Run docker-compose build in .gitlab-ci.yml

Question

I have a .gitlab-ci.yml file which contains following:

image: docker:latest

services:
  - docker:dind

before_script:
  - docker info
  - docker-compose --version

buildJob:
  stage: build
  tags:
    - docker
  script:
    - docker-compose build

But in ci-log I receive message:

$ docker-compose --version
/bin/sh: eval: line 46: docker-compose: not found

What am I doing wrong?

webmaster777 · Answer 1 · 2023-03-14T21:17:56.187

111

EDIT (2023): Do not keep using docker-compose! Please migrate to docker compose, which is now part of the docker image and a 99% drop-in replacement of docker-compose. Ironically, if you (now) use docker:latest, the question does not apply. See Docker's documentation on Compose V2. The answer below only applies to docker-compose, AKA Compose V1, which is to be discontinued/become unsupported.

Docker also provides an official docker-compose image: docker/compose

This is the ideal solution if you don't want to install it every pipeline.

Note that in the latest version of GitLab CI/Docker you will likely need to give privileged access to your GitLab CI Runner and configure/disable TLS. See Use docker-in-docker workflow with Docker executor

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

# Official docker compose image.
image:
  name: docker/compose:latest

services:
  - docker:dind

before_script:
  - docker version
  - docker-compose version

build:
  stage: build
  script:
    - docker-compose down
    - docker-compose build
    - docker-compose up tester-image

Note that in versions of docker-compose earlier than 1.25:

Since the image uses docker-compose-entrypoint.sh as entrypoint you'll need to override it back to /bin/sh -c in your .gitlab-ci.yml. Otherwise your pipeline will fail with No such command: sh

    image:
      name: docker/compose:latest
      entrypoint: ["/bin/sh", "-c"]

edited Mar 14 '23 at 21:17

answered Oct 10 '18 at 06:34

webmaster777

1,442
2
12
17

This works, but as your linked source says, it will need a GitLab **9.4**+ , because you **need* to overwrite the entrypoint in the CI yaml. – Christian Ulbrich Oct 24 '18 at 09:58
11

@webmaster777 I'm getting this error when the execution reaches "docker-compose version"... `Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running? ERROR: Job failed: exit code 1`... how can I solve it? – Marcello DeSales Oct 27 '18 at 05:28
7

It only worked when I added the variables `DOCKER_HOST: tcp://docker:2375/` and `DOCKER_DRIVER: overlay2` – Marcello DeSales Oct 27 '18 at 10:06
That depends on your docker/runner setup. See https://docs.gitlab.com/ee/ci/docker/using_docker_build.html There are various ways to set up the runner, so you should add any needed config where necessary. – webmaster777 Oct 29 '18 at 13:49
About the need to override the entrypoint, I opened an issue at https://github.com/docker/compose/issues/6614, some +1 could help to make it fix! thanks you. – Thomas Decaux Mar 31 '19 at 17:34
1

It works! If bind mounts or volumes are involved, some environment variables are needed as well -- `DOCKER_HOST=tcp://docker:2375` and `DOCKER_DRIVER=overlay2` (optional) -- to instruct docker to talk with the daemon started inside of the service (`docker:dind`). Refer to [Use docker-in-docker executor](https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#use-docker-in-docker-executor) for more information. – Jeremy Kao May 11 '19 at 03:50
4

This still doesn't work for me (shared gitlab.com runners) even when all variables are set, `Cannot connect to the Docker daemon at tcp://docker:2375/. Is the docker daemon running?` – The Godfather Jun 12 '20 at 13:58
@TheGodfather Probably because of TLS restrictions. See https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#troubleshooting – webmaster777 Jul 07 '20 at 15:22
@webmaster777 I did read this, but that's why I highlighted - shared Gitlab.com runners. What *they* provide. I have nothing to do with Docker version, TLS or whatsoever, I'm using shared runners and I expect them to work out-of-the-box. – The Godfather Jul 07 '20 at 16:29
I haven't used gitlab.com shared runners, but from what I can see [here](https://docs.gitlab.com/ee/user/gitlab_com/index.html#configtoml) they've disabled tls by default, but do mount `/certs/client` for some reason. Do you have the service definition `docker:dind` in your (job) config? And yes you do have to comply with whatever they are using, and properly run docker-in-docker, otherwise it will not work. Also, the dind version and docker-compose version should be compatible. Using "latest" tags is not recommended. – webmaster777 Jul 10 '20 at 07:16
With the shell executor, it worked without any environment variables, but my sysadmin had a custom MTU, so I had to change in `/etc/docker/daemon.json` the mtu value. – barfoos Mar 04 '21 at 10:58
1

