Testing Anti Patterns

Testing Anti-patterns (.NET)

Mocking DbContext is a smell

// non-compiling: illustrative
// WRONG: mocking what you don't own, re-implementing the ORM in Moq
var mockSet = new Mock<DbSet<Order>>();
mockSet.As<IQueryable<Order>>().Setup(m => m.Provider).Returns(orders.Provider);
mockSet.As<IQueryable<Order>>().Setup(m => m.Expression).Returns(orders.Expression);
mockSet.As<IQueryable<Order>>().Setup(m => m.GetEnumerator()).Returns(orders.GetEnumerator());

var mockCtx = new Mock<ApplicationDbContext>();
mockCtx.Setup(c => c.Orders).Returns(mockSet.Object);

var service = new OrderService(mockCtx.Object);

This tests LINQ-to-Objects semantics, not SQL translation. EF Core's in-memory provider has the same flaw: no relational behavior, no transactions, no constraint enforcement. It catches only the simplest bugs and gives false confidence.

Correct options, in order of preference:

1. Testcontainers with the production engine; one container per test class/collection, Respawn or transactions for cleanup.
2. SQLite in-memory only when the model is compatible and speed genuinely blocks you - and keep a Testcontainers suite for the critical queries.
3. If a service must be tested without a database, mock a repository interface YOU defined - never the DbContext.

// CORRECT: repository abstraction you control
public interface IOrderRepository
{
    Task<Order?> GetByIdAsync(int id, CancellationToken ct = default);
    Task AddAsync(Order order, CancellationToken ct = default);
    Task SaveChangesAsync(CancellationToken ct = default);
}

// Test doubles implement YOUR interface, not EF Core's
public class FakeOrderRepository : IOrderRepository
{
    private readonly List<Order> _orders = [];
    
    public Task<Order?> GetByIdAsync(int id, CancellationToken ct) =>
        Task.FromResult(_orders.FirstOrDefault(o => o.Id == id));
    
    public Task AddAsync(Order order, CancellationToken ct)
    {
        _orders.Add(order);
        return Task.CompletedTask;
    }
    
    public Task SaveChangesAsync(CancellationToken ct) => Task.CompletedTask;
}

Asserting implementation details

// non-compiling: illustrative
// WRONG: testing the mock, not the behavior
[Fact]
public void UpdateOrder_WhenCalled_CallsSaveChangesOnMock()
{
    var mockRepo = new Mock<IOrderRepository>();
    var service = new OrderService(mockRepo.Object);
    
    service.UpdateOrder(1, "New status");
    
    mockRepo.Verify(r => r.SaveChangesAsync(default), Times.Once); // 🚫 brittle
}

This test breaks when you refactor the service to call SaveChanges in a different method. It also doesn't verify the actual business outcome.

Assert outcomes, not interactions:

// CORRECT: assert the state change
[Fact]
public async Task UpdateOrder_WhenCalled_UpdatesOrderStatus()
{
    // Arrange
    var order = new Order { Id = 1, Status = OrderStatus.Pending };
    var repo = new FakeOrderRepository();
    await repo.AddAsync(order, default);
    var service = new OrderService(repo);
    
    // Act
    await service.UpdateOrder(1, OrderStatus.Processing, default);
    
    // Assert
    var updated = await repo.GetByIdAsync(1, default);
    updated.Status.Should().Be(OrderStatus.Processing); // ✅ verifies behavior
}

Task.Delay timing tests are flaky

// non-compiling: illustrative
// WRONG: timing-dependent test that fails unpredictably
[Fact]
public async Task ProcessQueue_WhenCalled_ProcessesAfterDelay()
{
    var queue = new BackgroundQueue();
    queue.Enqueue("item");
    
    await Task.Delay(100); // 🚫 100ms is arbitrary
    
    queue.Count.Should().Be(0);
}

Timing tests are environment-dependent: CI runners are slower than dev machines, some machines are slower than others. They pass on your machine and fail in CI - or vice versa. They also slow down the suite and make it non-deterministic.

