserTestCases))]
public void User_Add(string creatorUsername, Role role, ExceptionResult result)
{
var endpoint = new UserEndpoint(creatorUsername);
var person = GeneratePerson();
var request = GenerateCreateUserRequest(person, role);
// Assertion comes here
result(endpoint.Invoking(e => e.Add(request)));
}
private static IEnumerable AddUserTestCases
{
get
{
yield return new TestCaseData(TestUserEmail, Role.User, new ExceptionResult(x => x.Should().Throw<ApplicationException>())
.SetName("{m} (Regular User => Regular User)")
.SetDescription("User with Regular User role cannot add any users.");
yield return new TestCaseData(TestAdminEmail, Role.Admin, new ExceptionResult(x => x.Should().NotThrow())
)
.SetName("{m} (Admin => Admin)")
.SetDescription("User with Admin role adds another user with Admin role.");
}
}
No big issues with readability, besides, SetName() and SetDescription() methods in the test case source help with that.

How to perform async ModelState validation with FluentValidation in Web API using .NET Core

This question is a follow up to this post - How to perform async ModelState validation with FluentValidation in Web API?.
I was wondering if FluentValidation has a way to perform async ModelState validation in .net core web api. I have a FluentValidation Validator class which contains async validation methods such as "MustAsync", which means in my business service class I call the validator manually using "ValidateAsync". I also want to use this same validator class to validate the model coming in from the request. I went through the documents and read that the only way to do this is to manually call the "ValidateAsync()" method since the .net pipeline is synchronous. I would rather not manually have to call this method from within my controller, I would prefer to either register it in the startup (have the framework automatically call the the validator on my model) or decorate my request model with the validator.
Has anyone been able to achieve this?
Thanks!
Based on the linked question, I've adapted the code slightly to be compatible with ASP.NET Core (2.2, in my case). In general, this is using the IAsyncActionFilter interface. You can read about it in the official docs.
public class ModelValidationActionFilter : IAsyncActionFilter
{
private readonly IValidatorFactory _validatorFactory;
public ModelValidationActionFilter(IValidatorFactory validatorFactory) => _validatorFactory = validatorFactory;
public async Task OnActionExecutionAsync(ActionExecutingContext context, ActionExecutionDelegate next)
{
var allErrors = new Dictionary<string, object>();
// Short-circuit if there's nothing to validate
if (context.ActionArguments.Count == 0)
{
await next();
return;
}
foreach (var (key, value) in context.ActionArguments)
{
// skip null values
if (value == null)
continue;
var validator = _validatorFactory.GetValidator(value.GetType());
// skip objects with no validators
if (validator == null)
continue;
// validate
var result = await validator.ValidateAsync(value);
// if it's valid, continue
if (result.IsValid) continue;
// if there are errors, copy to the response dictonary
var dict = new Dictionary<string, string>();
foreach (var e in result.Errors)
dict[e.PropertyName] = e.ErrorMessage;
allErrors.Add(key, dict);
}
if (allErrors.Any())
{
// Do anything you want here, if the validation failed.
// For example, you can set context.Result to a new BadRequestResult()
// or implement the Post-Request-Get pattern.
}
else
await next();
}
}
If you want to apply this filter globally, you can add the filter to the AddMvc call in your Startup class. For example:
services.AddMvc(options =>
{
options.Filters.Add<ModelValidationActionFilter>();
// uncomment the following line, if you want to disable the regular validation
// options.ModelValidatorProviders.Clear();
});
I had trouble getting the code in #nachtjasmin's answer to work with newer versions of FluentValidation. Specifically, the trouble is that ValidateAsync now takes an IValidationContext instead of the model being validated, and the context can't be created without knowing the type of the model at compile time.
Eventually I stumbled upon this answer, which points out that the exact type is not important and uses object instead.
So, instead of:
var result = await validator.ValidateAsync(value);
You can use:
var context = new ValidationContext<object>(value);
var result = await validator.ValidateAsync(context);
Based on the answer above by #nachtjasmin, you can add this in two ways,
Using AddMvc
services.AddControllersWithViews(options =>
{
options.Filters.Add<FluentValidationActionFilter>();
});
Using AddControllersWithViews
services.AddControllersWithViews(options =>
{
options.Filters.Add<FluentValidationActionFilter>();
});
If your's is just a Web API and you don't have any Razor pages involved, then you can consider using AddControllersWithViews over AddMvc, as the AddMvc uses the AddControllersWithViews internally and add the services.AddRazorPages() on top of that.
You can see this info here for AddMvc and here for AddControllersWithViews