The `docker/compose` image is deadware. Hasn't recieved updates since 2 years ago. It still features docker 19.03 and the deprecated compose V1. – N1ngu Mar 13 '23 at 19:37
@N1ngu that is a valid point, but I'm guessing users of compose v2 don't need this, because it is part of the docker image now (`docker-compose` has been replaced with `docker compose`) see https://docs.docker.com/compose/compose-v2/ – webmaster777 Mar 14 '23 at 20:59

n2o · Answer 2 · 2020-06-12T21:05:51.063

99

Following the official documentation:

# .gitlab-ci.yml
image: docker
services:
  - docker:dind    
build:
  script:
    - apk add --no-cache docker-compose
    - docker-compose up -d

Sample docker-compose.yml:

version: "3.7"
services:
  foo:
    image: alpine
    command: sleep 3
  bar:
    image: alpine
    command: sleep 3

We personally do not follow this flow anymore, because you loose control about the running containers and they might end up running endless. This is because of the docker-in-docker executor. We developed a python-script as a workaround to kill all old containers in our CI, which can be found here. But I do not suggest to start containers like this anymore.

edited Jun 12 '20 at 21:05

answered Mar 09 '17 at 14:24

n2o

6,175
3
28
46

@tmaier I think i am stuck here too. What kind of "Runner-Installation" did you use? I mean which of the three ways did you go? https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#runner-configuration – Max Schindler May 08 '17 at 09:23
We are using the docker-in-docker executor https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#use-docker-in-docker-executor – n2o May 08 '17 at 12:27
3

While this works, it means that on **every** build docker-compose needs to be installed (and fetched...) and thus it will increase the build time. If you have multiple jobs needing docker-compose (and in a production scenario you must likely will), build time will considerably expand. – Christian Ulbrich Sep 12 '18 at 00:19
1

If you want to use it more frequently, you can define your own docker image für your CI with a small Dockerfile containing online three lines: ` FROM alpine \n RUN apk add --no-cache py-pip && pip install docker-compose ` But keep in mind to keep this image always up-to-date – n2o Sep 12 '18 at 10:21
Thank you very much for your answer. I just have an issue with the networking. It seems no container can communicate with another… Am I the only one in this situation? (using shared runners) – Einenlum Oct 18 '18 at 12:32
This solution is no longer working for me as of now. However alpine linux now has a docker-compose package in their edge branch, which I think is the way to go. – m0etaz Apr 09 '19 at 10:22
regarding your "Update from 2019-04-09", maybe you should also use `docker-compose down` to terminate all running services, after your CI task is done? – webmaster777 Apr 25 '19 at 06:04
try: `services: - docker:18-dind` – filthysocks Aug 06 '19 at 06:30
"because you loose control about the running containers". Why not? Can't you do `docker-compose down`? – Max O. Oct 09 '22 at 12:24

tmaier · Answer 3 · 2023-03-15T08:34:48.390

31

Deprecated

I created a simple docker container which has docker-compose installed on top of docker:latest. See https://hub.docker.com/r/tmaier/docker-compose/ or https://github.com/tmaier/docker-compose

Your .gitlab-ci.yml file would look like this:

image: tmaier/docker-compose:latest

services:
  - docker:dind

before_script:
  - docker info
  - docker-compose --version

buildJob:
  stage: build
  tags:
    - docker
  script:
    - docker-compose build

edited Mar 15 '23 at 08:34

answered Jan 04 '17 at 01:49

tmaier

953
9
17

getting this error in gitlab ci `error: command 'gcc' failed with exit status 1` – notacorn Jul 07 '20 at 18:27
Just installing the deprecated compose V1 on top of `docker:latest` when that already features compose V2 out-of-the-box. – N1ngu Mar 14 '23 at 07:31

n2o · Accepted Answer · 2023-03-14T08:39:25.800

15

Deprecated

docker-compose can be installed as a Python package, which is not shipped with your image. The image you chose does not even provide an installation of Python:

$ docker run --rm -it docker sh
/ # find / -iname "python"
/ #

Looking for Python gives an empty result. So you have to choose a different image, which fits to your needs and ideally has docker-compose installed or you maually create one.

