How do I push to a repo from within a gitlab CI pipeline?

Question

In my CI pipeline I am generating an artifact public/graph.png that visualises some aspect of my code. In a later step I want to commit that to the repo from within the CI pipeline. Here's the pertinent part of .gitlab-ci.yml:

commit-graph:
  stage: pages
  script:
    - git config user.email "cipipeline@example.com"
    - git config user.name "CI Pipeline"
    - cd /group/project
    - mv public/graph.png .
    - git add graph.png
    - git commit -m "committing graph.png [ci skip]"
    - echo $CI_COMMIT_REF_NAME
    - git push origin HEAD:$CI_COMMIT_REF_NAME

When the pipeline runs within gitlab it fails with:

$ git config user.email "cipipeline@dhgitlab.dunnhumby.co.uk"
$ git config user.name "CI Pipeline"
$ cd /group/project
$ mv public/graph.png .
$ git add graph.png
$ git commit -m "committing graph.png [ci skip]"
[detached HEAD 22a50d1] committing graph.png [ci skip]
1 file changed, 0 insertions(+), 0 deletions(-)
create mode 100644 graph.png
$ echo $CI_COMMIT_REF_NAME
jamiet/my-branch
$ git push origin HEAD:$CI_COMMIT_REF_NAME
fatal: unable to access 'https://gitlab-ci-token:xxxxxxxxxxxxxxxxxxxx@example.com/group/project/project.git/': server certificate verification failed. CAfile: /etc/ssl/certs/ca-certificates.crt CRLfile: none

Not sure what I'm doing wrong and don't know enough about SSL to understand that error. Can anyone advise?

We are hosting gitlab ourselves by the way.

Answer 1

From GitLab 17.2 on^[1], you can now allow your CI/CD job token to push to its repository. You do not longer have to maintain an entire account or an access token for that purpose. It is disabled by default (docs), but you can enable it, when needed:

To grant permission to job tokens generated in your project to push to the project's repository:

1. On the left sidebar, select Search or go to and find your project.
2. Select Settings > CI/CD.
3. Expand Job token permissions.
4. In the Permissions section, select Allow Git push requests to the repository.

You can then push changes like this:

  script:
    - ...
    - git config --global user.email "$GITLAB_USER_EMAIL"
    - git config --global user.name "$GITLAB_USER_NAME"
    - git remote set-url --push origin "https://gitlab-ci-token:${CI_JOB_TOKEN}@${CI_SERVER_HOST}/${CI_PROJECT_PATH}.git"
    - git add -A
    - 'git commit -m "refactor: apply changes from pipeline $CI_PIPELINE_ID"'
    - git push origin HEAD:$CI_COMMIT_REF_NAME

Note: You do not longer have to maintain any secrets for this, which makes work a lot easier.

---

^{[1] In GitLab 17.2 and later minor releases, this feature is hidden behind a feature flag. You have to enable allow_push_repository_for_job_token to make the option available in the UI. Currently, it is estimated that this feature will become GA in 17.6. For updates, have a look at this issue.}

Answer 2

Here's a less brittle way of pushing to the repository from its own CI, that I use in my daily work. The keys used here do not expire, as is the case with PAT in Gitlab v16 and up. It pushes to master directly from a detached head:

1. Generate an RSA key and add it as a Project Deploy Key with write access (the public part).
2. Put the private part into your CI/CD variables from inside of your project settings as SSH_PUSH_KEY. Make sure to set it as protected.
3. Add a CI_KNOWN_HOSTS variable, with the SSH fingerprint of your GitLab instance (remember that thing ssh asks you about the first time you try to connect to a host? That.).

Use ssh-keyscan <gitlab-host> to get it. It will look similar to this:

my.gitlab.instance.com ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEArlUMUmNj59PpoLyy4EsKbwhPUfXxuAzFN7dMKDXVvKMmN8344HqQV1tRx6fcmH+0BXK1JAP4f10V0VnYti3e1c5f9dhpl8pIqKLMJgdGDq3MLqjihL3bp5xm8nDsPTm5FoEPPYK1I3M2wr18pBB19evz64NHrK5R/HO5LyTrybVasFumt8cZoH6crnCFgfQWV1mHAG3j41Q0z4yxu6g8zBWESZcVVn90HxQH7+LDHx11122233344491MQGl5fZcKqVWsWQVEssaK87iBsWUxvsuoeVUrj4YRcmbi6F4+ZZZZZZZwwww3ZboWsSWxTk5ESR6WWHccBm8GQflXyY3ZQ==

