Azure Key Vault + MSI = failing Web API integration tests

Introduction

In a number of our ASP.NET Core Web APIs we’re using Azure Key Vault for keeping secrets such as connection strings and authentication credentials out of source control.

Initially we were using Azure Key Vault with a clientid and clientsecret, which are stored in appsettings.json like so:

  "KeyVault": {
    "Name": "MyApplicationDev",
    "ClientId": "abcdefg123-b292-4177-ba53-858227a9143c",
    "ClientSecret": "5m9g9cpuNc31abcJZcjkfP9/pDwJgQ+T82t/qCey7Nc="
  },

But this begs the question – what happens if the Key Vault ClientId and ClientSecret get compromised? To prevent this you can setup Key Vault to use Managed Service Identity (MSI). With that in place you don’t need to have the ClientId and ClientSecret in your appsettings.json, instead you only need the KeyVault url:

  "KeyVaultSettings": {
    "Url": "https://myapplicationdev.vault.azure.net"
  },

This is usually configured in the WebApi’s Program.cs, via something like:

public static IWebHostBuilder CreateWebHostBuilder(string[] args) =>
    WebHost.CreateDefaultBuilder(args)
        .ConfigureAppConfiguration((context, builder) =>
        {
            builder.SetBasePath(context.HostingEnvironment.ContentRootPath)
                    .AddEnvironmentVariables();

            var config = builder.Build();
            var tokenProvider = new AzureServiceTokenProvider();
            var keyvaultClient = new KeyVaultClient((authority, resource, scope)
               => tokenProvider.KeyVaultTokenCallback(authority, resource, scope));
                    
            builder.AddAzureKeyVault(config["KeyVaultSettings:Url"], keyvaultClient, new DefaultKeyVaultSecretManager());
        })
        .UseStartup<Startup>();

Under the covers, this will add an AzureKeyVaultConfigurationSource to the registered list of IConfigurationBuilders (as well as do other things).

The problem

In ASP.NET 2.1 we are using the new WebApplicationFactory<T> for running integration tests against an in-memory TestServer. When you run these tests locally it’ll use your (i.e. you, the developer’s) AD credentials to authenticate against the Key Vault, and so the tests should pass if you have access to the Key Vault. However, when the tests run on the Build server (as part of your CI pipeline) then they’ll probably fail because the Build agent does not have access to the Key Vault.

My solution

Firstly, I didn’t want to change the code of the system under test (i.e. the Web API), i.e. by adding configuration to determine whether to use Key Vault or not. Although thinking about it, that might have been easier! But it feels a bit dirty to change the application code to make integration tests easier.

The approach I took was to remove the AzureKeyVaultConfigurationSource from the list of IConfigurationBuilders from the system under test, in my custom WebApplicationFactory<T>, i.e:

protected override void ConfigureWebHost(IWebHostBuilder builder)
{
    builder.ConfigureAppConfiguration((_, configurationBuilder) =>
    {
        var keyVaultSource = configurationBuilder.Sources.FirstOrDefault(cs => cs.GetType().Name == "AzureKeyVaultConfigurationSource");
        if (keyVaultSource != null)
        {
            configurationBuilder.Sources.Remove(keyVaultSource);
        }
    });
}

This way, the code in the system under test is unchanged, instead we are just changing the configuration of the TestServer prior to it starting. Hope that helps someone.

Advertisement

Where to download vstest.console.exe

Often when I’m debugging builds or releases on VSTS, I will see that it uses the vstest.console.exe command line for running tests. Sometimes I need to run vstest.console.exe locally so that I can debug test run failures.

FYI, it is installed as part of Visual Studio 2017, and if you run the Visual Studio Developer Command prompt you will be able to run it from there. The executable lives in C:\Program Files (x86)\Microsoft Visual Studio\2017\Enterprise\Common7\IDE\CommonExtensions\Microsoft\TestWindow

If you don’t have VS2017 installed and you need to download it so that you can run it locally, you can find it in the Microsoft.TestPlatform NuGet package.

