Friday, September 16, 2011

Automatic web deployment from TFS build

Submit this story to DotNetKicks

We have recently started using Team Foundation Server 2010, and having a CI build running for every check-in was one of the things we were really enjoying. Then I saw Scott Hanselman's talk on web deployment - if you're using Xcopy, you're doing it wrong!

Of course, once I saw that , I  wanted an automated deployment to the development environment every night or so. To get that going, I had to do some research.

Just so you know, the project I'm talking about consists of  three web applications, one DAL and two presentation layers. Of course all three need different configurations for the different environments (dev, test and production).

Web.config transformations
And as it turns out, there is a neat function in visual studio 2010 called "web.config transformations" for exactly this purpose. You can read about it here at Scott's blog.

The most obvious example for using web.config transformations is probably database connectionstrings, but you  can replace most things using this simple syntax:

<setting name="Your_Setting_name" serializeAs="String" xdt:Transform="Replace" xdt:Locator="Match(name)">        <value>Your value here!</value>      </setting>
The web.config translates directly to the build configuration. So if you're building in "Debug" you're going to use the web.Debug.config file for transformations.  In our setup, this is the CI build. For the nightly automated deploy, we're going to need a separate target, so we created one called  Dev-Snapshot, (perhaps it would have been even more apt with Nighly-Dev - but let's ute Dev-Snapshot for the remainder of this post).

Then you need a build defintion on the TFS that uses the new build configuration. For instance, call it Dev-Snapshot like the build configuration, under Process and "Items to build", select the correct project and your newly defined build configuration. Now whenever you build using the Dev-Snapshot in Visual Studio, the original web.config will be used. Transformations are only applied when you publish - it will automagically transform the web.config using the web.Dev-Snapshot.config into the final web.config inside the deployment package.

Publish settings
Now we have a working build of the development environment, tailored for deployment on our development server.  So how do we do the automatic deployment? Maybe there's a switch in TFS Build for it? No such luck.

The first thing you need to look at is your web project's publish settings. They too relate to the build configuration, so make sure you have selected the right one - again, for instance, Dev-Snapshot.

Most interesting bits: Under items to deploy, select "only files needed to run the application" - no .cs files, .csproj etc are not included in the package, just the "necessary files" (more on that later).

You can choose to include database settings (out of scope for this post) and run any setup or change scripts directly. However, this probably fits best for any single-server solution, where you have one database server and one web server - in a web farm, you pretty much need to update all nodes whenever you change the db, though I suspect you might be able to run the db package only for the first node or something similar.

Under Web Deployment Package settings you may choose to package as a zip file. I think this is useful. You need to specify where to create the package, and the name of the web site. This name must match the actual name of the web site you're planning to update on the server.

After these settings are set, you may choose to test that they are actually working. Do this by rightclicking the project and select "Publish". What the TFS will do later on is actually build a deployment package - which you can do by right-clicking the project and select "Build deployment package". Then go look in the destination you provided in the settings - you should have some files there.

Read more about this in Scott Gu's blog post.

Deploying from TFS
Now that we have a working web.config transformation and web publish settings, the stage is set for the TFS Build server (and MSBuild in the background). In Team Explorer, go to Builds and right-click the one you're working with. Then go to the Process section, and under "Items to build" make sure you have the correct project with the correct configuration selected (in this example - Dev-Snapshot configuration).

Open the "3. Advanced" section and go down to MSBuild arguments. Enter
/p:DeployOnBuild=True /p:IsAutoBuild=True
The first one tells MSBuild to build a deployment package after the build has completed, the second tells TFS that this is an automated build, as opposed to a manually triggered one.

If you try queueing this build now, it should be working fine. You can open the build drop location to see the deployment package. But there's one step missing  - the deployment itself! /p:DeployOnBuild=True  as mentioned, only tells MSbuild to create the deployment package. It does not run the deployment script. That's not really automatic deployment...

So what I found out is that you can use a post-build event to call the script. The post-build event is found by right-clicking the project and selecting Properties, then Build events. Surprise, surprise, the Build Events are common to all configurations! which means, you can't simply run the script here, you have to run it according to the configuration. Or else, you could end up with any build ending up on your production server. Bad news.

Enter a condition that will be true for your build - $(ConfigurationName) is a build variable that matches the Build configuration - and then create the command needed to execute the script:
if "$(ConfigurationName)" == "Dev-snapshot" "$(TargetDir)_PublishedWebsites\Projectname_Package\Projectname.deploy.cmd" /Y /u:publishuser /p:secret! /M:yourwebserver
Of course, your configuration will vary from this, but the basic idea should be clear, and while a little messy this can be extended to as many builds as you like, just keep on adding conditions.

If you have some common settings for some builds - let's say your development and test environment has some common tasks - you can distinguish them with the IsAutoBuild parameter - $IsAutoBuild == 'True'.

Now, if you queue up this build, you should actually have a ready published web site. Or a failed build. Make sure you have installed the web deployment package on the remote server, opened the correct port in the firewall and that the service is indeed running! Details here.

