Running GitLab Runner behind a proxy

This guide aims specifically to making GitLab Runner with Docker executor to work behind a proxy.

Before proceeding further, you need to make sure that you’ve already installed Docker and GitLab Runner on the same machine.

Configuring CNTLM

Note: If you already use a proxy without authentication, this section is optional and you can skip straight to configuring Docker. Configuring CNTLM is only needed if you are behind a proxy with authentication, but it’s recommended to use in any case.

CNTLM is a Linux proxy which can be used as a local proxy and has 2 major advantages compared to adding the proxy details everywhere manually:

  • One single source where you need to change your credentials
  • The credentials can not be accessed from the Docker Runners

Assuming you have installed CNTLM, you need to first configure it.

Make CNTLM listen to the docker0 interface

For extra security, and to protect your server from the outside world, you can bind CNTLM to listen on the docker0 interface which has an IP that is reachable from inside the containers. If you tell CNTLM on the Docker host to bind only to this address, Docker containers will be able to reach it, but the outside world won’t be able to.

  1. Find the IP that Docker is using:

     ip -4 -oneline addr show dev docker0
    

    This is usually 172.17.0.1, let’s call it docker0_interface_ip.

  2. Open the config file for CNTLM (/etc/cntlm.conf). Enter your username, password, domain and proxy hosts, and configure the Listen IP address which you found from the previous step. It should look like this:

     Username     testuser
     Domain       corp-uk
     Password     password
     Proxy        10.0.0.41:8080
     Proxy        10.0.0.42:8080
     Listen       172.17.0.1:3128 # Change to your docker0 interface IP
    
  3. Save the changes and restart its service:

     sudo systemctl restart cntlm
    

Configuring Docker for downloading images

Note: The following apply to OSes that have systemd support.

Follow Docker’s documentation how to use a proxy.

The service file should look like this:

[Service]
Environment="HTTP_PROXY=http://docker0_interface_ip:3128/"
Environment="HTTPS_PROXY=http://docker0_interface_ip:3128/"

Adding Proxy variables to the Runner config

The proxy variables need to also be added the Runner’s config, so that it can get builds assigned from GitLab behind the proxy.

This is basically the same as adding the proxy to the Docker service above:

  1. Create a systemd drop-in directory for the gitlab-runner service:

     mkdir /etc/systemd/system/gitlab-runner.service.d
    
  2. Create a file called /etc/systemd/system/gitlab-runner.service.d/http-proxy.conf that adds the HTTP_PROXY environment variable(s):

     [Service]
     Environment="HTTP_PROXY=http://docker0_interface_ip:3128/"
     Environment="HTTPS_PROXY=http://docker0_interface_ip:3128/"
    
  3. Save the file and flush changes:

     systemctl daemon-reload
    
  4. Restart GitLab Runner:

     sudo systemctl restart gitlab-runner
    
  5. Verify that the configuration has been loaded:

     systemctl show --property=Environment gitlab-runner
    

    You should see:

       Environment=HTTP_PROXY=http://docker0_interface_ip:3128/ HTTPS_PROXY=http://docker0_interface_ip:3128/
    

Adding the proxy to the Docker containers

After you registered your Runner, you might want to propagate your proxy settings to the Docker containers (for git clone and other stuff).

To do that, you need to edit /etc/gitlab-runner/config.toml and add the following to the [[runners]] section:

pre_clone_script = "git config --global http.proxy $HTTP_PROXY; git config --global https.proxy $HTTPS_PROXY"
environment = ["https_proxy=docker0_interface_ip:3128", "http_proxy=docker0_interface_ip:3128", "HTTPS_PROXY=docker0_interface_ip:3128", "HTTP_PROXY=docker0_interface_ip:3128"]

Where docker0_interface_ip is the IP address of the docker0 interface. You need to be able to reach it from within the Docker containers, so it’s important to set it right.

Note: In our examples, we are setting both lower case and upper case variables because certain programs expect HTTP_PROXY and others http_proxy. Unfortunately, there is no standard on these kinds of environment variables.

Proxy settings when using dind service

When using the docker-in-docker executor (dind), it can be necessary to specify docker:2375,docker:2376 in the NO_PROXY environment variable. This is because the proxy intercepts the TCP connection between:

  • dockerd from the dind container.
  • docker from the client container.

The ports can be required because otherwise docker push will be blocked as it originates from the IP mapped to docker. However, in that case, it is meant to go through the proxy.

When testing the communication between dockerd from dind and a docker client locally (as described here: https://hub.docker.com/_/docker/), dockerd from dind is initially started as a client on the host system by root, and the proxy variables are taken from /root/.docker/config.json.

For example:

{
    "proxies": {
        "default": {
            "httpProxy": "http://proxy:8080",
            "httpsProxy": "http://proxy:8080",
            "noProxy": "docker:2375,docker:2376"
        }
    }
}

However, the container started for executing .gitlab-ci.yml scripts will have the environment variables set by the settings of the gitlab-runner configuration (/etc/gitlab-runner/config.toml). These are available as environment variables as is (in contrast to .docker/config.json of the local test above) in the dind containers running dockerd as a service and docker client executing .gitlab-ci.yml. In .gitlab-ci.yml, the environment variables will be picked up by any program honouring the proxy settings from default environment variables. For example, wget, apt, apk, docker info and docker pull (but not by docker run or docker build as per: https://github.com/moby/moby/issues/24697#issuecomment-366680499). docker run or docker build executed inside the container of the docker executor will look for the proxy settings in $HOME/.docker/config.json, which is now inside the executor container (and initally empty). Therefore, docker run or docker build executions will have no proxy settings. In order to pass on the settings, a $HOME/.docker/config.json needs to be created in the executor container. For example:

before_script:
  - mkdir -p $HOME/.docker/
  - 'echo "{ \"proxies\": { \"default\": { \"httpProxy\": \"$HTTP_PROXY\", \"httpsProxy\": \"$HTTPS_PROXY\", \"noProxy\": \"$NO_PROXY\" } } }" > $HOME/.docker/config.json'

Because it is confusing to add additional lines in a .gitlab-ci.yml file that are only needed in case of a proxy, it is better to move the creation of the $HOME/.docker/config.json into the configuration of the gitlab-runner (/etc/gitlab-runner/config.toml) that is actually affected:

[[runners]]
  pre_build_script = "mkdir -p $HOME/.docker/ && echo \"{ \\\"proxies\\\": { \\\"default\\\": { \\\"httpProxy\\\": \\\"$HTTP_PROXY\\\", \\\"httpsProxy\\\": \\\"$HTTPS_PROXY\\\", \\\"noProxy\\\": \\\"$NO_PROXY\\\" } } }\" > $HOME/.docker/config.json"
Note: An additional level of escaping " is needed here because this is the creation of a JSON file with a shell specified as a single string inside a TOML file. Because it is not YAML anymore, do not escape the :.

Note that if the NO_PROXY list needs to be extended, wildcards * only work for suffixes but not for prefixes or CIDR notation. For more information, see https://github.com/moby/moby/issues/9145 and https://unix.stackexchange.com/questions/23452/set-a-network-range-in-the-no-proxy-environment-variable.