Building Docker Images on CircleCI 2.0


As part of your build you might want to build Docker images for deploying elsewhere or for further testing. This document explains how to do that on CircleCI 2.0.

For security reasons, the Docker Executor doesn’t allow building Docker images within a job space.

To help users build, run, and publish new images, we’ve introduced a special feature which creates a separate environment for each build. This environment is remote, fully-isolated and has been configured to execute Docker commands.


If your build requires docker or docker-compose commands, you’ll need to add a special step into your .circleci/config.yml:

      # ... steps for building/testing app ...

      - setup_remote_docker

When setup_remote_docker executes, a remote environment will be created, and your current primary container will be configured to use it. Then, any docker-related commands you use will be safely executed in this new environment.


Here’s an example where we build and push a Docker image for our demo docker project:

version: 2
      - image: golang:1.6.4   # (1)
    working_directory: /go/src/
      - checkout
      # ... steps for building/testing app ...

      - setup_remote_docker   # (2)

      # use a primary image that already has Docker (recommended)
      # or install it during a build like we do here
      - run:
          name: Install Docker client
          command: |
            set -x
            curl -L -o /tmp/docker-$VER.tgz$VER.tgz
            tar -xz -C /tmp -f /tmp/docker-$VER.tgz
            mv /tmp/docker/* /usr/bin

      # build and push Docker image
      - run: |
          docker build -t   CircleCI-Public/circleci-demo-docker:$TAG .      # (3)
          docker login -u $DOCKER_USER -p $DOCKER_PASS         # (4)
          docker push CircleCI-Public/circleci-demo-docker:$TAG

Let’s break down what’s happening during this build’s execution:

  1. All commands are executed in the primary container.
  2. Once setup_remote_docker is called, a new remote environment is created, and your primary container is configured to use it.
  3. All docker-related commands are also executed in your primary container, but building/pushing images and running containers happens in the remote Docker Engine.
  4. We use project environment variables to store credentials for Docker Hub.

Separation of Environments

Since the job space and remote docker are separated environments, there’s one caveat: containers running in your job space can’t directly communicate with containers running in remote docker.

Accessing Services

It’s impossible to start a service in remote docker and ping it directly from a primary container (and vice versa). To solve that, you’ll need to interact with a service from remote docker, as well as through the same container:

# start service and check that it’s running
- run: |
    docker run -d --name my-app my-app
    docker exec my-app curl --retry 10 --retry-connrefused http://localhost:8080

A different way to do this is to use another container running in the same network as the target container:

- run: |
    docker run -d --name my-app my-app
    docker run --network container:my-app appropriate/curl --retry 10 --retry-connrefused http://localhost:8080

Mounting Folders

It’s not possible to mount a folder from your job space into a container in Remote Docker (and vice versa). But you can use docker cp command to transfer files between these two environments. For example, you want to start a container in Remote Docker and you want to use a config file from your source code for that:

- run: |
    # creating dummy container which will hold a volume with config
    docker create -v /cfg --name configs alpine:3.4 /bin/true
    # copying config file into this volume
    docker cp path/in/your/source/code/app_config.yml configs:/cfg
    # starting application container using this volume
    docker run --volumes-from configs app-image:1.2.3

In the same way, if your application produces some artifacts that need to be stored, you can copy them from Remote Docker:

- run: |
    # starting container with our application
    # make sure you're not using `--rm` option otherwise container will be killed after finish
    docker run --name app app-image:1.2.3

- run: |
    # once application container finishes we can copy artifacts directly from it
    docker cp app:/output /path/in/your/job/space

If you have any questions, head over to our community forum for support from us and other users.