Use synchronization primitives instead:

// CORRECT: await the actual condition
[Fact]
public async Task ProcessQueue_WhenCalled_ProcessesItem()
{
    var queue = new BackgroundQueue();
    var cts = new CancellationTokenSource(TimeSpan.FromSeconds(5));
    queue.Enqueue("item");
    
    // Wait for the queue to be processed
    while (queue.Count > 0 && !cts.IsCancellationRequested)
    {
        await Task.Delay(10, cts.Token); // small poll interval
    }
    
    queue.Count.Should().Be(0);
}

// Even better: expose a completion signal
public class BackgroundQueue
{
    private readonly Channel<string> _channel = Channel.CreateUnbounded<string>();
    public int Count => _channel.Reader.Count;
    
    public async Task ProcessAsync(CancellationToken ct)
    {
        await foreach (var item in _channel.Reader.ReadAllAsync(ct))
        {
            await ProcessItemAsync(item, ct);
        }
    }
}

// Test uses the signal
[Fact]
public async Task ProcessQueue_WhenCalled_ProcessesItem()
{
    var queue = new BackgroundQueue();
    var processing = queue.ProcessAsync(default);
    queue.Enqueue("item");
    
    await queue.WaitForCompletionAsync(); // ✅ deterministic
    
    // Assert state...
}

Missing cancellation tokens in resilience tests

// non-compiling: illustrative
// WRONG: resilience policy without cancellation token flow
[Fact]
public async Task MakePayment_WhenServiceUnavailable_Retries()
{
    var handler = new MockHttpMessageHandler(HttpStatusCode.ServiceUnavailable);
    var client = new HttpClient(handler);
    var resilience = new ResiliencePipelineBuilder()
        .AddRetry(new RetryStrategyOptions { MaxRetryAttempts = 3 })
        .Build();
    
    await resilience.ExecuteAsync(async _ => 
        await client.GetAsync("https://api.example.com/payments"), 
        CancellationToken.None); // 🚫 cancellation ignored
}

A timeout policy without the request token keeps calling dependencies for clients that already disconnected. The cancellation token is part of the contract between caller and callee.

Flow the caller's token:

// CORRECT: propagate cancellation
[Fact]
public async Task MakePayment_WhenClientCancels_StopsRetrying()
{
    var handler = new MockHttpMessageHandler(HttpStatusCode.ServiceUnavailable);
    var client = new HttpClient(handler);
    var cts = new CancellationTokenSource();
    
    var resilience = new ResiliencePipelineBuilder()
        .AddRetry(new RetryStrategyOptions { MaxRetryAttempts = 100 })
        .Build();
    
    // Start the operation
    var task = resilience.ExecuteAsync(async ct => 
        await client.GetAsync("https://api.example.com/payments", ct),
        cts.Token);
    
    // Cancel immediately
    cts.CancelAfter(TimeSpan.Zero);
    
    await Assert.ThrowsAnyAsync<OperationCanceledException>(() => task);
}

See also: http-resilience-and-outbound-calls for timeout and retry policy rules.

Testing async void and fire-and-forget

// non-compiling: illustrative
// WRONG: async void cannot be awaited, exceptions are lost
public class OrderController : ControllerBase
{
    [HttpPost("orders")]
    public async void Post(OrderDto dto) // 🚫 async void
    {
        await _service.CreateOrderAsync(dto);
    }
}

// Test cannot catch the exception
[Fact]
public void Post_WhenInvalid_ReturnsBadRequest()
{
    var controller = new OrderController(_service);
    controller.ModelState.AddModelError("Name", "Required");
    
    controller.Post(new OrderDto()); // Exception swallowed
    
    Assert.Equal(400, controller.Response.StatusCode);
}

Async void exceptions are lost - they don't propagate to the test. They also make the controller non-composable (you can't await the endpoint).

Return Task instead:

// CORRECT: async Task
[HttpPost("orders")]
public async Task<IActionResult> Post(OrderDto dto) // ✅ async Task
{
    if (!ModelState.IsValid)
        return BadRequest(ModelState);
    
    await _service.CreateOrderAsync(dto, HttpContext.RequestAborted);
    return Created(...);
}

Testing synchronous code with async

// non-compiling: illustrative
// WRONG: sync-over-async in tests
[Fact]
public void GetOrder_WhenCalled_ReturnsOrder()
{
    var service = new OrderService(_repo);
    var order = service.GetOrder(1).Result; // 🚫 .Result deadlocks under load
    
    Assert.NotNull(order);
}

.Result and .Wait() in tests cause deadlocks when the async code needs the synchronization context (which tests have). The thread pool is blocked waiting for itself.