4. Set up your job inside .gitlab-ci.yml as follows. Set stage and resource_group options appropriately - without the latter you might run into race conditions. Also, make sure to set only properly, as otherwise your CI might trigger itself:

"This CI job pushes to its own repo":
    stage: my_push_stage
    resource_group: this_option_comes_handy_when_pushing
    only:
        - triggers
    before_script:
        - mkdir ~/.ssh/
        - echo "${CI_KNOWN_HOSTS}" > ~/.ssh/known_hosts
        - echo "${SSH_PUSH_KEY}" > ~/.ssh/id_rsa
        - chmod 600 ~/.ssh/id_rsa
        - git config user.email "ci@example.com"
        - git config user.name "CI"
        - git remote remove ssh_origin || true  # Local repo state may be cached
        - git remote add ssh_origin "git@$CI_SERVER_HOST:$CI_PROJECT_PATH.git"
    script:
        - touch "xyz"  # Make an edit
        - git add "xyz"
        - git commit -m "My CI commit"
        - git push ssh_origin HEAD:master  # ❗ this pushes to master, 
                                           # use $CI_COMMIT_REF_NAME if you want to push to current branch
        - git tag MyCiTag  # If you need to add a tag you can do that too
        - git push --tags ssh_origin

Answer 3

None of these worked immediately for me but I merged them all together to come up with this:

- git config user.email "example@example.com"
- git config user.name "example"
- git remote remove dev || true
- git remote add dev https://example:$PROJECT_ACCESS_TOKEN@gitlab.com/demo/versioning-test.git
- git remote set-url --push origin "https://example:$PROJECT_ACCESS_TOKEN@gitlab.com/demo/versioning-test.git"
- git pull origin dev
- git checkout dev 
- git reset --hard origin/dev
- git rm build_number.json || true
#grabs the build number from tags and pipes it into text file json formatted
- echo "`echo {'"build_number"':` `git describe --tags` `echo }`" | sed -e 's/\s\+/"/g' >> build_number.json
- git add build_number.json
- git commit -m "push back from pipeline [skip ci]"
- git push origin dev 

My use case was that we wanted to bump to a version file and push that back into the repo.

Answer 4

I can commit from Gitlab-CI with a selected user with a minor change based on tsr's answer https://stackoverflow.com/a/57800614/5269825 :

# set remote URL to https://oauth2:<AccessToken>@server.com/project.git
CI_PUSH_REPO=`echo "$CI_REPOSITORY_URL $ACCESS_TOKEN_PARAM" | sed 's/^.*\(@.*\)\s\(.*\)/https:\/\/oauth2:\2\1/g'`
git config http.sslverify false
git remote set-url --push origin "${CI_PUSH_REPO}"
git config user.name "Token Owner"
git config user.email "tokenowner@email.com"

# runner runs on a detached HEAD, create a temporary local branch for editing
git checkout -b ci_processing
# make your changes

# push changes
# always return true so that the build does not fail if there are no changes
git push origin ci_processing:${CI_COMMIT_REF_NAME} || true

The ACCESS_TOKEN_PARAM must be configured at the project's CI/CD Variables configuration.

The idea of using Oauth2 and Access Token was taken from https://stackoverflow.com/a/52074198/5269825 and https://stackoverflow.com/a/52154378/5269825.

Also, pushing changes can trigger a new pipeline!

Answer 5

TL;DR;

If you just want to make a release or version bump commits when stuff is merged to main, there are probably existing CLI tools that you can use in a much more simpler and well documented way.
(See Use the gitlab api for an example)

How to do it

The necessary steps:

commitingJob:
  stage: some_stage
  script:
    # in order to commit we have to have a user set
    # this would also make it easy to distinct the CI-made commits
    - git config user.name "CI Pipeline"
    - git config user.email "cipipeline@example.com"
    # stage some changes
    - git add src/*
    # We can use `-o ci.skip`, but AFAIK it doesn't work for Merge Request pipelines, while having it inside the commit message works there as well
    - git commit -m "A commit message [skip ci]"
    # we're on a detached head but we can push the commits we made to the remote branch like so:
    - git push HEAD:$CI_COMMIT_REF_NAME