Once you’ve downloaded the NuGet package, rename it from microsoft.testplatform.15.8.0.nupkg to microsoft.testplatform.15.8.0.zip. Then open it, and you’ll find vstest.console.exe in the tools\net451\Common7\IDE\Extensions\TestPlatform folder.

Add an authorization header to your swagger-ui with Swashbuckle (revisited)

Just over a year ago I blogged a simple way to add an authorization header to your swagger-ui with Swashbuckle. Although that works, Swagger-UI and Swashbuckle support a better way, which I’ll describe below.

Before starting I assume you’ve already got OAuth2 setup correctly on your application (using bearer tokens), and you have decorated your controllers and actions with [Authorize] attributes. If you haven’t, that is beyond the scope of this blog post. Here all I’m doing is explaining how to configure Swashbuckle.

First, you need to tell Swashbuckle what security your API has:

services.AddSwaggerGen(options =>
{
    options.AddSecurityDefinition("oauth2", new ApiKeyScheme
    {
        Description = "Standard Authorization header using the Bearer scheme. Example: \"bearer {token}\"",
        In = "header",
        Name = "Authorization",
        Type = "apiKey"
    });

This adds a securityDefinition to the bottom of the Swagger document, which Swagger-UI renders as an “Authorize” button:

Clicking that brings up a dialog box where you can put your bearer token:

The next thing we need to do is tell Swashbuckle which of our actions require Authorization. To do that you can use the SecurityRequirementsOperationFilter:

services.AddSwaggerGen(options =>
{
    options.AddSecurityDefinition("oauth2", new ApiKeyScheme
    {
        Description = "Standard Authorization header using the Bearer scheme. Example: \"bearer {token}\"",
        In = "header",
        Name = "Authorization",
        Type = "apiKey"
    });

    options.OperationFilter<SecurityRequirementsOperationFilter>();

You can either download the SecurityRequirementsOperationFilter from here, or, if you’re using ASP.NET Core you can install my Swashbuckle.AspNetCore.Filters package from NuGet, which includes it (and other filters).

The SecurityRequirementsOperationFilter adds a security property to each operation in the Swagger document, which renders in Swagger-UI as a padlock next to the operation:

Once you’ve done that, when you “Try it out” using the Swagger-UI, the authorization header with your bearer token should be sent to your API.

Show Swagger documentation on Azure Service Fabric

Another blog post in what seems to be becoming a series of posts on Swagger.

Swashbuckle allows you to include XML comments on your API’s Swagger page. To do this you need to set your Build to output an XML file, which Swashbuckle reads the XML comments from.
Don’t forget to add that XML file to the “Release” build configuration too, otherwise you won’t have XML comments when your application is actually deployed to an environment.

Here’s a handy code snippet for your .NET Core .csproj file:

  <!-- Begin Swagger documentation file-->
  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|AnyCPU'">
    <DocumentationFile>bin\Debug\net46\win10-x64\swagger.xml</DocumentationFile>
    <NoWarn>1701;1702;1705;1591</NoWarn>
  </PropertyGroup>

  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
    <DocumentationFile>bin\Debug\net46\win10-x64\swagger.xml</DocumentationFile>
    <NoWarn>1701;1702;1705;1591</NoWarn>
  </PropertyGroup>

  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|AnyCPU'">
    <DocumentationFile>bin\Release\net46\win10-x64\swagger.xml</DocumentationFile>
    <NoWarn>1701;1702;1705;1591</NoWarn>
  </PropertyGroup>

  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
    <DocumentationFile>bin\Release\net46\win10-x64\swagger.xml</DocumentationFile>
    <NoWarn>1701;1702;1705;1591</NoWarn>
  </PropertyGroup>
  <!-- End Swagger documentation file-->

Additionally, if your application is deployed on Azure Service Fabric, you will also need to copy the XML file to your PublishDir, with the following snippet in your .csproj:

  <Target Name="PrepublishScript" BeforeTargets="PrepareForPublish">
    <ItemGroup>
      <DocFile Include="bin\x64\$(Configuration)\$(TargetFramework)\win10-x64\*.xml" />
    </ItemGroup>
    <Copy SourceFiles="@(DocFile)" DestinationFolder="$(PublishDir)" SkipUnchangedFiles="false" />
  </Target>

And of course you need to tell Swashbuckle to use that XML file via the .IncludeXmlComments() method mentioned above.