The docker image you chose uses Alpine Linux. You can use it as a base for your own image or try a different one first if you are not familiar with Alpine Linux.

I had the same issue and created a Dockerfile in a public GitHub repository and connected it with my Docker Hub account and chose an automated build to build my image on each push to the GitHub repository. Then you can easily access your own images with the GitLab CI.

edited Mar 14 '23 at 08:39

answered Oct 05 '16 at 08:31

n2o

6,175
3
28
46

5

Can you delete one of your answers? Confusing to find two, plus lots of "EDIT" messages. (We can see edit history if we need to). – Duncan Jones Jun 10 '19 at 08:35
This would install the deprecated compose V1 – N1ngu Mar 14 '23 at 07:26
2

I cannot delete an accepted answer :-/ – n2o Mar 14 '23 at 08:39

score 10 · Answer 5 · edited Oct 25 '18 at 14:37

If you don't want to provide a custom docker image with docker-compose preinstalled, you can get it working by installing Python during build time. With Python installed you can finally install docker-compose ready for spinning up your containers.

image: docker:latest

services:
- docker:dind

before_script:
- apk add --update python py-pip python-dev && pip install docker-compose # install docker-compose
- docker version
- docker-compose version

test:
  cache:
    paths:
    - vendor/
  script:
  - docker-compose up -d
  - docker-compose exec -T php-fpm composer install --prefer-dist
  - docker-compose exec -T php-fpm vendor/bin/phpunit --coverage-text --colors=never --whitelist src/ tests/

Use docker-compose exec with -T if you receive this or a similar error:

$ docker-compose exec php-fpm composer install --prefer-dist
Traceback (most recent call last):
  File "/usr/bin/docker-compose", line 9, in <module>
    load_entry_point('docker-compose==1.8.1', 'console_scripts', 'docker-compose')()
  File "/usr/lib/python2.7/site-packages/compose/cli/main.py", line 62, in main
    command()
  File "/usr/lib/python2.7/site-packages/compose/cli/main.py", line 114, in perform_command
    handler(command, command_options)
  File "/usr/lib/python2.7/site-packages/compose/cli/main.py", line 442, in exec_command
    pty.start()
  File "/usr/lib/python2.7/site-packages/dockerpty/pty.py", line 338, in start
    io.set_blocking(pump, flag)
  File "/usr/lib/python2.7/site-packages/dockerpty/io.py", line 32, in set_blocking
    old_flag = fcntl.fcntl(fd, fcntl.F_GETFL)
ValueError: file descriptor cannot be a negative integer (-1)
ERROR: Build failed: exit code 1

Not a common way, actually you can get "apk: command not found" — Wishmaster, Apr 06 '19 at 06:50
this no longer works, possibly because of the newer docker images. now `python` and `python-dev` are no longer valid apk package adds, and then when you specify python2, `gcc` is missing somehow — notacorn, Jul 07 '20 at 18:23
Outdated. Official docker images already feature compose V2. Instead, this would shadow that with the deprecated compose V1 — N1ngu, Mar 14 '23 at 07:27

score 5 · Answer 6 · answered Nov 02 '20 at 22:19

5

there is tiangolo/docker-with-compose which works:

image: tiangolo/docker-with-compose

stages:
  - build
  - test
  - release
  - clean

  
before_script:
  - docker login -u gitlab-ci-token -p $CI_BUILD_TOKEN registry.gitlab.com

build:
  stage: build
  script:
    - docker-compose -f docker-compose-ci.yml build --pull 


test1:
    stage: test
    script:
        - docker-compose -f docker-compose-ci.yml up -d
        - docker-compose -f docker-compose-ci.yml exec -T php ...

answered Nov 02 '20 at 22:19

olidem

1,961
2
20
45

1

This does not work for me. I get "error during connect: Post http://docker:2375/v1.40/auth: dial tcp: lookup docker on 169.254.169.254:53: no such host". webmaster777's solution, however, worked for me and seems like the best solution as the main reason tiangolo/docker-with-compose was created was because docker/compose previously had an entrypoint problem. Now that that is not an issue, I think this *might* no longer be a good solution. – chaitan94 May 07 '21 at 10:15

score 3 · Answer 7 · edited May 01 '20 at 06:26

3

I think most of the above are helpful, however i needed to collectively apply them to solve this problem, below is the script which worked for me

I hope it works for you too

Also note, in your docker compose this is the format you have to provide for the image name

