Teamcity running build steps even when tests fail

Question

I am having problems with Teamcity, where it is proceeding to run build steps even if the previous ones were unsuccessful.

The final step of my Build configuration deploys my site, which I do not want it to do if any of my tests fail.

Each build step is set to only execute if all previous steps were successful.

In the Build Failure Conditions tab, I have checked the following options under Fail build if:

-build process exit code is not zero
-at least one test failed
-an out-of-memory or crash is detected (Java only)

This doesn't work - even when tests fail TeamCity deploys my site, why?

I even tried to add an additional build failure condition that will look for specific text in the build log (namely "Test Run Failed.")

When viewing a completed test in the overview page, you can see the error message against the latest build:

"Test Run Failed." text appeared in build log

But it still deploys it anyway.

Does anyone know how to fix this? It appears that the issue has been running for a long time, here.

Apparently there is a workaround:

So far we do not consider this feature as very important as there is an obvious workaround: the script can check the necessary condition and do not produce the artifacts as configured in TeamCity.

e.g. a script can move the artifacts from a temporary directory to the directory specified in the TeamCity as publish artifacts from just before the finish and in case the build operations were successful.

But that is not clear to me on exactly how to do that, and doesn't sound like the best solution either. Any help appreciated.

Edit: I was also able to workaround the problem with a snapshot dependency, where I would have a separate 'deploy' build that was dependent on the test build, and now it doesn't run if tests fail.

This was useful for setting the dependency up.

Groundwork

Even though you fixed your own problem with a workaround, I think Adams answer is better for the scenario you describe. Consider marking his answer as accepted. – Skep

yeah I wasn't sure only because my workaround was simpler, but it was an excellent answer and I have reassigned. – Groundwork

Answer 1

This is a known problem as of TeamCity 7.1 (cf. http://youtrack.jetbrains.com/issue/TW-17002) which has been fixed in TeamCity 8.x+ (see this answer).

TeamCity distinguishes between a failed build and a failed build step. While a failing unit test will fail the build as a whole, unfortunately TeamCity still considers the test step itself successful because it did not return a non-zero error code. As a result, subsequent steps will continue running.

A variety of workarounds have been proposed, but I've found they either require non-trivial setup or compromise on the testing experience in TeamCity.

However, after reviewing a suggestion from @arex1337, we found an easy way to get TeamCity to do what we want. Just add an extra Powershell build step after your existing test step that contains the following inline script (replacing YOUR_TEAMCITY_HOSTNAME with your actual TeamCity host/domain):

$request = [System.Net.WebRequest]::Create("http://YOUR_TEAMCITY_HOSTNAME/guestAuth/app/rest/builds/%teamcity.build.id%")
$xml = [xml](new-object System.IO.StreamReader $request.GetResponse().GetResponseStream()).ReadToEnd()
Microsoft.PowerShell.Utility\Select-Xml $xml -XPath "/build" | % { $status = $_.Node.status }

if ($status -eq "FAILURE") {
    throw "Failing this step because the build itself is considered failed. This is our way to workaround the fact that TeamCity incorrectly considers a test step to be successful even if there are test failures. See http://youtrack.jetbrains.com/issue/TW-17002"
}

This inline PowerShell script is just using the TeamCity REST API to ask whether or not the build itself, as a whole, is considered failed (the variable %teamcity.build.id%" will be replaced by TeamCity with the actual build id when the step is executed). If the build as a whole is considered failed (say, due to a test failure), then this PowerShell script throws an error, causing the process to return a non-zero error code which results in the individual build step itself to be considered unsuccessful. At that point, subsequent steps can be prevented from running.

Note that this script uses guestAuth, which requires the TeamCity guest account to be enabled. Alternately, you can use httpAuth instead, but you'll need to update the script to include a TeamCity username and password (e.g. http://USERNAME:PASSWORD@YOUR_TEAMCITY_HOSTNAME/httpAuth/app/rest/builds/%teamcity.build.id%).

So, with this additional step in place, all subsequent steps set to execute "Only if all previous steps were successful" will be skipped if there are any previous unit test failures. We're using this to prevent automated deployment if any of our NUnit tests are not successful until JetBrains fixes the problem.

Thanks to @arex1337 for the idea.

Jacelynjacenta

There is an easier way to get build status directly without parsing XML: /guestAuth/app/rest/builds/%teamcity.build.id%/status – Cinque

Throwing an error from PowerShell does not seem to return an error code. I used "exit 1" instead. – Cinque

As @Cinque mentioned, you need to output an non-zero exit code for TeamCity to detect build failed. Use 'exit 1' rather than throwing the error. – Profluent

Answer 2

Just to prevent confusion, this issue is fixed in Team City v8.x, We don't need those workarounds now.

You can specify the step execution policy via the Execute step option:

Only if build status is successful - before starting the step, the build agent requests the build status from the server, and skips the step if the status is failed.

https://confluence.jetbrains.com/display/TCD8/Configuring+Build+Steps

Of course you need to fail the build if at least one unit test failed:

https://confluence.jetbrains.com/display/TCD8/Build+Failure+Conditions

On the Build Failure Conditions page, the Fail build if area, specify when TeamCity will fail builds:

at least one test failed: Check this option to mark the build as failed if the build fails at least one test.

Denudate

This is the correct answer. Checking the boxes for "at least one test failed" and "an error message is logged by build runner" (just in case) fixed this issue for us. Saved time, effort, and maintenance confusion to avoid the powershell route! – Luxor

Answer 3

This is (as you have found) a known issue with TeamCity, there are a set of linked issues in their Issue Tracker. This issue is hopefully scheduled to be resolved in the next release of TeamCity (version 8.x)

In the mean time, the way we identified to resolve the issue (for version 6.5.5) was to download the test results file as part of the later steps. This was then parsed to check for any test failures, returning an error code and hence breaking the build properly (performing any cleanup we needed as part of that failure) which would probably work for you.

Gameto

Answer 4

TeamCity build failure does not mean that it will stop the build and it will publish the artifacts if your build is providing the the build output files as required by TeamCity. It will only update the build status properly.

But, you can very well stop the build process by modification to your build script to stop the build on test case failure. If you are using MSBuild, then ContinueOnError="false" will do that.

Mazuma

I set this value in the Deployment build step, @RinoTom, and it still deployed – Groundwork

Sorry for the confusion if I made any. ContinueOnError is something which is set by default to false. So you may not need to explicitly specify that in any of the build steps. But, in some of your build steps it might be set as true. So you need to check for that one with true and change that back. Most probably, the place where the build is failing right at the moment. Deployment build step can continue as such without any change. – Mazuma

Answer 5

In the end, I was able to solve the problem with a snapshot dependency, where I would have a separate 'deploy' build that was dependent on the test build, and now it doesn't run if tests fail.

This was useful for setting the dependency up.

Groundwork