Migrating from Jenkins
A lot of GitLab users have successfully migrated to GitLab CI/CD from Jenkins. To make this easier if you're just getting started, we've collected several resources here that you might find useful before diving in. Think of this page as a "GitLab CI/CD for Jenkins Users" guide.
First of all, our Quick Start Guide contains a good overview of how GitLab CI/CD works. You may also be interested in Auto DevOps which can potentially be used to build, test, and deploy your applications with little to no configuration needed at all.
For advanced CI/CD teams, templates can enable the reuse of pipeline configurations.
Otherwise, read on for important information that will help you get the ball rolling. Welcome to GitLab!
If you have questions that are not answered here, the GitLab community forum can be a great resource.
Managing the organizational transition
An important part of transitioning from Jenkins to GitLab is the cultural and organizational changes that comes with the move, and successfully managing them. There are a few things we have found that helps this:
- Setting and communicating a clear vision of what your migration goals are helps your users understand why the effort is worth it. The value will be clear when the work is done, but people need to be aware while it's in progress too.
- Sponsorship and alignment from the relevant leadership team helps with the point above.
- Spending time educating your users on what's different, sharing this document with them, and so on will help ensure you are successful.
- Finding ways to sequence or delay parts of the migration can help a lot, but you don't want to leave things in a non-migrated (or partially-migrated) state for too long. To gain all the benefits of GitLab, moving your existing Jenkins setup over as-is, including any current problems, will not be enough. You need to take advantage of the improvements that GitLab offers, and this requires (eventually) updating your implementation as part of the transition.
Important product differences
There are some high level differences between the products worth mentioning:
With GitLab you don't need a root
pipelinekeyword to wrap everything.
The way pipelines are triggered and trigger other pipelines is different than Jenkins. GitLab pipelines can be triggered:
You can control which jobs run in which cases, depending on how they are triggered, with the
GitLab pipeline scheduling concepts are also different than with Jenkins.
You can reuse pipeline configurations using the
includekeyword and templates. Your templates can be kept in a central repo (with different permissions), and then any project can use them. This central project could also contain scripts or other reusable code.
You can also use the
extendskeyword to reuse configuration within a single pipeline configuration.
All jobs within a single stage always run in parallel, and all stages run in sequence. We are planning to allow certain jobs to break this sequencing as needed with our directed acyclic graph feature.
parallelkeyword can automatically parallelize tasks, like tests that support parallelization.
Normally all jobs within a single stage run in parallel, and all stages run in sequence. There are different pipeline architectures that allow you to change this behavior.
rulessyntax is the recommended method of controlling when different jobs run. It is more powerful than the
One important difference is that jobs run independently of each other and have a fresh environment in each job. Passing artifacts between jobs is controlled using the
dependencieskeywords. When finished, the planned Workspaces feature will allow you to more easily persist a common workspace between serial jobs.
.gitlab-ci.ymlfile is checked in to the root of your repository, much like a Jenkinsfile, but is in the YAML format (see complete reference) instead of a Groovy DSL. It's most analogous to the declarative Jenkinsfile format.
GitLab comes with a container registry, and we recommend using container images to set up your build environment. For example, set up one pipeline that builds your build environment itself and publish that to the container registry. Then, have your pipelines use this instead of each building their own environment, which will be slower and may be less consistent. We have extensive docs on how to use the Container Registry.
A central utilities repository can be a great place to put assorted scheduled jobs or other manual jobs that function like utilities. Jenkins installations tend to have a few of these.
Agents vs. Runners
Both Jenkins agents and GitLab Runners are the hosts that run jobs. To convert the Jenkins agent, simply uninstall it and then install and register the runner. Runners do not require much overhead, so you can size them similarly to the Jenkins agents you were using.
There are some important differences in the way Runners work in comparison to agents:
- Runners can be set up as shared across an instance, be added at the group level, or set up at the project level. They will self-select jobs from the scopes you've defined automatically.
- You can also use tags for finer control, and associate runners with specific jobs. For example, you can use a tag for jobs that require dedicated, more powerful, or specific hardware.
- GitLab has autoscaling for Runners which will let configure them to be provisioned as needed, and scaled down when not. This is similar to ephemeral agents in Jenkins.
If you are using
gitlab.com, you can take advantage of our shared Runner fleet
to run jobs without provisioning your own Runners. We are investigating making them
available for self-managed instances
Groovy vs. YAML
Jenkins Pipelines are based on Groovy, so the pipeline specification is written as code.
GitLab works a bit differently, we use the more highly structured YAML format, which
places scripting elements inside of
script: blocks separate from the pipeline specification itself.
This is a strength of GitLab, in that it helps keep the learning curve much simpler to get up and running and avoids some of the problem of unconstrained complexity which can make your Jenkinsfiles hard to understand and manage.