Fix slow NCrunch build

NCrunch is a great unit test runner and has been part of my developer toolbox for many years.

Recently I was having an issue on a medium-sized project where the NCrunch’s Build step was taking almost 2 minutes to complete. That spinning “B” was taking forever!

NCrunch progress

The tests themselves would run in under one minute.

The problem was we use a custom Autofixture [AutoData] attribute on every test.

Our attribute’s constructor creates a TestServer for running the web tests against.
We then inject the test server into each test, i.e.

[Scenario]
[WebHostAutoData]
public void CreateJob_ShouldCreateSuccessfully(
    WebApiClientFixture apiFixture

As it turns out, NCrunch enumerates all test attributes during Analysis, so you should not have any expensive code in your attribute’s constructor.

In my case, I fixed this by introducing a base class for all of my tests, which creates the TestServer (during test run). I then dropped our custom [AutoData] attribute.

Add an upload button to your swagger page

The swagger adventures continue…

In .NET Core the current recommended way to upload a file in ASP.NET is with an IFormFile.

If you’re using an IFormFile in your ASP.NET Web Api like so:

[HttpPost("upload")]
public IActionResult UploadFile(IFormFile file)
{

N.B. Update December 2018. As of Swashbuckle 4.0, IFormFile is supported out of the box, so the filter below is no longer needed.

If you do this, the Swagger page as rendered by Swashbuckle won’t look correct:

However, if you install version 2.5.0 or later of my Swashbuckle.AspNetCore.Examples NuGet package, then you can add the [AddSwaggerFileUploadButton] attribute to your controller action, like so:

[HttpPost("upload")]
[AddSwaggerFileUploadButton]
public IActionResult UploadFile(IFormFile file)
{

and then enable my filter in your Startup.cs:

services.AddSwaggerGen(c =>
{
    c.OperationFilter<AddFileParamTypesOperationFilter>();

Then you’ll get a nice upload button:

N.B. Update December 2018. As of Swashbuckle 4.0, IFormFile is supported out of the box, so the filter above is no longer needed.

Speed up development in a NuGet package-centric solution

As microservices architectures become more popular, so increases usage of NuGet as a way of sharing code amongst separate services. The last few projects I’ve worked on have typically contained a number of “Core” NuGet packages with shared code and interfaces that are consumed by one or more services in the solution. Our build pipeline will publish these NuGet packages to an in-house NuGet server.

This can cause a lot of developer friction when you’re working in one of the services and you decide you need to change one of the Core projects. Typically you’ll

  1. make the change to Core
  2. commit it
  3. (Perhaps raise a PR and wait for approval)
  4. wait for the build to publish new versions of the package
  5. and finally updated the package in your service.

Probably at least 20 min wait for all that to happen. And then you realise that your change didn’t quite work, so you have to go through the cycle again with another change.

Well, here’s a shortcut to that cycle and it’s pretty darn simple. Note that this only works if you’re using the newer .NET Core style .csproj containing <PackageReference>. Simply

  1. make the change in Core and compile it locally
  2. Copy the resulting dll(s) to your local NuGet packages feed folder. NOT the bin folder of your service, or the packages folder of your service. On my machine the local NuGet packages feed folder is C:\Users\matt\.nuget\packages\.

As soon as the dll is copied, Visual Studio detects the change – you don’t need to tell it to update the NuGet packages or even compile!

Another surprise bonus of this – while debugging you can step into the Core code! You don’t need to mess around copying PDBs around or anything. I don’t know how it works, but Visual Studio seems to somehow know that the DLL you’ve copied into the NuGet folder came from your local machine and thus it knows where the source code for it is.

Obviously once you’re happy that your changes to Core are working correctly, then you can can commit them and consume an updated version of the package in your service as usual.

Anyway, hopefully this will speed up your local development as much as it has mine.

