Docker and tooling for using ember-cli in a reproducable way.

Our ember-cli builds have not been as reproducable as we'd have wanted them to be. Tooling can differ across machines because the operating systems are different, therefore yielding different versions of nodejs/iojs or different versions of sass bindings. The sass bindings are what pushed us over the edge to try out Dockers for sharing our build environment.

With the advent of user-namespaces in Docker, mounting volumes with the right privileges has become transparant. (see http://www.jrslv.com/docker-1-10/#usernamespacesindocker for some basic info)

The arguments you need to pass to the Docker run command for it to be useful are too cumbersome, hence we've created scripts to help you out.

We have 4 commands, each for a different use-case. Run them at the root of your project.

ed is your default friend. ed helps you install npm & bower dependencies, install new ember dependencies and run any other non-interactive ember command.

# Install a dependency
ed ember install ember-cli-coffeescript
# Install all current node modules
ed npm install
# Install bower components
ed bower install

eds launches the ember server for you.

# No nonsense ember server
eds
# Proxying to your localhost (note it's been renamed from localhost to host)
eds --proxy=http://host:8080
# Serving on a non default port
eds --port=4000 --live-reload-port=64000

edi is the interactive version of ed. It can ask you questions and you can provide interactive answers.

# Generate a route
edi ember generate route epic-win
# Release a new minor version
edi ember release --minor

edl is your friend when developing addons. It provides a replacement for npm link and npm unlink that works in docker-ember.

# Create a global symlink of your addon
cd your-ember-addon
edl
# Use that addon in another project
cd your-ember-project
edl your-ember-addon
# Remove the global symlink of your addon
cd your-ember-addon
edl -u

Note: edl assumes edi is available on your PATH

Assuming you have docker set up correctly, simply clone this repository and add the bin folder to your path.

git clone https://github.com/madnificent/docker-ember.git
echo "export PATH=\$PATH:`pwd`/docker-ember/bin" >> ~/.bashrc
source ~/.bashrc

We suggest to use brew installation scripts to account for specific issues related to docker for mac. See: https://github.com/mu-semtech/homebrew-scripts

You can configure the Ember version in ~/.config/edi/settings using the VERSION variable.

VERSION="3.15.1"

By default ed* commands run as root in the docker container, this means newly created files will be owned as root as well. To avoid this you can use user namespaces to map the container's root user to your own user. This requires some minimal configuration.

Note: on ubuntu 16.04 your user needs to part of the docker group so that it has access to /var/run/docker.sock

Assuming systemd and access to the id command the following steps should suffice:

echo "$( whoami ):$(id -u):65536" |  sudo tee -a /etc/subuid
echo "$( whoami ):$(id -g):65536" |  sudo tee -a /etc/subgid

For systemd you can use the following command:

systemctl edit docker.service

The config file might look this:

[Service]
ExecStart=
ExecStart=/usr/bin/dockerd --userns-remap="your-user-name"

More information on user namespaces is available in the docker documentation

Mac uses a login shell when launching the default terminal app, which slightly changes the desired setup. Sharing the ssh-agent socket currently doesn't work, and thus requires a workaround.

Docker for Mac creates files under the right username automatically. Mac does use a login shell when launching the default terminal app, rather than an interactive shell. These shells don't read the standard ~/.bashrc file, but rather the ~/.bash_profile file. Make sure the following is present in your ~/.bash_profile so ~/.bashrc is always read.

if [ -f ~/.bashrc ]; then
   source ~/.bashrc
fi

The ssh-agent's socket can't be shared with Docker for Mac at the time of writing. A common workaround is to use a Docker container in which a new ssh-agent is ran. We advise the use of the https://github.com/10eTechnology/docker-ssh-agent-forward and have integrated this in the supplied scripts. On mac, this solution is assumed to be installed.

It's important to note that since NPM v7 peer dependencies are installed by default when executing npm install. As of v3.27 the docker-ember image contains NPM >= v7.x. Using this version may lead to resolution conflict errors on peer dependencies in existing projects. NPM provides a --legacy-peer-deps flag to make npm install behave like in previous versions, not installing peer dependencies by default. This can help teams to gradually fix the peer dependency version conflicts in their project.

The --legacy-peer-deps flag can be enabled project-wide by adding legacy-peer-deps=true to .npmrc. In that case it's import to copy that file inside your project's Docker build before running npm install.

Example Dockerfile:

FROM madnificent/ember
COPY package.json .
COPY .npmrc .   # <--- this line must be added
RUN npm install

Some experimental features have been added which optimize the way the Docker daemon is called. These features may behave oddly when developing addons. It may be required to restart certain daemons after using edl, or to disable features when using edl.

Some systems take more time than necessary to spin up a new docker daemon. For these cases, you may choose to keep a daemon alive and send commands to the daemon. Set EDI_USE_EDI_DAEMON to a non-empty string to enable this feature.

Note: You will have to restart the daemon after using edl.

The ember docker links locally developed node modules. Some optimizations are possible in this regard but they break edl.

We mount all available node modules in a consistent way when you use edl. Mounting many volumes may lead to a slow-down on some systems. You may choose to mount only the used linked modules by setting EDI_MOUNT_ONLY_USED_LINKED_MODULES to a non-empty string.

When using older versions of node when developing nested node modules the builds may fail because the right node submodules are not included. This is not the case when the symlinks are removed. Setting both EDI_MOUNT_ONLY_USED_LINKED_MODULES and EDI_MOUNT_USED_NODE_MODULES_WITHOUT_SYMLINKS will mount the used node_modules directly. This may also have a positive performance impact, but we did not run benchmarks.

These optimizations should be disabled when running edl as edl will not be able to find the addons to link.

We assume you are running an SSH agent container as mentioned earlier. If your application never reaches to the outside world using your ssh key, you may disable this feature.

On Mac, the socket of the native SSH agent can't be shared to the Docker image like we do in Linux. We assume the necessary tooling is available on Mac to share the SSH agent. Should you want to disable this feature, set EDI_SSH_AGENT_CONTAINER to an empty string. If you want to force it to be turned on on Linux, then set it to a non-empty string.

When you disable this option, your locally running socket will be shared. When Docker for Mac starts supporting this feature, that will be the superior option.