Use async all the way:

// CORRECT: async test
[Fact]
public async Task GetOrder_WhenCalled_ReturnsOrder()
{
    var service = new OrderService(_repo);
    var order = await service.GetOrderAsync(1, default); // ✅ proper async
    
    order.Should().NotBeNull();
}

Testing private methods directly

// non-compiling: illustrative
// WRONG: testing implementation, not behavior
public class OrderService
{
    private bool IsValid(Order order) => order.Total > 0;
}

[Fact]
public void IsValid_WhenTotalPositive_ReturnsTrue()
{
    var service = new OrderService();
    var method = typeof(OrderService).GetMethod("IsValid", BindingFlags.NonPublic | BindingFlags.Instance);
    var result = (bool)method.Invoke(service, [new Order { Total = 100 }]); // 🚫 brittle
    
    Assert.True(result);
}

Private methods are an implementation detail. If the behavior matters, expose it through a public method or extract a class. Testing private methods couples tests to refactoring.

Test through public seams:

// CORRECT: test the public behavior
public class OrderService
{
    public Result ValidateOrder(Order order) // ✅ public method
    {
        if (order.Total <= 0)
            return Result.Fail("Total must be positive");
        return Result.Ok();
    }
}

[Fact]
public void ValidateOrder_WhenTotalZero_ReturnsFailure()
{
    var service = new OrderService();
    var result = service.ValidateOrder(new Order { Total = 0 });
    
    result.IsSuccess.Should().BeFalse();
}

Testing concurrency with locks instead of synchronization

// non-compiling: illustrative
// WRONG: testing lock behavior instead of business outcome
[Fact]
public void ProcessOrdersConcurrently_WhenCalled_UsesLocks()
{
    var service = new ConcurrentOrderService();
    var tasks = Enumerable.Range(1, 100)
        .Select(i => Task.Run(() => service.ProcessOrder(i)))
        .ToArray();
    
    Task.WaitAll(tasks); // 🚫 tests locks, not correctness
    
    service.OrdersProcessedCount.Should().Be(100);
}

This tests that locks exist, not that the business rule is correct. It also doesn't verify thread safety under real contention.

Test the outcome, not the mechanism:

// CORRECT: test the invariant
[Fact]
public async Task ProcessOrdersConcurrently_WhenCalled_ProcessesAllOrders()
{
    var service = new ConcurrentOrderService();
    var orders = Enumerable.Range(1, 100).Select(i => new Order { Id = i }).ToList();
    
    var tasks = orders.Select(o => service.ProcessOrderAsync(o, default));
    await Task.WhenAll(tasks);
    
    var processed = await service.GetProcessedOrdersAsync(default);
    processed.Should().HaveCount(100); // ✅ verifies the invariant
}

See also: concurrency-and-shared-state for lock discipline and async mutual exclusion rules.

Testing background services without proper cleanup

// non-compiling: illustrative
// WRONG: background service test without proper lifecycle
[Fact]
public async Task BackgroundService_WhenStarted_ProcessesItems()
{
    var service = new OrderProcessingService(_repo);
    var host = new HostBuilder()
        .ConfigureServices(sc => sc.AddHostedService(_ => service))
        .Build();
    
    await host.StartAsync();
    await Task.Delay(100); // 🚫 arbitrary delay
    await host.StopAsync();
    
    _repo.Verify(r => r.SaveChangesAsync(default), Times.AtLeastOnce);
}

Background service tests need proper lifecycle management: start, wait for work, verify, stop. Arbitrary delays make tests flaky and slow.

Use synchronization primitives:

// CORRECT: use completion signals
[Fact]
public async Task BackgroundService_WhenStarted_ProcessesItems()
{
    var service = new OrderProcessingService(_repo);
    var host = new HostBuilder()
        .ConfigureServices(sc => sc.AddHostedService(_ => service))
        .Build();
    
    await host.StartAsync();
    
    // Signal that work is complete
    await service.WaitForProcessingCompleteAsync(default);
    
    await host.StopAsync(TimeSpan.FromSeconds(5));
    
    _repo.Verify(r => r.SaveChangesAsync(default), Times.AtLeastOnce);
}

See also: background-work-and-hosted-services for background service lifecycle rules.

