We wrote about why we're replacing GitLab CI jobs with a .gitlab-ci.yml file.
As we've started on implementing this large change, we wanted to share the details
of that process with you and would love to hear what you think.
To recap the previous article:
currently you are required to write out your CI jobs in GitLab CI's interface.
We're replacing this with a single file .gitlab-ci.yml, that you place in the root
of your repository.
Schema change
Currently, on a push to GitLab, GitLab sends a web-hook to the CI Coordinator. The coordinator creates a build based on the jobs that are defined in its UI, which can then be executed by the connected Runners.
In the new schema, GitLab sends the web-hook and the .gitlab-ci.yml contents
to the CI Coordinator, which creates builds based on the yml file. In turn,
these builds are executed by the Runners as before.
Migrating to new style
Keeping two different ways of doing things would be a strain on development and support, not to mention confusing. So we're not just deprecating the old style of defining jobs, we're removing it entirely and will migrate existing jobs.
Upon upgrading your existing jobs defined in the GitLab CI Coordinator will be converted into a YAML file with the new syntax. You can download this file at any time from the project settings.
When the GitLab webhook triggers and doesn't transmit the content from .gitlab-ci.yml,
the coordinator will use the converted YAML file instead.
This makes migrating to the new style very easy. You can start by simply copy-pasting the contents of the converted YAML file to the root of your repository. Existing projects will continue to build successfully, yet new projects do not have the option to use anything else.
An example .gitlab-ci.yml
To get an idea of how the .gitlab-ci.yml will look, we've prepared an example
for a Ruby on Rails project (such as GitLab itself). Of course, this is due to
change as we're still working on this.
# Refs to skip
skip_refs: “deploy*”
# Run before each script
# Refs to skip
skip_refs: “deploy*”
# Run before each script
before_script:
- export PATH=$HOME/bin:/usr/local/bin:/usr/bin:/bin
- gem install bundler
- cp config/database.yml.mysql config/database.yml
- cp config/gitlab.yml.example config/gitlab.yml
- touch log/application.log
- touch log/test.log
- bundle install --without postgres production --jobs $(nproc)
- “bundle exec rake db:create RAILS_ENV=test”
# Parallel jobs, each line is a parallel build
jobs:
- script: “rake spec”
runner: “ruby,postgres”
name: “Rspec”
- script: “rake spinach”
runner: “ruby,mysql”
name: “Spinach”
tags: true
branches: false
# Parallel deploy jobs
on_success:
- “cap deploy production”
- “cap deploy staging”
UPDATE
Dmitriy and Sytse spend some time thinking about file syntax. Scripting should be simple and memorable. Thats why we come with better proposal:
before_script:
- gem install bundler
- bundle install
- bundle exec rake db:create
rspec:
test: "rake spec"
tags:
- ruby
- postgres
only:
- branches
spinach:
test: "rake spinach"
tags:
- ruby
- mysql
except:
- tags
staging:
deploy: "cap deploy stating"
tags:
- capistrano
- debian
except:
- stable
production:
deploy:
- cap deploy production
- cap notify
tags:
- capistrano
- debian
only:
- master
- /^deploy-.*$/
Contribute
GitLab is nothing without its community. Contribute or follow the development in the GitLab CI repository.