<registry base url>/<username>/<repo name>/<image name>:<tag>

image:
  name: docker/compose:latest
  entrypoint: ["/bin/sh", "-c"]

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

services:
  - docker:dind

stages:
- build_images

before_script:
  - docker version
  - docker-compose version
  - docker login -u $CI_REGISTRY_USER -p $CI_JOB_TOKEN $CI_REGISTRY

build:
  stage: build_images
  script:
    - docker-compose down
    - docker-compose build
    - docker-compose push

edited May 01 '20 at 06:26

Philipp Panik

254
1
4
15

answered Mar 09 '20 at 04:11

Basav

3,176
1
22
20

The `docker/compose` image is deadware, no updates for 2 years and still featuring docker 19 and the deprecated compose V1 – N1ngu Mar 14 '23 at 07:25
1

Never said docker-compose was dead, but the `docker/compose` image is. No hard feelings, just trying to signal that an answer should be avoided. The answer was good advice 3 years ago, just not anymore. – N1ngu Mar 21 '23 at 10:26

score 2 · Answer 8 · answered Jun 12 '20 at 14:18

It really took me some time to get it working with Gitlab.com shared runners.

I'd like to say "use docker/compose:latest and that's it", but unfortunately I was not able to make it working, I was getting Cannot connect to the Docker daemon at tcp://docker:2375/. Is the docker daemon running? error even when all the env variables were set.

Neither I like an option to install five thousands of dependencies to install docker-compose via pip.

Fortunately, for the recent Alpine versions (3.10+) there is docker-compose package in Alpine repository. It means that @n2o's answer can be simplified to:

test:
  image: docker:19.03.0

  variables:
    DOCKER_DRIVER: overlay2
    # Create the certificates inside this directory for both the server
    # and client. The certificates used by the client will be created in
    # /certs/client so we only need to share this directory with the
    # volume mount in `config.toml`.
    DOCKER_TLS_CERTDIR: "/certs"

  services:
    - docker:19.03.0-dind

  before_script:
    - apk --no-cache add docker-compose      # <---------- Mind this line
    - docker info
    - docker-compose --version

  stage: test
  script:
      - docker-compose build

This worked perfectly from the first try for me. Maybe the reason other answers didn't was in some configuration of Gitlab.com shared runners, I don't know...

This is really the solution on 2021, all others didn't work with a custom runner — Vlax, Mar 12 '21 at 14:23

score 0 · Answer 9 · answered Mar 14 '23 at 11:31

As a consequence of https://github.com/docker-library/docker/pull/361 just do

# docker-compose.yml
version: '2'
services:
  hello:
    image: hello-world

# .gitlab-ci.yml
myjob:
  image: docker:latest
  services:
    - docker:dind
  script:
    - docker --version
    - docker compose version

If you are not using the official docker "docker" image, do not install compose with a package manager as most distributions are outdated. Official install instructions point to either use the downloads.docker.com repository in its diverse forms for each Linux distribution or to simply download the github release artifact, which I think is leaner for a CI usage:

myjob:
    - mkdir -p ~/.docker/cli-plugins
    - (wget --no-verbose --no-clobber https://github.com/docker/compose/releases/latest/download/docker-compose-linux-x86_64 --output-document ~/.docker/cli-plugins/docker-compose || true) # ideally you would cache this binary
    - chmod a+x ~/.docker/cli-plugins/docker-compose
    - ln --symbolic ~/.docker/cli-plugins/docker-compose /usr/local/bin/docker-compose
  script:
    - docker-compose version

score -1 · Answer 10 · answered Apr 09 '19 at 10:25

-1

Alpine linux now has a docker-compose package in their "edge" branch, so you can install it this way in .gitlab-ci.yml


a-job-with-docker-compose:
  image: docker
  services:
    - docker:dind
  script:
    - apk add docker-compose --update-cache --repository http://dl-3.alpinelinux.org/alpine/edge/testing/ --allow-untrusted
    - docker-compose -v

answered Apr 09 '19 at 10:25

m0etaz

953
1
8
21

not working.. needs to update answer.. When run `docker-compose` throws `pkg_resources.DistributionNotFound: The 'docker-compose==1.24.0' distribution was not found and is required by the application` – Carlos.V Jun 20 '19 at 18:54
This installs the deprecated compose V1 – N1ngu Mar 14 '23 at 07:24

Run docker-compose build in .gitlab-ci.yml

10 Answers10

Linked