That said, we do of course still value DRY (don't repeat yourself) principles and want to ensure that
behaviors of your jobs can be codified once and applied as needed. You can use the
extends: syntax to
reuse configuration in your jobs, and
be used to reuse pipeline configurations in pipelines
in different projects:
.in-docker: tags: - docker image: alpine rspec: extends: - .in-docker script: - rake rspec
Artifacts may work a bit differently than you've used them with Jenkins. In GitLab, any job can define
a set of artifacts to be saved by using the
artifacts: keyword. This can be configured to point to a file
or set of files that can then be persisted from job to job. Read more on our detailed
pdf: script: xelatex mycv.tex artifacts: paths: - ./mycv.pdf - ./output/ expire_in: 1 week
Additionally, we have package management features like a built-in container, NPM, and Maven registry that you can leverage. You can see the complete list of packaging features (which includes links to documentation) in the Packaging section of our documentation.
Where you may have used plugins to get things like code quality, unit tests, security scanning, and so on working in Jenkins, GitLab takes advantage of our connected ecosystem to automatically pull these kinds of results into your Merge Requests, pipeline details pages, and other locations. You may find that you actually don't need to configure anything to have these appear.
If they aren't working as expected, or if you'd like to see what's available, our CI feature index has the full list of bundled features and links to the documentation for each.
For advanced CI/CD teams, project templates can enable the reuse of pipeline configurations, as well as encourage inner sourcing.
In self-managed GitLab instances, you can build an Instance Template Repository. Development teams across the whole organization can select templates from a dropdown menu. A group administrator is able to set a group to use as the source for the custom project templates, which can be used by all projects in the group. An instance adminsitrator can set a group as the source for instance project templates, which can be used by projects in that instance.
Converting Declarative Jenkinsfiles
Declarative Jenkinsfiles contain "Sections" and "Directives" which are used to control the behavior of your pipelines. There are equivalents for all of these in GitLab, which we've documented below.
This section is based on the Jenkinsfile syntax documentation and is meant to be a mapping of concepts there to concepts in GitLab.
The agent section is used to define how a pipeline will be executed. For GitLab, we use the GitLab Runner to provide this capability. You can configure your own runners in Kubernetes or on any host, or take advantage of our shared runner fleet (note that the shared runner fleet is only available for GitLab.com users.) The link above will bring you to the documentation which will describe how to get up and running quickly. We also support using tags to direct different jobs to different Runners (execution agents).
agent section also allows you to define which Docker images should be used for execution, for which we use
image keyword. The
image can be set on a single job or at the top level, in which
case it will apply to all jobs in the pipeline:
my_job: image: alpine ...
post section defines the actions that should be performed at the end of the pipeline. GitLab also supports
this through the use of stages. You can define your stages as follows, and any jobs assigned to the
after_pipeline stages will run as expected. You can call these stages anything you like:
stages: - before_pipeline - build - test - deploy - after_pipeline
Setting a step to be performed before and after any job can be done via the
default: before_script: - echo "I run before any jobs starts in the entire pipeline, and can be responsible for setting up the environment."
GitLab CI also lets you define stages, but is a little bit more free-form to configure. The GitLab
is a top level setting that enumerates the list of stages, but you are not required to nest individual jobs underneath
stages section. Any job defined in the
.gitlab-ci.yml can be made a part of any stage through use of the
Note that, unless otherwise specified, every pipeline is instantiated with a
which are run in that order. Jobs that have no
stage defined are placed by default in the
Of course, each job that refers to a stage must refer to a stage that exists in the pipeline configuration.
stages: - build - test - deploy my_job: stage: build ...
steps section is equivalent to the
script section of an individual job. This is
a simple YAML array with each line representing an individual command to be run:
my_job: script: - echo "hello! the current time is:" - time ...
In GitLab, we use the
variables keyword to define different variables at runtime.
These can also be set up through the GitLab UI, under CI/CD settings. See also our general documentation on variables,
including the section on protected variables which can be used
to limit access to certain variables to certain environments or runners:
variables: POSTGRES_USER: user POSTGRES_PASSWORD: testing_password
Here, options for different things exist associated with the object in question itself. For example, options related to jobs are defined in relation to the job itself. If you're looking for a certain option, you should be able to find where it's located by searching our complete configuration reference page.
GitLab does not require you to define which variables you want to be available when starting a manual job. A user can provide any variables they like.
Because GitLab is integrated tightly with Git, SCM polling options for triggers are not needed. We support an easy to use syntax for scheduling pipelines.
GitLab does not support a separate
tools directive. Our best-practice recommendation is to use pre-built
container images, which can be cached, and can be built to already contain the tools you need for your pipelines. Pipelines can
be set up to automatically build these images as needed and deploy them to the container registry.
If you're not using container images with Docker/Kubernetes, for example on Mac or FreeBSD, then the
shell executor does require you to
set up your environment either in advance or as part of the jobs. You could create a
action that handles this for you.
Similar to the
parameters keyword, this is not needed because a manual job can always be provided runtime
GitLab does support a
when keyword which is used to indicate when a job should be
run in case of (or despite) failure, but most of the logic for controlling pipelines can be found in
our very powerful
only/except rules system
(see also our advanced syntax):
my_job: only: [branches]