One thing missing here is authenticating with the remote. It can be setup in a number of ways (ssh, token, api). Other answers discuss a few different options. My preference would be to use an ACCESS_TOKEN, see how and why bellow

Summary and critique on the available authentication options

Adding or copying an ssh key inside the CI steps

Compared to just using an access token, going for the ssh way seems like too much work, for the same result

• generate an ssh key
• save the public key as Project deploy token
• save the private key as a Project CI variable
• dedicate a few CI steps on
  • adding the private keys
  • adding known hosts
  • add/update the ssh remote

Overall it seems like something you'd do for a user and not for a pipeline

You have to add the private key as a Project Level CI variable - due to pattern restrictions the variable can't be masked. It can be added as a file, but then you'd have to chmod for proper permissions...

The positive thing I see here is the ssh key is restricted only to the projects you add it to as deploy token

---

The easiest option seems to be to use an ACCESS_TOKEN

Using a token to push commits is as easy as updating the push remote like this:

git remote set-url --push origin "https://$TOKEN_NAME:$ACCESS_TOKEN@gitlab.com/<project>.git"

TOKEN_NAME - the name you set for the token in gitlab (preferably pick a name without space characters) ACCESS_TOKEN - this can be a personal access token (Free Tier) or a project access token (Premium Tier)

It would have to be stored as Project Level CI variable, and it can be masked. I would just save a variable like CI_COMMITTER_USER_AND_TOKEN containing both <token_name>:<access_token> content

⚠ Warning about Personal Access Tokens

This is too broad permission to give to a CI Runner even if the runner should commit to multiple repositories, I'll prefer separate access tokens per project with only the minimum permissions needed

Keep in mind that if someone takes hold of your personal access token, they can do stuff (like committing in this case) on your behalf. Giving the write_repository permission means you can use the token to write to any repository you're account has access too.

CI_JOB_TOKEN

If you're wondering whether you can use the CI_JOB_TOKEN to push commits - you can't. You can only pull with that thing
(You can also use it to publish packages and images, but not push commits)

Why git remote set-url --push origin

We can add a separate remote for CI made commits, but keep in mind it might get cached, so if we just do

script: 
  - git remote add ci "https://$TOKEN_NAME:$ACCESS_TOKEN@gitlab.com/<project>.git"

the next time the pipeline runs we'll get an error from git remote add, because the remote we're trying to add is already added

set-url on the existing origin makes sure you use the latest url

To achieve the same with git remote add we'd have to first try to remove the ci remote and then add it back

script: 
  - git remote remove ci || true
  - git remote add ci "https://$TOKEN_NAME:$ACCESS_TOKEN@gitlab.com/<project>.git"

---

Use the gitlab api to push changes

This works pretty much the same way as the ACCESS_TOKEN way, but instead of the write_repository the token has the api permission

If the reason to commit and push as part of CI is releasing a new version on merge to main then using CLI tools and a guide like this might be a better option than setting up the committing and pushing yourself: https://docs.gitlab.com/ee/ci/examples/semantic-release.html

The above guide setups a version bump, package updates, and release notes. Note how this step instructs us to create an Access Token with api permission in order for the underlying CLIs to commit the changes: https://docs.gitlab.com/ee/ci/examples/semantic-release.html#set-up-cicd-variables

Answer 6

Nowadays there is a much cleaner way to solve this without using SSH but using a project scoped access token, also see this answer.

In the GitLab project create an project scoped access token so it is linked to the project, not to an individual. Next store this token as an GitLab CI/CD variable. You can now connect using the following:

push-back-to-remote:
  script:
    - git config user.email "my-email@email.com"
    - git config user.name "ci-bot"
    - git remote add gitlab_origin https://oauth2:$ACCESS_TOKEN@gitlab.com/path-to-project.git
    - git add .
    - git commit -m "push back from pipeline"
    - git push gitlab_origin HEAD:main -o ci.skip # prevent triggering pipeline again

Answer 7

You can add the CI_SERVER_CLS_CA_FILE to sslCAInfo git config.

checkout alchemy:
    stage: prepare
    script:
        - git config --global "http.${CI_SERVER_URL}.sslCAInfo" "$CI_SERVER_TLS_CA_FILE"
        - git clone https://gitlab-ci-token:${CI_JOB_TOKEN}@${CI_SERVER_HOST}/sparklemuffin/alchemy.git

@Sjoerd's approach, in the comments, to export GIT_SSL_CAINFO instead, is a bit shorter.

export GIT_SSL_CAINFO=$CI_SERVER_TLS_CA_FILE

Gitlab creates CI_SERVER_TLS_CA_FILE and configures git to use it for initially cloning the repository. For some reason this configuration is no longer available later on.

