Cannot connect to the Docker daemon at unix:///var/run/docker.sock in gitlab CI

Question

I looked at any other questions but can't find my own solution! I setting up a CI in gitlab and use the gitlab's shared runner. In build stage I used docker image as base image but when i use docker command it says :

Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?

I looked at this topic but still don't understand what should I do?

.gitlab-ci.yml :

stages:
  - test
  - build
  - deploy

job_1:
  image: python:3.6
  stage: test
  script:
    - sh ./sh_script/install.sh
    - python manage.py test -k

job_2:
  image: docker:stable
  stage: build
  before_script:
    - docker info
  script:
    - docker build -t my-docker-image .

I know that the gitlab runner must registered to use docker and share /var/run/docker.sock! But how to do this when using the gitlab own runner?

Answer 1

Ahh, that's my lovely topic - using docker for gitlab ci. The problem you are experiencing is better known as docker-in-docker.

Before configuring it, you may want to read this brilliant post: http://jpetazzo.github.io/2015/09/03/do-not-use-docker-in-docker-for-ci/

That will give you a bit of understanding what is the problem and which solution best fits you. Generally there are 2 major approaches: actual installation of docker daemon inside docker and sharing host's daemon to containers. Which approach to choose - depends on your needs.

In gitlab you can go in several ways, I will just share our experience.

Way 1 - using docker:dind as a service.

It is pretty simple to setup. Just add docker:dind as a shared service to your gitlab-ci.yml file and use docker:latest image for your jobs.

image: docker:latest  # this sets default image for jobs
services:
  - docker:dind

Pros:

• simple to setup.
• simple to run - your source codes are available by default to your job in cwd because they are being pulled directly to your docker runner

Cons: you have to configure docker registry for that service, otherwise you will get your Dockerfiles built from scratch each time your pipeline starts. As for me, it is unacceptable, because can take more than an hour depending on the number of containers you have.

Way 2 - sharing /var/run/docker.sock of host docker daemon

We setup our own docker executor with docker daemon and shared the socket by adding it in /etc/gitlab-runner/config.toml file. Thus we made our machine's docker daemon available to docker cli inside containers. Note - you DONT need privileged mode for executor in this case.

After that we can use both docker and docker-compose in our custom docker images. Moreover, we dont need special docker registry because in this case we share executor's registry among all containers.

Cons

You need to somehow pass sources to your containers in this case, because you get them mounted only to docker executor, but not to containers, launched from it. We've stopped on cloning them with command like git clone $CI_REPOSITORY_URL --branch $CI_COMMIT_REF_NAME --single-branch /project

Answer 2

this worked for me. specifically the DOCKER_HOST variable was the final straw to get success. couldn't tell you why though, sorry!

# .gitlab-ci.yml

# github runs docker right out of the box
# gitlab... not so much apparently
# https://stackoverflow.com/a/70813945/5563327
services:
  - docker:dind

variables:
  DOCKER_HOST: tcp://docker:2375
  DOCKER_TLS_CERTDIR: ""
  DOCKER_DRIVER: overlay2