Automation, bash, CI/CD, Docker

Taming Unit Test logging in Docker pipeline for .NET

Avatar of author Kees C. Bakker To profile Kees C. Bakker Written by Kees C. Bakker, written on , 4 minute read.

Automated unit tests are the backbone of the testing pyramid. Unit tests enable us to solve bugs using a test driven development strategy. They express both intent and usage and will become more valuable by time.

Last week I changed the way we integrated unit tests in our CI/CD pipeline. Here's our situation:

  • All projects are written in .NET Core / C#.
  • Every project has a matching unit test project.
  • Unit tests are written in Xunit.
  • Asserts are expressed as Fluent Assertions.
  • The CI script runs in a microsoft/aspnetcore-build:2.0.0 Docker container.
  • Unit tests are part of the script: restore, build, test, publish, pack, persist.
Unit tests are the base of the testing pyramid!

Logging practices

The main problem of a CI/CD pipeline - in my opinion - is logging. When it logs too little, you don't know what's going on, when it logs too much you can't see the trees to the forest. That's why I adhere to the following principles:

  1. Only. Show. Relevant. Information.
  2. Divide the pipeline into clear steps with output like "Step 4/9: Testing".
  3. Restore and build once. This speeds up the process and prevents extra logging.
  4. Always show errors.
  5. Terminate on failure with a clear reason.

The problem: too much output

Let's view the output of a typical unit test project (xunit) that ran by dotnet test --no-build --no-restore:

Now imagine if you have 9 of these projects. Imagine you don't have color-coded logs.

Get to greps with output

What can we to reduce the output of the unit tests? A couple of things:

  • Don't show the error stream, the standard stream has enough information.
  • Show lines that start with Failed, because they contain the name of the failed test.
  • Show lines that start with Total, because they contain a summary of all the tests.
  • Show lines that contain the word Expect, because they contain the reason the unit test is failing.
  • Show lines that contain the word Exception, because they contain exception information.
  • Discard all other lines.

Let's use grep with a regular expression to parse the output of the test:

dotnet test --no-build --no-restore 2> /dev/null |
grep -Po '^(Failed|Total|\s+Expect|[^:]+Exception).*$' --color=never

This will produce the following information:

Note that the information is both compact and relevant.

CI/CD implementation

Let's take a look at the CI/CD implementation. In most pipelines a restore and build is done before the test is started. To speed things up, I like to test the release binaries that were generated by the previous steps. Just add -c Release to the test runner command. I also want to output the name of the project before testing it. And I like to have a summary at the end of the test step.

Let's examine the bash script:

echo -n > test.txt

for project in `find ./ -wholename "*UnitTests*.csproj"`; do
    echo "Testing $(basename $project)"
    dotnet test $project --no-build --no-restore -c Release 2> /dev/null | 
    grep -Po '^(Failed|Total|\s+Expect|[^:]+Exception).*$' --color=never |
    tee -a test.txt
    echo ""
done

cat test.txt | grep -oP '(?<=Total tests: )\d+' | awk '{ sum += $1 } END { print "Total tests: "sum }'
cat test.txt | grep -oP      '(?<=Passed: )\d+' | awk '{ sum += $1 } END { print "Passed:      "sum }'
cat test.txt | grep -oP      '(?<=Failed: )\d+' | awk '{ sum += $1 } END { print "Failed:      "sum }'
cat test.txt | grep -oP     '(?<=Skipped: )\d+' | awk '{ sum += $1 } END { print "Skipped:     "sum }'

echo ""

TESTFAILED=$(cat test.txt | grep -oP '(?<=Failed: )\d+' | awk '{ sum += $1 } END { print sum }')
rm test.txt

if [[ $TESTFAILED -ne 0 ]] ; then
    echo "Unit test(s) failed."
    exit 1
fi

Commands like tee, cat, grep and awk are part of any Linux installation. When the script runs it looks way better:

So there you have it. Turn off the noise; be more effective.

Note: the screenshots are taken from bash on Windows. I don't use Linux to develop, only to do CI/CD. Installing bash on Windows will help to debug your CI/CD scripts.

expand_less