moby/docs/contributing/test.md
Tibor Vass f1c1cd436a integration-cli: move each test suite to its own TestX testing function
Signed-off-by: Tibor Vass <tibor@docker.com>
2019-09-18 18:26:36 +00:00

8.8 KiB

Run tests

Contributing includes testing your changes. If you change the Moby code, you may need to add a new test or modify an existing test. Your contribution could even be adding tests to Moby. For this reason, you need to know a little about Moby's test infrastructure.

This section describes tests you can run in the dry-run-test branch of your Docker fork. If you have followed along in this guide, you already have this branch. If you don't have this branch, you can create it or simply use another of your branches.

Understand how to test Moby

Moby tests use the Go language's test framework. In this framework, files whose names end in _test.go contain test code; you'll find test files like this throughout the Moby repo. Use these files for inspiration when writing your own tests. For information on Go's test framework, see Go's testing package documentation and the go test help.

You are responsible for unit testing your contribution when you add new or change existing Moby code. A unit test is a piece of code that invokes a single, small piece of code (unit of work) to verify the unit works as expected.

Depending on your contribution, you may need to add integration tests. These are tests that combine two or more work units into one component. These work units each have unit tests and then, together, integration tests that test the interface between the components. The integration and integration-cli directories in the Docker repository contain integration test code. Note that integration-cli tests are now deprecated in the Moby project, and new tests cannot be added to this suite - add integration tests instead using the API client.

Testing is its own specialty. If you aren't familiar with testing techniques, there is a lot of information available to you on the Web. For now, you should understand that, the Docker maintainers may ask you to write a new test or change an existing one.

Run tests on your local host

Before submitting a pull request with a code change, you should run the entire Moby Engine test suite. The Makefile contains a target for the entire test suite, named test. Also, it contains several targets for testing:

Target What this target does
test Run the unit, integration, and docker-py tests
test-unit Run just the unit tests
test-integration Run the integration tests
test-docker-py Run the tests for the Docker API client

Running the entire test suite on your current repository can take over half an hour. To run the test suite, do the following:

  1. Open a terminal on your local host.

  2. Change to the root of your Docker repository.

    $ cd moby-fork
    
  3. Make sure you are in your development branch.

    $ git checkout dry-run-test
    
  4. Run the make test command.

    $ make test
    

    This command does several things, it creates a container temporarily for testing. Inside that container, the make:

    • creates a new binary
    • cross-compiles all the binaries for the various operating systems
    • runs all the tests in the system

    It can take approximate one hour to run all the tests. The time depends on your host performance. The default timeout is 60 minutes, which is defined in hack/make.sh (${TIMEOUT:=60m}). You can modify the timeout value on the basis of your host performance. When they complete successfully, you see the output concludes with something like this:

    Ran 68 tests in 79.135s
    

Run targets inside a development container

If you are working inside a development container, you use the hack/test/unit script to run unit-tests, and hack/make.sh script to run integration and other tests. The hack/make.sh script doesn't have a single target that runs all the tests. Instead, you provide a single command line with multiple targets that does the same thing.

Try this now.

  1. Open a terminal and change to the moby-fork root.

  2. Start a Moby development image.

    If you are following along with this guide, you should have a dry-run-test image.

    $ docker run --privileged --rm -ti -v `pwd`:/go/src/github.com/docker/docker dry-run-test /bin/bash
    
  3. Run the unit tests using the hack/test/unit script.

    # hack/test/unit
    
  4. Run the tests using the hack/make.sh script.

    # hack/make.sh dynbinary binary cross test-integration test-docker-py
    

    The tests run just as they did within your local host.

    Of course, you can also run a subset of these targets too. For example, to run just the integration tests:

    # hack/make.sh dynbinary binary cross test-integration
    

    Most test targets require that you build these precursor targets first: dynbinary binary cross

Run unit tests

We use golang standard testing package or gocheck for our unit tests.

You can use the TESTDIRS environment variable to run unit tests for a single package.

$ TESTDIRS='github.com/docker/docker/opts' make test-unit

You can also use the TESTFLAGS environment variable to run a single test. The flag's value is passed as arguments to the go test command. For example, from your local host you can run the TestValidateIPAddress test with this command:

$ TESTFLAGS='-test.run ^TestValidateIPAddress$' make test-unit

On unit tests, it's better to use TESTFLAGS in combination with TESTDIRS to make it quicker to run a specific test.

$ TESTDIRS='github.com/docker/docker/opts' TESTFLAGS='-test.run ^TestValidateIPAddress$' make test-unit

Run integration tests

We use gocheck for our integration-cli tests. You can use the TESTFLAGS environment variable to run a single test. The flag's value is passed as arguments to the go test command. For example, from your local host you can run the TestBuild test with this command:

$ TESTFLAGS='-test.run TestDockerSuite/TestBuild*' make test-integration

To run the same test inside your Docker development container, you do this:

# TESTFLAGS='-test.run TestDockerSuite/TestBuild*' hack/make.sh binary test-integration

Test the Windows binary against a Linux daemon

This explains how to test the Windows binary on a Windows machine set up as a development environment. The tests will be run against a daemon running on a remote Linux machine. You'll use Git Bash that came with the Git for Windows installation. Git Bash, just as it sounds, allows you to run a Bash terminal on Windows.

  1. If you don't have one open already, start a Git Bash terminal.

    Git Bash

  2. Change to the moby source directory.

    $ cd /c/gopath/src/github.com/docker/docker
    
  3. Set DOCKER_REMOTE_DAEMON as follows:

    $ export DOCKER_REMOTE_DAEMON=1
    
  4. Set DOCKER_TEST_HOST to the tcp://IP_ADDRESS:2376 value; substitute your Linux machines actual IP address. For example:

    $ export DOCKER_TEST_HOST=tcp://213.124.23.200:2376
    
  5. Make the binary and run the tests:

    $ hack/make.sh binary test-integration
    

    Some tests are skipped on Windows for various reasons. You can see which tests were skipped by re-running the make and passing in the TESTFLAGS='-test.v' value. For example

    $ TESTFLAGS='-test.v' hack/make.sh binary test-integration
    

    Should you wish to run a single test such as one with the name 'TestExample', you can pass in TESTFLAGS='-test.run /TestExample'. For example

    $ TESTFLAGS='-test.run /TestExample' hack/make.sh binary test-integration
    

You can now choose to make changes to the Moby source or the tests. If you make any changes, just run these commands again.

Public CI infrastructure

The current infrastructure is maintained here: Moby ci job. The Jenkins infrastructure is for the Moby project is maintained and managed by Docker Inc. All contributions against the Jenkinsfile are appreciated and welcomed! However we might not be able to fully provide the infrastructure to test against various architectures in our CI pipelines. All jobs can be triggered and re-ran by the Moby maintainers

Where to go next

Congratulations, you have successfully completed the basics you need to understand the Moby test framework.