270 lines (183 sloc) 8.56 KB

Contributing to machine

GoDoc Build Status Windows Build Status Coverage Status

Want to hack on Machine? Awesome! Here are instructions to get you started.

Machine is a part of the Docker project, and follows the same rules and principles. If you're already familiar with the way Docker does things, you'll feel right at home.

Otherwise, please read Docker's contributions guidelines.


The requirements to build Machine are:

  1. A running instance of Docker or a Golang 1.10 development environment
  2. The bash shell
  3. Make

Build using Docker containers

To build the docker-machine binary using containers, simply run:

$ export USE_CONTAINER=true
$ make build

Local Go development environment

Make sure the source code directory is under a correct directory structure; Example of cloning and preparing the correct environment GOPATH:

$ mkdir docker-machine
$ cd docker-machine
$ export GOPATH="$PWD"
$ go get
$ cd src/

If you want to use your existing workspace, make sure your GOPATH is set to the directory that contains your src directory, e.g.:

$ export GOPATH=/home/yourname/work
$ mkdir -p $GOPATH/src/
$ cd $GOPATH/src/ && git clone
$ cd machine        

At this point, simply run:

$ make build

Built binary

After the build is complete a bin/docker-machine binary will be created.

You may call:

$ make clean

to clean-up build results.

Tests and validation

We use the usual go tools for this, to run those commands you need at least the linter which you can install with go get -u

To run basic validation (dco, fmt), and the project unit tests, call:

$ make test

If you want more indepth validation (vet, lint), and all tests with race detection, call:

$ make validate

If you make a pull request, it is highly encouraged that you submit tests for the code that you have added or modified in the same pull request.

Code Coverage

To generate an html code coverage report of the Machine codebase, run:

make coverage-serve

And navigate to http://localhost:8000 (hit CTRL+C to stop the server).

Native build

Alternatively, if you are building natively, you can simply run:

make coverage-html

List of all targets

High-level targets

make clean
make build
make test
make validate

Advanced build targets

Build for all supported OSes and architectures (binaries will be in the bin project subfolder):

make build-x

Build for a specific list of OSes and architectures:

TARGET_OS=linux TARGET_ARCH="amd64 arm" make build-x

You can further control build options through the following environment variables:

DEBUG=true # enable debug build
STATIC=true # build static (note: when cross-compiling, the build is always static)
VERBOSE=true # verbose output
PREFIX=folder # put binaries in another folder (not the default `./bin`)

Scrub build results:

make build-clean

Coverage targets

make coverage-html
make coverage-serve
make coverage-send
make coverage-generate
make coverage-clean

Tests targets

make test-short
make test-long
make test-integration

Validation targets

make fmt
make vet
make lint
make dco

Managing dependencies

When you make a fresh copy of the repo, all the dependencies are in vendor/ directory for the build to work. This project uses golang/dep as vendor management tool. Please refer to dep documentation for further details.

  1. Verify the changes in your repo, commit and submit a pull request

Integration Tests


We use BATS for integration testing, so, first make sure to install it.

Basic Usage

You first need to build, calling make build.

You can then invoke integration tests calling DRIVER=foo make test-integration TESTSUITE, where TESTSUITE is one of the test/integration subfolder, and foo is the specific driver you want to test.


$ DRIVER=virtualbox make test-integration test/integration/core/core-commands.bats
 ✓ virtualbox: machine should not exist
 ✓ virtualbox: create
 ✓ virtualbox: ls
 ✓ virtualbox: run busybox container
 ✓ virtualbox: url
 ✓ virtualbox: ip
 ✓ virtualbox: ssh
 ✓ virtualbox: docker commands with the socket should work
 ✓ virtualbox: stop
 ✓ virtualbox: machine should show stopped after stop
 ✓ virtualbox: machine should now allow upgrade when stopped
 ✓ virtualbox: start
 ✓ virtualbox: machine should show running after start
 ✓ virtualbox: kill
 ✓ virtualbox: machine should show stopped after kill
 ✓ virtualbox: restart
 ✓ virtualbox: machine should show running after restart

17 tests, 0 failures
Cleaning up machines...
Successfully removed bats-virtualbox-test

To invoke a directory of tests recursively:

$ DRIVER=virtualbox make test-integration test/integration/core/

Extra Create Arguments

In some cases, for instance to test the creation of a specific base OS (e.g. RHEL) as opposed to the default with the common tests, you may want to run common tests with different create arguments than you get out of the box.

Keep in mind that Machine supports environment variables for many of these flags. So, for instance, you could run the command (substituting, of course, the proper secrets):

$ DRIVER=amazonec2 \
  AWS_VPC_ID=vpc-xxxxxxx \
  AWS_SECRET_ACCESS_KEY=yyyyyyyyyyyyy \
  AWS_ACCESS_KEY_ID=zzzzzzzzzzzzzzzz \
  AWS_AMI=ami-12663b7a \
  AWS_SSH_USER=ec2-user \
  make test-integration test/integration/core

in order to run the core tests on Red Hat Enterprise Linux on Amazon.


The test/integration directory is layed out to divide up tests based on the areas which the test. If you are uncertain where to put yours, we are happy to guide you.

At the time of writing, there is:

  1. A core directory which contains tests that are applicable to all drivers.
  2. A drivers directory which contains tests that are applicable only to specific drivers with sub-directories for each provider.
  3. A cli directory which is meant for testing functionality of the command line interface, without much regard for driver-specific details.


The best practices for writing integration tests on Docker Machine are still a work in progress, but here are some general guidelines from the maintainers:

  1. Ideally, each test file should have only one concern.
  2. Tests generally should not spin up more than one machine unless the test is deliberately testing something which involves multiple machines, such as an ls test which involves several machines, or a test intended to create and check some property of a Swarm cluster.
  3. BATS will print the output of commands executed during a test if the test fails. This can be useful, for instance to dump the magic $output variable that BATS provides and/or to get debugging information.
  4. It is not strictly needed to clean up the machines as part of the test. The BATS wrapper script has a hook to take care of cleaning up all created machines after each test.


Docker Machine has several included drivers that supports provisioning hosts in various providers. If you wish to contribute a driver, we ask the following to ensure we keep the driver in a consistent and stable state:

  • Address issues filed against this driver in a timely manner
  • Review PRs for the driver
  • Be responsible for maintaining the infrastructure to run unit tests and integration tests on the new supported environment
  • Participate in a weekly driver maintainer meeting

Note: even if those are met does not guarantee a driver will be accepted. If you have questions, please do not hesitate to contact us on IRC.