OVERVIEW

In .NET it’s really easy to create your own interfaces and implementations. Likewise, it’s seemingly effortless to register them for dependency injection. But it is not always obvious how to override existing implementations. Let’s discuss various aspects of “dependency injection” and how you can override the “framework-provided services”.

As an example, let’s take a recent story on our product backlog for building a security audit of login attempts. The story involved the capture of attempted usernames along with their corresponding IP addresses. This would allow system administrators to monitor for potential attackers. This would require our ASP.NET Core application to have custom logging implemented.

LOGGING

Luckily ASP.NET Core Logging is simple to use and is a first-class citizen within ASP.NET Core.

In the Logging repository there is an extension method namely AddLogging, here is what it looks like:

public static IServiceCollection AddLogging(this IServiceCollection services)
{
    if (services == null)
    {
        throw new ArgumentNullException(nameof(services));
    }

    services.TryAdd(ServiceDescriptor.Singleton<ILoggerFactory, LoggerFactory>());
    services.TryAdd(ServiceDescriptor.Singleton(typeof(ILogger<>), typeof(Logger<>)));

    return services;
}

As you can see, it is rather simple. It adds two ServiceDescriptor instances to the IServiceCollection, effectively registering the given service type to the corresponding implementation type.

FOLLOWING THE RABBIT DOWN THE HOLE

When you create a new ASP.NET Core project from Visual Studio, all the templates follow the same pattern. They have the Program.cs file with a Main method that looks very similar to this:

public static void Main(string[] args)
{
    var host = new WebHostBuilder()
        .UseKestrel()
        .UseContentRoot(Directory.GetCurrentDirectory())
        .UseIISIntegration()
        .UseStartup<Startup>()
        .UseApplicationInsights()
        .Build();

    host.Run();
}

TEMPLATES 

One thing that is concerning about a template like this is that the IWebHost is an IDisposable, so why then is this statement not wrapped in a using you ask? The answer is that the Run extension method internally wraps itself in a using. If you were wondering where the AddLogging occurs, it is a result of invoking the Build function.

[ Microsoft.AspNetCore.Hosting.WebHostBuilder ]
    public IWebHost Build() ...
        private IServiceCollection BuildCommonServices() ...
            creates services then invokes services.AddLogging()

A FEW WORDS ON THE SERVICE DESCRIPTOR

The ServiceDescriptor class is an object that describes a service, and this is used by dependency injection. In other words, instances of the ServiceDescriptor are descriptions of services. The ServiceDescriptor class exposes several static methods that allow its instantiation.

The ILoggerFactory interface is registered as a ServiceLifetime.Singleton and its implementation is mapped to the LoggerFactory. Likewise, the generic type typeof(ILogger<>) is mapped to typeof(Logger<>). This is just one of the several key “Framework-Provided Services” that are registered.

PUTTING IT TOGETHER

Now we know that the framework is providing all implementations of ILogger<T>, and resolving them as their Logger<T>. We also know that we could write our own implementation of the ILogger<T>interface. Being that this is open-source we can look to their implementation for inspiration.

public class RequestDetailLogger<T> : ILogger<T>
{
    private readonly ILogger _logger;

    public RequestDetailLogger(ILoggerFactory factory,
                               IRequestCategoryProvider requestCategoryProvider)
    {
        if (factory == null)
        {
            throw new ArgumentNullException(nameof(factory));
        }
        if (requestCategoryProvider == null)
        {
            throw new ArgumentNullException(nameof(requestCategoryProvider));
        }

        var category = requestDetailCategoryProvider.CreateCategory<T>();
        _logger = factory.CreateLogger(category);
    }

    IDisposable ILogger.BeginScope<TState>(TState state)
        => _logger.BeginScope(state);

    bool ILogger.IsEnabled(LogLevel logLevel)
        => _logger.IsEnabled(logLevel);

    void ILogger.Log<TState>(LogLevel logLevel,
                             EventId eventId,
                             TState state,
                             Exception exception,
                             Func<TState, Exception, string> formatter)
        => _logger.Log(logLevel, eventId, state, exception, formatter);
}

The IRequestCategoryProvider is defined and implemented as follows:

using static Microsoft.Extensions.Logging.Abstractions.Internal.TypeNameHelper;

public interface IRequestCategoryProvider
{
    string CreateCategory<T>();
}

public class RequestCategoryProvider : IRequestCategoryProvider
{
    private readonly IPrincipal _principal;
    private readonly IPAddress _ipAddress;

    public RequestCategoryProvider(IPrincipal principal,
                                   IPAddress ipAddress)
    {
        _principal = principal;
        _ipAddress = ipAddress;
    }

    public string CreateCategory<T>()
    {
        var typeDisplayName = GetTypeDisplayName(typeof(T));

        if (_principal == null || _ipAddress == null)
        {
            return typeDisplayName;
        }

        var username = _principal?.Identity?.Name;
        return $"User: {username}, IP: {_ipAddress} {typeDisplayName}";
    }
}

If you’re curious how to get the IPrincipal and IPAddress into this implementation (with DI). It is pretty straight-forward. In the Startup.ConfigureServices method do the following:

public void ConfigureServices(IServiceCollection services)
{
    // ... omitted for brevity

    services.AddTransient<IRequestCategoryProvider, RequestCategoryProvider>();
    services.AddTransient<IHttpContextAccessor, HttpContextAccessor>();
    services.AddTransient<IPrincipal>(
        provider => provider.GetService<IHttpContextAccessor>()
                           ?.HttpContext
                           ?.User);
    services.AddTransient<IPAddress>(
        provider => provider.GetService<IHttpContextAccessor>()
                           ?.HttpContext
                           ?.Connection
                           ?.RemoteIpAddress);
}

Finally, we can Replace the implementations for the ILogger<T> by using the following:

public void ConfigureServices(IServiceCollection services)
{
    // ... omitted for brevity
    services.Replace(ServiceDescriptor.Transient(typeof(ILogger<>),
                                                 typeof(RequestDetailLogger<>)));
}

Notice that we replace the framework-provided service as a ServiceLifetime.Transient. Opposed to the default ServiceLifetime.Singleton. This is more or less an extra precaution. We know that with each request we get the HttpContext from the IHttpContextAccessor, and from this we have the User. This is what is passed to each ILogger<T>.

CONCLUSION

This approach is valid for overriding any of the various framework-provided service implementations. It is simply a matter of knowing the correct ServiceLifetime for your specific needs. Likewise, it is a good idea to leverage the open-source libraries of the framework for inspiration. With this you can take finite control of your web-stack.