Testing HTTP clients without IHttpClientFactory

// non-compiling: illustrative
// WRONG: direct HttpClient construction
[Fact]
public async Task PaymentClient_WhenCalled_PostsToApi()
{
    var handler = new MockHttpMessageHandler();
    var client = new HttpClient(handler) { BaseAddress = new Uri("https://api.example.com") }; // 🚫 socket exhaustion
    var client = new PaymentClient(client);
    
    await client.ProcessPaymentAsync(new PaymentRequest(), default);
    
    handler.Protected().Verify(
        "SendAsync",
        Times.Once(),
        ItExpr.IsAny<HttpRequestMessage>(),
        ItExpr.IsAny<CancellationToken>());
}

Direct new HttpClient() construction causes socket exhaustion (each instance owns its handler and connections). In tests, this leaks resources between tests.

Use IHttpClientFactory in tests too:

// CORRECT: factory-based construction
[Fact]
public async Task PaymentClient_WhenCalled_PostsToApi()
{
    var handler = new MockHttpMessageHandler();
    var services = new ServiceCollection();
    services.AddHttpClient<PaymentClient>(c => 
        c.BaseAddress = new Uri("https://api.example.com"))
        .AddHttpMessageHandler(() => handler);
    
    var provider = services.BuildServiceProvider();
    var client = provider.GetRequiredService<PaymentClient>();
    
    await client.ProcessPaymentAsync(new PaymentRequest(), default);
    
    handler.Protected().Verify(
        "SendAsync",
        Times.Once(),
        ItExpr.IsAny<HttpRequestMessage>(),
        ItExpr.IsAny<CancellationToken>());
}

See also: http-resilience-and-outbound-calls for client construction and resilience policy rules.

Testing without proper test data builders

// non-compiling: illustrative
// WRONG: test with raw constructors and property soup
[Fact]
public async Task CreateOrder_WhenValid_CreatesOrder()
{
    var order = new Order
    {
        Id = 1,
        CustomerId = 42,
        Total = 100.50m,
        Status = OrderStatus.Pending,
        CreatedAt = DateTime.UtcNow,
        Items = [
            new OrderItem { ProductId = 1, Quantity = 2, Price = 50.25m },
            new OrderItem { ProductId = 2, Quantity = 1, Price = 0.00m }
        ]
    };
    
    var repo = new FakeOrderRepository();
    var service = new OrderService(repo);
    
    await service.CreateOrderAsync(order, default);
    
    var saved = await repo.GetByIdAsync(1, default);
    saved.Should().NotBeNull();
}

Raw constructors create brittle tests that break on every schema change. Setup duplication rots the suite.

Use builders/object mothers:

// CORRECT: builder pattern
public static class OrderBuilder
{
    public static OrderBuilder Default => new OrderBuilder()
        .WithCustomer(42)
        .WithItem("Widget", 2, 50.25m);
    
    private readonly Order _order = new();
    
    public OrderBuilder WithCustomer(int id)
    {
        _order.CustomerId = id;
        return this;
    }
    
    public OrderBuilder WithItem(string name, int qty, decimal price)
    {
        _order.Items.Add(new OrderItem { ProductId = 1, Quantity = qty, Price = price });
        _order.Total += qty * price;
        return this;
    }
    
    public Order Build() => _order;
}

[Fact]
public async Task CreateOrder_WhenValid_CreatesOrder()
{
    var order = OrderBuilder.Default.Build();
    var repo = new FakeOrderRepository();
    var service = new OrderService(repo);
    
    await service.CreateOrderAsync(order, default);
    
    var saved = await repo.GetByIdAsync(1, default);
    saved.Should().NotBeNull();
    saved.Total.Should().Be(100.50m);
}

Structural rules

• One assertion per test named for the behavior: UpdateOrder_WhenAlreadyCancelled_ReturnsConflict. A test named Test1 fails uninformatively.
• Deterministic tests: inject TimeProvider (assert against FakeTimeProvider), fixed seeds, no Task.Delay-based synchronization - await a real condition or use TaskCompletionSource.
• No logic in tests: a loop or if in a test needs its own test. Parameterize with [Theory]/[InlineData] instead.
• Test data setup via builders/object mothers, not 30 lines of property soup copied between tests - setup duplication is why test suites rot after schema changes.
• Flaky tests are P1: quarantine within a day, fix or delete within a sprint. A suite people retry is a suite people ignore.

See also

• testing-strategy - Decide what to test at each layer of a .NET service
• concurrency-and-shared-state - Lock discipline and async mutual exclusion
• http-resilience-and-outbound-calls - Timeout and retry policy rules
• background-work-and-hosted-services - Background service lifecycle rules