Moving From CruiseControl.NET to TeamCity
I admit to loving me some good ol’ CruiseControl.NET for continuous integration but there’s just something to be desired from the UI and ease-of-use/ease-of-administration standpoint. CI Factory brings it a step closer to what I’m looking for, but after doing some evaluation, we ended up trying out TeamCity.
It’s easy to set up, easy to administer, flexible, customizable… really, the experience thus far has been pretty great. At work we’re testing it out with the free Professional edition, but if it works out on this project, we’ll be getting the full Enterprise edition. I know that if I set up a continuous integration server for CR_Documentor it’ll definitely be TeamCity.
That said, there are some differences in how you have to do things in TeamCity than if you’re working in CruiseControl.NET. Here’s what I’ve found so far. (Note - I’m using MSBuild in my code snippets. If you’re using NAnt or something else, you get to translate.)
$(CCNetLabel) becomes $(BUILD_NUMBER). In CruiseControl.NET you access the current build number (e.g., “22.214.171.1244”) with the variable $(CCNetLabel). In TeamCity, it’s $(BUILD_NUMBER). Something we’ve done to make this easier is creating a variable $(BuildLabel) that we copy the appropriate value into at the beginning of the script. Doing that allows you to separate your build script from the build server proprietary variables.
Builds are generally done with agents. In CruiseControl.NET, if you want to run a private build you drop to the command line and build it yourself. In TeamCity, you can ship your changes off to the server to run a personal build. There is a variable provided $(BUILD_IS_PERSONAL) that will equal “true” if the build is running for an individual rather than as part of the “official” process. You’ll want to update your build script to accommodate both personal builds via agent and via command line.
This sample shows some MSBuild logic that sets a $(BuildLabel) property based on whether a build is personal or not. It also sets a $(CCNetLabel) property for backwards compatibility with scripts that might be using that.
<Project InitialTargets="__EnvironmentSetup" xmlns="http://schemas.microsoft.com/developer/msbuild/2003"> <Target Name="__EnvironmentSetup"> <CreateProperty Condition="'$(BUILD_NUMBER)' == ''" Value="true"> <Output TaskParameter="Value" PropertyName="BUILD_IS_PERSONAL"/> </CreateProperty> <CreateProperty Condition="'$(BUILD_IS_PERSONAL)' == 'true'" Value="0.0.0.0"> <Output TaskParameter="Value" PropertyName="BuildLabel"/> </CreateProperty> <CreateProperty Condition="'$(BUILD_IS_PERSONAL)' != 'true'" Value="$(BUILD_NUMBER)"> <Output TaskParameter="Value" PropertyName="BuildLabel"/> </CreateProperty> <CreateProperty Value="$(BuildLabel)"> <Output TaskParameter="Value" PropertyName="CCNetLabel"/> </CreateProperty> </Target> </Project>
Use the TeamCity test runner for better integration. While you can still use a straight call to, say, NUnit-Console.exe, you get very rich interaction and reporting in the TeamCity UI if you use the TeamCity version of the runner for your tests. You’ll want your script to “sniff” and see if it’s running on a developer box or in a TeamCity environment and use the appropriate runner.
This sample shows some MSBuild logic that builds an appropriate test runner command line for running NUnit 2.4.8 tests based on the environment. It assumes you have a list of test assemblies in a @(TestAsemblies) collection.
<PropertyGroup Condition="'$(BuildLabel)'!='0.0.0.0'"> <TestCommandLineExe>$(teamcity_dotnet_nunitlauncher)</TestCommandLineExe> <TestCommandLineArgs>v2.0 x86 NUnit-2.4.8 @(TestAssemblies->'%(FullPath)', ' ')</TestCommandLineArgs> </PropertyGroup> <PropertyGroup Condition="'$(BuildLabel)'=='0.0.0.0'"> <TestCommandLineExe>path\to\NUnit-Console.exe</TestCommandLineExe> <TestCommandLineArgs>@(TestAssemblies->'%(FullPath)', ' ') /xml:TestResults.xml</TestCommandLineArgs> </PropertyGroup>
There are two ways to integrate FxCop into the build. The first is to use the built-in FxCop build runner. This will run just FxCop though, and as a separate step. If you’re OK with not running FxCop as part of your CI build, that’s the easiest way to go. If you want to run the FxCop command line, though, and get the report into the UI, you need to use a “service message” - a message to Console.Out in a special format - to tell TeamCity where to get the report.