Answer 8

Another way:

1. Create dedicated Gitlab account
2. Add CI/CD variable GIT_CICDUSER_PASSWORD containing it's password to the project inside the main account running the pipeline

Then a job can look like:

task_requiring_a_push:
  stage: some-stage
  variables:
    GIT_STRATEGY: none
  script:
    - git config --global user.email "user@mail.com"
    - git config --global user.name "CI/CD User"
    - git clone https://<dedicated_gitlab_user_username>:$GIT_CICDUSER_PASSWORD@gitlab.com/$CI_PROJECT_PATH .
    - git checkout $CI_COMMIT_REF_NAME
    - # do something
    - git add *
    - git commit -m "message"
    - git push --push-option=ci.skip origin $CI_COMMIT_REF_NAME
only:
    - release-*

Answer 9

This is a working example as of today environment: k8s gitlab: 13.x gitlab runner: 13.x goal of this job named convert:- Converts the excel files to json files and commits and updates the branch of the repo.

convert:
  variables:
    REPO: "gitlab.com/group/myproject.git" # example
    BRANCH: "BRANCHNAME" # example
    # recommended to store the following as project variables to hide secrets away from the gitlab ci file.
    GITLAB_USER_ID: "gitlab_user" # example
    CI_USERNAME: "gitlab_user" # example
    CI_PUSH_TOKEN: "<api token from gitlab" # example
    GITLAB_USER_EMAIL: "gitlab_user@company.com" # example
  stage: convert
  image:
    name: python:3.7-buster
    entrypoint: ["/bin/ash"]
  before_script:
    - pip3 install openpyxl
    - ls -altr
  script:
    - echo 'converting excel to json'
    - python excel2json.py
    - git remote set-url origin https://${CI_USERNAME}:${CI_PUSH_TOKEN}@$REPO
    - git config --global user.email '${GITLAB_USER_EMAIL}'
    - git config --global user.name '${GITLAB_USER_ID}'
    - git add -A && git commit -m 'added/updated json files'
    - git push origin HEAD:$BRANCH

Note: CI_USERNAME == GITLAB_USER_ID. Both are same in my case.

Answer 10

Instead of declaring CI_KNOWN_HOST, you can try get it realtime:

- 'which ssh-agent || ( apt-get update -y && apt-get install openssh-client -y )'
- ssh-keyscan -t rsa $CI_SERVER_HOST >> ~/.ssh/known_hosts

Answer 11

I found this GitLab forum link helpful As suggested by the user you need to generate SSH key, associate it with new GitLab user dedicated for this job and add key to the runner. Small drawback is you need to use swap origin in gitlab for original ssh source (instead of sandboxed one used inside the job) which leads to committer being changed to mentioned new account instead of person who triggered pipeline. Source from link:

# for your information
whoami
printenv

# we need to extract the ssh/git URL as the runner uses a tokenized URL
export CI_PUSH_REPO=`echo $CI_REPOSITORY_URL | perl -pe 's#.*@(.+?(\:\d+)?)/#git@\1:#'`

# runner runs on a detached HEAD, create a temporary local branch for editing
git checkout -b ci_processing
git config --global user.name "My Runner"
git config --global user.email "runner@gitlab.example.org"
git remote set-url --push origin "${CI_PUSH_REPO}"

# make your changes
touch test.txt

# push changes
# always return true so that the build does not fail if there are no changes
git push origin ci_processing:${CI_COMMIT_REF_NAME} || true

Just with current version of GitLab you need to change source variable name as follows:

export CI_PUSH_REPO=`echo $CI_REPOSITORY_URL | perl -pe 's#.*@(.+?(\:\d+)?)/#git@\1:#'`

Answer 12

In my case deploy keys option was optimal (compared to personal tokens or CI token - only supports basic auth) with Gitlab Shell Runner. In case someone is struggling with pushing to from Gitlab CI, this can be done sharing the public key of the runner with the Gitlab server

Answer 13

Solved it. Issuing git config --global http.sslverify "false" prior to the push solved that particular problem (it exposed another problem but that's for another thread :) )