How to resolve a Collection of Types from within the IoC Container

We're using MvvmCross in our app, and using the MvxSimpleIoCContainer
In the app startup, we register all of our Migrations.
it's easy do do since all migrations inherit from IMigration
typeof (IMigration)
.Assembly
.CreatableTypes()
.Inherits<IMigration>()
.AsTypes()
.RegisterAsLazySingleton();
After the migrations are registered, we need to run them consecutively, and therefore the MigrationRunner looks a little something like this.
Mvx.Resolve<IMigrationRunner>().RunAll(SystemRole.Client, new List<IMigration>
{
Mvx.IocConstruct<Migration001>(),
Mvx.IocConstruct<Migration002>()
});
as you can see, I'm explicitely constructing each Migration using Mvx. This get's tedious and is prone to mistakes when a bunch of migrations end up in the app.
What I'd prefer to be able to do is resolve the entire collection in one fell swoop, and not have to touch it every time I create a new Migration.
Is there a way to do this via MvvmCross?
Pseudo Code
Mvx.Resolve<IMigrationRunner>()
.RunAll(SystemRole.Client, Mvx.ResolveAll<IMigration>());
I would use LINQ to get the list of types. Unfortunately there's no way to get a list of registered types, so you'll have to enumerate the types again like you do for registration. You can even sort by type name. Now that you have a list of types, you can create a new list of instantiated/resolved types to pass into RunAll(). Something like:
var migrationTypes = typeof (IMigration)
.Assembly
.CreatableTypes()
.Inherits<IMigration>()
.AsTypes()
.OrderBy(t => t.Name)
.ToList();
Mvx.Resolve<IMigrationRunner>()
.RunAll(SystemRole.Client,
migrationTypes.Select(t => Mvx.Resolve(t)).ToList());
This is "browser" code, so no guarantees, but you get the gist.
Ok, so reflection is the answer to this problem for now, and eventually, I'd like to either extend our custom MvxServiceLocator : IServiceLocator to include something like
public IEnumerable<object> GetAllInstances(Type serviceType){...}
but for now I've just got a RunMigrations() method in the app
private void RunMigrations()
{
var migrationType = typeof (IMigration); // IMigration is in a separate assembly
var migrations = GetType().Assembly
.GetTypes()
.Where(
t => migrationType.IsAssignableFrom(t) && !t.IsAbstract)
.OrderBy(t => t.Name)
.Select(m => _serviceLocator.GetInstance(m) as IMigration)
.ToList();
var migrationRunner = new MigrationRunner(Mvx.Resolve<IDbProvider>());
migrationRunner.RunAll(SystemRole.Client, migrations);
}
where _serviceLocator.GetInstance(m) just lives in our custom MvxServiceLocator
public object GetInstance(Type serviceType)
{
return _ioCProvider.Resolve(serviceType);
}
Edit: here's how I extended our service locator wrapper.
public class MvxServiceLocator : IServiceLocator
{
private readonly IMvxIoCProvider _ioCProvider;
public MvxServiceLocator(IMvxIoCProvider ioCProvider)
{
_ioCProvider = ioCProvider;
}
public IEnumerable<TService> GetAllInstances<TService>()
{
var serviceType = typeof(TService);
var registrations = GetType().Assembly
.GetTypes()
.Where(
t => serviceType.IsAssignableFrom(t) && !t.IsAbstract)
.Select(m => (TService)_ioCProvider.Resolve(m));
return registrations;
}
}

Rhino AutoMocker and Stubs

I am using Joshua Flanagan article “Auto mocking Explained” as a guide. In the article there is a section called “The same tests, with automocker would look like this”. I used this information to build code to run the automocker.
As you can see below answer is a list returned from the BLL. Answer does have one row in it; however, all fields are null. So the test for boo fails. Any tips and hints would be greatly appreciated.
[Test]
public void GetStaffListAndRolesByTeam_CallBLLWithDALStub()
{
// Build List<> data for stub
List<StaffRoleByTeamCV> stubData = new List<StaffRoleByTeamCV>();
StaffRoleByTeamCV stubRow = new StaffRoleByTeamCV();
stubRow.Role = "boo";
stubRow.StaffId = 12;
stubRow.StaffName = "Way Cool";
stubData.Add(stubRow);
// create the automocker
var autoMocker = new RhinoAutoMocker<PeteTestBLL>();
// get instance of test class (the BLL)
var peteTestBllHdl = autoMocker.ClassUnderTest;
// stub out call to DAL inside of BLL
autoMocker.Get<IPeteTestDAL>().Stub(c => c.GetStaffListAndRolesByTeam("4146")).Return(stubData);
// make call to BLL this should return stubData
List<StaffRoleByTeamCV> answer = peteTestBllHdl.GetStaffListAndRolesByTeam("4146");
// do simple asserts to test stubData present
// this passes
Assert.IsTrue(1 == answer.Count, "Did not find any rows");
// this fails
Assert.IsTrue(answer[0].Role == "boo", "boo was not found");
}
I tried using MockMode.AAA but still no joy
An new version of AutoMocker (1.0.3) is available. The new version supports relay mode as in this example..
[TestMethod]
public void ShouldSupportOrderedTest()
{
//Arrange
var autoMocker = new RhinoAutoMocker<CustomerUpdater>();
var mockRepository = autoMocker.Repository;
using (mockRepository.Ordered())
{
autoMocker.Get<ICustomerDataProvider>().Expect(x => x.GetCustomer(Arg<int>.Is.Anything)).Return(new CustomerItem());
autoMocker.Get<ICustomerDataProvider>().Expect(x => x.UpdateCustomer(Arg<CustomerItem>.Is.Anything));
autoMocker.Get<ILogWriter>().Expect(x => x.Write(Arg<string>.Is.Anything));
autoMocker.Get<ILogWriter>().Expect(x => x.Write(Arg<string>.Is.Anything));
autoMocker.Get<IMailSender>().Expect(x => x.SendMail(Arg<string>.Is.Anything, Arg<string>.Is.Anything));
}
//Act
autoMocker.ClassUnderTest.UpdateCustomerName(1, "Frank", "frank#somecompany.com");
//Assert
ExceptionAssert.Throws<ExpectationViolationException>(mockRepository.VerifyAll,"IMailSender.SendMail(anything, anything); Expected #1, Actual #0.\r\nILogWriter.Write(anything); Expected #1, Actual #0.\r\n");
}
I haven't tried, but this article suggests that by default all the mocks created by automocker are not replayed:
http://www.lostechies.com/blogs/joshuaflanagan/archive/2008/09/25/arrange-act-assert-with-structuremap-rhinoautomocker.aspx
Yes that was true for the previous version. But was changed to support ordered tests in version 1.0.3.