This sample shows what a service message to get FxCop command line output into TeamCity might look like.
<Message Text="##teamcity[importData id='FxCop' file='fxcop-report.xml']" Condition="'$(BuildLabel)'!='0.0.0.0'" />
NCover support is manual. The stock TeamCity code coverage support is for EMMA. If you want NCover reporting to show up in the UI, there’s a little work to do. This blog entry shows a great in-depth how-to, but the simple version boils down to:
- Set up your HTML coverage report (e.g., the output from NCoverExplorer) to be published as a build artifact (along with your binary/compiled output).
- Update your TeamCity main-config.xml file to include the report. You’ll most likely want to comment out the other code coverage related report tabs since you probably won’t be using them.
For example, if you publish the report as “CoverageReport.html” then your main-config.xml file might have a line in it like:
<report-tab title="Code Coverage" basePath="" startPage="CoverageReport.html" />
You will probably also want to see a trend report of code coverage over time. To do that, you need to have your build script publish the data to TeamCity (again using “service messages”) and you need to add the custom graph to the build dashboard.
The MSBuild logic to publish the code coverage overall percentage to TeamCity looks like this. Note that in this example, it’s assumed that the output from NCoverExplorer has been written to “CoverageReport.xml” - adjust your paths as needed.
<XmlRead XPath="//coverageReport/project/@coverage" XmlFileName="CoverageReport.xml" Condition="Exists('CoverageReport.xml')"> <Output TaskParameter="Value" PropertyName="CoveragePercent"/> </XmlRead> <Message Text="##teamcity[buildStatisticValue key='coveragePercent' value='$(CoveragePercent)']" Condition="'$(BuildLabel)'!='0.0.0.0'" />
In your main-config.xml file, you then need to add the custom graph so it shows up in the build statistics page:
<graph title="Code Coverage"> <valueType key="coveragePercent" title="% Coverage" /> </graph>
The important bit there is that the value of “key” in the “service message” matches the value of “key” in the graph description.
Users set up their own notifications. In CruiseControl.NET, the build configuration specifies who gets alerted and to which build events. I hate that. I get all sorts of notifications I want because I happen to be on a mailing alias that was added administratively to the notification list. I can’t get out of it. In TeamCity, I get to configure my personal preferences for my notifications. This can be weird for some users who wonder why they’re not getting notified anymore - they need to do the work themselves to subscribe.
Users need to tie their TeamCity accounts to version control system accounts. There’s a lot of functionality in TeamCity that allows you to deal with “changes you made.” The way TeamCity figures this out is by each user providing a mapping of their TeamCity account to their username on the various version control systems they use. The “my changes” features won’t work for people that don’t set this up. (There is no analogous behavior in CruiseControl.NET.)
Builds aren’t done from working copies. In CruiseControl.NET when the build server checks out the source, the build is actually run from a live Subversion working copy. In TeamCity, the code still gets checked out from Subversion, but none of the Subversion admin directories (.svn) are there - it’s not a working copy, it’s an exported checkout. This is important if your build script tries to do anything with the version control system like inspect the location from which the current working copy was checked out. You can mitigate this in TeamCity by providing a custom environment variable (e.g., $(BUILD_VCS_ROOT)) that contains this information. It won’t be dynamic, but it’s the best you can do.
You can’t serialize builds. In CruiseControl.NET there’s a notion of a “build queue” such that if one build is running, it can “lock” a semaphore and other builds that require that semaphore won’t run until it’s free. That sort of thing is helpful if your build script dynamically registers third-party assemblies in the GAC for the duration of the build, for example. There is no such serialization mechanism in TeamCity.
You can get notified of build events via Google Talk. Technically the “Jabber” protocol, but Google Talk nonetheless. You’ll need to get your build server a Google account, but once you do, people can add that service account to their friends list and configure Google Talk notifications. They tell you how to do this in the TeamCity docs but they don’t tell you what to do if port 5222 (the Jabber port) is blocked off at your firewall - switch to port 80. Here’s what works for me:
- Server: talk.google.com
- Port: 80
- Server user: firstname.lastname@example.org
All in all, TeamCity is the advance in the build server that I was hoping would come along for CC.NET. It’s well worth the time to check out.