Run a Service Fabric solution locally without deploying to Service Fabric

This is a similar piece to another post of mine from a few years ago Run a Windows Azure cloud service locally without the Azure compute emulator

So you’re working on a Service Fabric application which has an ASP.NET Web api host project. I find the debugging experience painful, for two reasons:

  1. Time to start debugging the project is a minimum of 45 seconds, every time, because the app gets deployed to a local service fabric cluster which takes forever.
  2. You need to remember to run Visual Studio as administrator in order for the above local deployment to succeed.

If either of these things bug you, then here’s a possible solution. Once we’re done you’ll be able to set the Web project in your solution as the StartUp project instead of the Service Fabric application, for much faster debugging, and you’ll no longer need to run VS as admin.

First, change the Program.cs in the Web project:

private static void Main()
{
	if (UseServiceFabric())
	{
		StartServiceFabric();
	}
	else
	{
		StartWebHost();
	}
}

private static bool UseServiceFabric()
{
	var webHostBuilder = new WebHostBuilder();
	var environment = webHostBuilder.GetSetting("environment");

	return environment != "Development";
}

private static void StartWebHost()
{
	var builder = new WebHostBuilder()
		.UseKestrel()
		.UseContentRoot(Directory.GetCurrentDirectory())
		.UseStartup<Startup>();

	var host = builder.Build();
	host.Run();
}

private static void StartServiceFabric()
{
	try
	{
		// The ServiceManifest.XML file defines one or more service type names.
		// Registering a service maps a service type name to a .NET type.
		// When Service Fabric creates an instance of this service type,
		// an instance of the class is created in this host process.

		ServiceRuntime.RegisterServiceAsync("Web1Type",
			context => new WebHost(context)).GetAwaiter().GetResult();

		ServiceEventSource.Current.ServiceTypeRegistered(Process.GetCurrentProcess().Id, typeof(Web1).Name);

		// Prevents this host process from terminating so services keeps running. 
		Thread.Sleep(Timeout.Infinite);
	}
	catch (Exception e)
	{
		ServiceEventSource.Current.ServiceHostInitializationFailed(e.ToString());
		throw;
	}
}

If the ASPNETCORE_ENVIRONMENT setting is Development, then it won’t use Service Fabric at all and will just use a plain ol’ ASP.NET Core WebHostBuilder to start the web host.

You’ll also need to change your Debug target to be the Web project, instead of IISExpress, via the VS Standard toolbar.

And just like that the app startup time has shrunk from around 40 seconds to around 5 seconds. Or from an unbearable 100 seconds for the application the team I just joined is working on.

Add an authorization header to your swagger-ui with Swashbuckle

Edit July 2018: I’ve blogged a better way to do this. Add an authorization header to your swagger-ui with Swashbuckle (revisited).

Out of the box there’s no way to add an Authorization header to your API requests from swagger-ui. Fortunately (if you’re using ASP.NET), Swashbuckle 5.0 is extendable, so it’s very easy to add a new IOperationFilter to do it for us:

public class AddAuthorizationHeaderParameterOperationFilter : IOperationFilter
{
    public void Apply(Operation operation, SchemaRegistry schemaRegistry, ApiDescription apiDescription)
    {
        if (operation.parameters != null)
        {
            operation.parameters.Add(new Parameter
            {
                name = "Authorization",
                @in = "header",
                description = "access token",
                required = false,
                type = "string"
            });
        }
    }
}

Now all you need to do is register it in your EnableSwagger call:

configuration
    .EnableSwagger(c =>
    {
        c.SingleApiVersion("v1", "Commerce Services - Discounts");

        foreach (var commentFile in xmlCommentFiles)
        {
            c.IncludeXmlComments(commentFile);
        }

        c.OperationFilter<ExamplesOperationFilter>();
        c.OperationFilter<AddAuthorizationHeaderParameterOperationFilter>();
    })
    .EnableSwaggerUi(config => config.DocExpansion(DocExpansion.List));

Once that’s done it’ll give you an input field where you can paste your Authorization header. Don’t forget to add the word “bearer” if you’re using a JWT token:

Edit July 2018: I’ve blogged a better way to do this. Add an authorization header to your swagger-ui with Swashbuckle (revisited).