Missing files?
Well the deployment is now working - new files are appearing on your web server when you build. But in my case there were at least some files that weren't appearing. Also there were some empty directories, that would be populated at runtime, suddenly missing. Strange.

So you may remember that setting for "Only files needed to run the application". It determines by itself what is needed. Turns out, .class files, .pdf's, .zip and .csv files are "not needed". In addition, empty directories are not included in the deployment. It actually half makes sense, though there should be some way for the developer to specify the details.

However, there is not in the GUI - but there is in the build process, if you use the build targets inside the project file! Sam Stephens has a blog post on this. You can either add this directly to your project file, or you can create a separate .target file for tidyness. A word of caution for the latter, while more tidy, it requires you to close and reopen the solution (!) for changes to be effective. Inside the project file you "only" need to unload and reload the project.

Here's the target configuration for the missing files in my project (the \**\ just means all directories, recursively):


   <PropertyGroup>
    <CopyAllFilesToSingleFolderForPackageDependsOn>
      CustomCollectClassFiles;
      $(CopyAllFilesToSingleFolderForPackageDependsOn);
    </CopyAllFilesToSingleFolderForPackageDependsOn>
  </PropertyGroup>


<Target Name="CustomCollectClassFiles">
    <ItemGroup>
      <_CustomClassFilesForRootFolder Include=".\**\*.class;.\**\*.zip;.\**\*.csv;.\**\*.pdf">
        <DestinationRelativePath>%(RecursiveDir)%(Filename)%(Extension)</DestinationRelativePath>
      </_CustomClassFilesForRootFolder>
      <FilesForPackagingFromProject Include="%(_CustomClassFilesForRootFolder.Identity)">
        <DestinationRelativePath>.\%(RecursiveDir)%(Filename)%(Extension)</DestinationRelativePath>
      </FilesForPackagingFromProject>
    </ItemGroup>
  </Target>


That covers the missing files, but the empty folders need another target. Or, you can simply add an empty file in the folder for it to be picked up, but that is really a hack.

I found a better way of doing it through the AfterAddIisSettingAndFileContentsToSourceManifest - a target that is defined by the web deployment publishing pipeline. It's kind of hard to find a complete list of the targets involved, but feel free add one in the comments.


<PropertyGroup>
    <AfterAddIisSettingAndFileContentsToSourceManifest>
      MakeEmptyFolders
    </AfterAddIisSettingAndFileContentsToSourceManifest>
  </PropertyGroup>


<Target Name="MakeEmptyFolders">
    <Message Text="Adding empty folder to hold snapshots...$(_MSDeployDirPath_FullPath)\SnapshotImages" />
    <MakeDir Directories="$(_MSDeployDirPath_FullPath)\SnapshotImages"/>
  </Target>



There. Missing files and folders no more.

Minifying javascript Build-Time
Since this is a web application, I will add a final note about the minifying task. I used the Ajax Minify from Microsoft, but I'm sure the same applies to the other tools as well, just with a different command line. Before I started on the automatic deployment project, we were using the AfterBuild target for minify operations, which worked out well, when a batch script (using robocopy!) was doing the deployment.

But when the deployment package is built, the AfterBuild target is not yet entered. So we switched to BeforeBuild to get around that - and it works perfectly. Unless you are emitting javascript files from your build, the same approach should be fine for you.

Here's what happens: First, delete the old files. Then, concatenate any separate files into one big .js file. Then, minify that file. In addition, you obviously have to reference that minified .js from your master page or what ever other page you need it in. I'd also suggest renaming the .js file for each version to avoid caching problems, but that's another story.



<Target Name="BeforeBuild">
     <!-- we need to minify the .js files before build, because they must be included in the deployment package -->
    <Message Text="Going to delete old JS files..." />
    <Delete Files=".\Controls\Javascript\Concatenated.js;.\Controls\Javascript\Concatenated.min.js" />
    <Message Text="Concatenating JavaScript files..." />
    <ItemGroup>
      <InFiles Include=".\Controls\Javascript\*.js" />     
      <InFiles Include=".\Controls\JsFolder\*.js" />
    </ItemGroup>
    <ReadLinesFromFile File="%(InFiles.Identity)">
      <Output TaskParameter="Lines" ItemName="lines" />
    </ReadLinesFromFile>
    <WriteLinesToFile File=".\Controls\Javascript\Concatenated.js" Lines="@(Lines)" Overwrite="true" />
    <Message Text="Minifying JavaScript files..." />
    <ItemGroup>
      <JS Include=".\Controls\Javascript\Concatenated.js" />
    </ItemGroup>
    
    <!-- Minify javascript files in .\Controls\Javascript -->
    <AjaxMin SourceFiles="@(JS)" SourceExtensionPattern="\.js$" TargetExtension=".min.js" CollapseToLiteral="True" LocalRenaming="CrunchAll" OutputMode="SingleLine" RemoveUnneededCode="True" StripDebugStatements="True" EvalsAreSafe="True" InlineSafeSettings="True" CombineDuplicateLiterals="True" CatchAsLocal="True" />
  </Target>