The Docker Registry comes with a sample configuration file, config_sample.yml. Copy this to config.yml to provide a basic configuration:

cp config_sample.yml config.yml

Inside the config.yml file we can see a selection of configuration headings called flavors: common, dev, prod, etc.

You can specify what flavor to run with the SETTINGS_FLAVOR environment variable.

$ export SETTINGS_FLAVOR=prod

The common flavor overrides and is inherited by all other flavors. If you don't specify a flavor when running the Docker Registry the dev flavor will be used.

docker run samalba/docker-registry

NOTE: The container will try to allocate the port 5000 by default, if the port is already taken, find out which one has been taken by running "docker ps"

Install the system requirements for building a Python library:

sudo apt-get install build-essential python-dev libevent-dev python-pip

Then install the Registry app:

sudo pip install -r requirements.txt

sudo yum install python-devel libevent-devel python-pip

NOTE: On RHEL and CentOS you will need the EPEL repostitories enabled. Fedora should not require the additional repositories.

Then install the Registry app:

sudo python-pip install -r requirements.txt

gunicorn --access-logfile - --debug -k gevent -b 0.0.0.0:5000 -w 1 wsgi:application

The first time someone tries to push to your registry, it will prompt them for a username, password, and email.

The recommended setting to run the Registry in a prod environment is gunicorn behind a nginx server which supports chunked transfer-encoding (nginx >= 1.3.9).

You could use for instance supervisord to spawn the registry with 8 workers using this command:

gunicorn -k gevent --max-requests 100 --graceful-timeout 3600 -t 3600 -b localhost:5000 -w 8 wsgi:application

Note that when using multiple workers, the secret_key for the Flask session must be set explicitly in config.yml. Otherwise each worker will use its own random secret key, leading to unpredictable behavior.

The nginx configuration will look like:

location / {
  proxy_pass        http://localhost:5000;
  proxy_set_header  X-Real-IP  $remote_addr;
}

And you might want to add Basic auth on Nginx to protect it (if you're not using it on your local network):

NOTE: The central Registry runs on the dotCloud platform:

cd dotcloud-registry/
dotcloud create myregistry
dotcloud push

$ cd test
$ python -m unittest discover

The file workflow.py is bit special since it's a functional test (not a unit test). It requires a server to be running in order to succeed.

$ DOCKER_CREDS="user:password" python -m unittest workflow

DOCKER_CREDS contains user credentials information to connect to the staging index server.

If you want to submit a pull request, an important point is to clear up all flake8 warning you could introduce (ignore the one about registry/init.py).

$ pip install flake8
$ find . -name '*.py' -exec flake8 {} \;