Gingter Ale

Fluent stories of a software engineer

Custom deployment scripts - with mstest - for Windows Azure Website git deployment

2013-06-11

I just started another project. It is hosted on Windows Azure and I’m using Git deployment for this website.

This was very fine and I am extremely impressed how easy it was to get started with it. Then I ran into a little problem.

Sidenote: My project relies on NuGet packages, and I, personally, have the strong opinion that compiled stuff does not belong into my source code versioning system. This is why I did not check in the NuGet.exe into Git, but just the NuGet.config and NuGet.Targets files configured to download NuGet.exe when it’s missing. Of course I make my build dependent on a NuGet package server, but since I could host my own gallery on a custom domain, and configure that domain in my NuGet.config, I could take control over this dependency at any time.

I wanted my project to incorporate information about the Git commit hash it is built from, the Git branch it was built from and other little details. For that the MsBuild Community Tasks project offers some nice helpers. So I added the NuGet package of this project to my solution.

The Problems

Now there is this chicken-egg problem: When MsBuild encounters a UsingTask declaration, it automatically loads the assembly that contains the task. If that assembly is not there, using the task will fail. Now, the NuGet download of the packages - including the task library - happens as part of the build. That is, after the project files are loaded. So the fresh downloaded file was not found when importing the projects and… the build fails.

To avoid this problem, I cheated a little bit on MsBuild: I added another project to my solution that also has the MsBuild Community Task project listed in it’s packaged.config. Then I manually set my web application project to be build after this ‘BuildSupport’ project. Now the BuildSupport project build downloads the community task library, which is then available when the project import is defined in the web application’s project file. It’s just a small cheat, though.

Then the next problem: The BuildSupport project is not actually ‘required’ to build the website project, and so the Git deployment build process simply does not build it. The task library is not downloaded prior to executing the actual build process of the application, and so it fails. I could not get the project to build the ‘BuildSupport’ project before the actual web application on Azure.

The Solution

After a little bit of research I found this can be achieved by using a custom deployment script. I was a bit afraid that I had to figure out how the actual deployment works to add a step just in front of the actual compile, but there is some infrastructure in place to help us out with that.

For a .NET developer this will feel strange, but you’ll need node.js in the first place. The Windows Azure Command Line Tools are a node.js package, and we’ll need that to get started with the actual live deployment script. So, after installing node.js, we’re going to install the package:

[shell]npm install azure-cli -g[/shell]

This will globally install the Azure CLI for use on our console. Now we navigate to our solution directory and let the Azure CLI generate the deployment script that will automatically run to deploy our application to Azure if we don’t do anything custom:

[shell]azure site deploymentscript –aspWAP ApplicationFolderApplication.csproj -s Solution.sln[/shell]

This will generate two files for you. First there is a .deployment file. This is a file structured like a oldfashioned .ini configuration file, telling Azure that there is a custom deployment file and how its name is. It’s content simply is: [text] [config] command = deploy.cmd[/text]

It also reveals the second generated file, the actual deployment script called deploy.cmd. This is the interesting part for us so far. I’m not posting the full script but rather go through the sections.

First there is a check that node.js is available. It is assumed that this is available on Azure, but to test the deployment script locally you’ll also need node.js. We just installed it, so we’re all set, but the next one checking out the solution could be missing node.

Then the script defines some environment variables for folders. Like where the build artifacts will be placed and where the actual files to deploy will be placed. This defaults to /artifacts/wwwroot and can be overridden by setting the corresponding environment variables before the deployment.

In a thirds step, the script checks if kudu is installed. Kudu is the actual deployment engine running on Azure, and is also capable of running on your machine. After that additional paths are configured.

In the fourth step the actual compiling and deployment work is done, and the fifth is just some error handling.

So, let’s have a look at the actual important stuff in the file:

[shell]:: 1. Build to the temporary path %MSBUILD_PATH% “%DEPLOYMENT_SOURCE%MyApplication.WebMyApplication.Web.csproj” /nologo /verbosity:m /t:Build /t:pipelinePreDeployCopyAllFilesToOneFolder /p:_PackageTempDir="%DEPLOYMENT_TEMP%";AutoParameterizationWebConfigConnectionStrings=false;Configuration=Release /p:SolutionDir="%DEPLOYMENT_SOURCE%.\" %SCM_BUILD_ARGS% IF !ERRORLEVEL! NEQ 0 goto error

:: 2. KuduSync call %KUDU_SYNC_CMD% -v 50 -f “%DEPLOYMENT_TEMP%” -t “%DEPLOYMENT_TARGET%” -n “%NEXT_MANIFEST_PATH%” -p “%PREVIOUS_MANIFEST_PATH%” -i “.git;.hg;.deployment;deploy.cmd” IF !ERRORLEVEL! NEQ 0 goto error[/shell]

Now, thats actualy very easy: MsBuild is called for the web application project, and then Kudu is launched to do the actual deployment.

What we want to achieve now is to build the full solution upfront to have all required NuGet packages downloaded before the actual project is being built. And while we are actually getting our hands dirty in a custom deployment script, why don’t add running the unit tests of the project as part of the deployment? So, if a test fails, deployment will fail too. I think that’s a good idea.

So what I did was adding these two steps just in front of the two default steps:

[shell]:: 1. Build solution echo Build solution %MSBUILD_PATH% “%DEPLOYMENT_SOURCE%MySolution.sln” /nologo /verbosity:m /t:Build /p:_PackageTempDir="%DEPLOYMENT_TEMP%";AutoParameterizationWebConfigConnectionStrings=false;Configuration=Release /p:SolutionDir="%DEPLOYMENT_SOURCE%.\" %SCM_BUILD_ARGS% IF !ERRORLEVEL! NEQ 0 goto error

:: 2. Running tests echo Running tests vstest.console.exe “%DEPLOYMENT_SOURCE%MyApplication.Web.TestsbinReleaseMyApplication.Web.Tests.dll” IF !ERRORLEVEL! NEQ 0 goto error[/shell]

That’s it. I just copied the build line and pointed it to my solution, and I added a call to the MsTest tooling to run my tests.

So, with very little tweaking I could remove all dependencies to actual binaries I would have to check in otherwise and I have the Azure git deployment run my unit tests on every deployment. That’s what I call easy and convenient.