As developers, we all know that it is a good idea to reuse code as much as possible. We all know the DRY mantra - Don't Repeat Yourself. Functions, classes or web components abstract away logic, parameterize data, allow code to be reusable, maintainable and extendable.
With GitLab CI/CD, pipelines are defined in YAML. It's a human-readable data- serialization language (Source), so its focus is clear and concise data delivery rather than efficiency optimization. It's often not obvious how we can go about reusing code in YAML.
Today, we're going to learn how to create CI template library - a library comprised of reusable job templates that can be shared, extended, and overridden by multiple projects 👌
If you're a developer working on a side project on GitLab, or if your team is relatively small and your projects are diverse in programming languages and build processes, then it's probably good enough to stick to a single
On the other hand, as a company grows, it often has standardized testing, building, and deploying processes that are applied to most internal projects. A CI template library increases time efficiency in pipeline development and decreases update and maintenance effort across the ecosystem.
To build a template library, let's first dive into the basic component of it - the job template ⬇️
job is a basic building block of a pipeline. It usually has a single purpose, executed in isolation and, most of the time, independent of other jobs.
For example, let's say we have an awesome Node.js app called
awesome-node-app and we need a job to install dependencies before building it. The
install job would look something like this:
- node_modules/ # cache node_modules/ in subsequent pipeline
# install dependencies in CI mode
- npm ci
Now, a job template is essentially a job, but it has the following extra properties:
- Generic: It is project-agnostic which means it does not contain any data that pertains to a specific project.
- Importable: It is easily imported and used directly in a project.
- Customizable: It can be extended or overridden.
If we examine the
install job above once again, it seems like we could, and should, transform it into a template. We might want to reuse it in another JS application in the future!
It's already generic enough, but not quite importable or customizable. Let's change that.
To create a template, all we need to do is move the job to a new file,
- npm ci
Then, we need to figure a way to "import" and use this job in
.gitlab-ci.yml file. Luckily, GitLab has a pretty sweet keyword
include that allows us to do exactly that!
includeallows us to include and use content declared in an external
yamlfile - either locally or remotely.
We already created
install.yml locally, so let's
include it at the top of our
.gitlab-ci.yml, like so:
- local: 'install.yml' # path to `install.yml`
The pipeline now has the job named
install that does the same thing:
Using local file here probably won't make sense since we might want to reuse the template in another project and we do not want to just copy and paste the job definition. Remember, keep it DRY.
So let's go ahead and make a new template library that only stores templates! We can then refer to this library whenever we need to
Create a new repository
ci-templates that's within the same group as
awesome-node-app. Then, add
install.yml at the root of the project. By now, you would have this:
Then, commit and push to
master. Cool! Now your library is up and accessible to other repositories.
Let's go back to
install.yml again, this time using
# group name is your username if the project is under personal account
- project: '<my-group>/ci-templates'
💡We can modify
refto point to any other branch, commit SHA, or version tag of the file in Git history as we'd like. It's good practice to keep track of version history for your template file.
Great! We just basically told GitLab to "include this file
ci-templates repo on
master branch into the pipeline". We now have
install job imported from
ci-templates to our
We can run the pipeline as-is -
install is activated automatically without any further configuration.
However, what if we do want to change or add configuration?
Scenario: Right now,
install only looks for dependencies declared in the
package.json file at the root of a project. What if we have
another-awesome-node-app that is a monorepo, and we want to run
install multiple times in various locations?
We need to parameterize our
install template to take in some sort of data that holds information about the location of the
package.json file we're looking for.
The most powerful way to parameterize a template is by using environment variables.
Environment variables come in two flavours:
- Predefined environment variables: Variables provided by GitLab out of the box and ready to use without any specification.
💡They are references to branch names, merge request IDs, jobs info, and much, much more.
Predefined environment variables are incredibly powerful. We can do things like conditionally skipping a job in a pipeline, allowing jobs to run on certain branches, leveraging custom variables, and so on.
This topic deserves a separate article of its own, so if you're interested in knowing more about their use cases and real-world implementation, let me know in the comments below 💚
Custom environment variables: Variables defined in
.gitlab-ci.yml(you can also define them in GitLab UI and via the API).
⚠️ Make sure to avoid name collision with predefined variables when naming your variable.
Custom environment variables work in great harmony with job templates. The syntax is as follow:
# declare a key/value pair
... # declare as many variables as you want
# call its value, this outputs "hello" in the runner
- echo $MY_VARIABLE
Notice that the variable is declared within the job scope. This means that the variable is only accessible within the job and inaccessible from pipeline level.
Going back to our example, let's create a new custom
INSTALL_DIRECTORY and call it in our
INSTALL_DIRECTORY: '.' # default to root directory
# cd to the directory of package.json
- cd $INSTALL_DIRECTORY
# install dependencies in CI mode
- npm ci
One more thing before we move on, let's make the job hidden by default by changing the job name from
.install. I'll explain how this works in just a bit.
Cool! Now we're ready to use this template in
another-awesome-node-app. Let's include the template again:
- project: '<my-group>/ci-templates'
We've just included
.install, but this time, it's hidden, which means it's disabled by default. If you try running this pipeline in GitLab, it will not run simply because the pipeline is empty - there is no job!
So how do we use our template then?
Turns out, we can create a new job that
extends our template to inherit its configuration.
Let's create two jobs
install_project_two that extend
.install. After that, we also need to change the default value of
INSTALL_DIRECTORY in each job to the expected path:
Awesome! Now both
install_project_two inherit the script from
.install, but they find the
package.json file in two different locations just like we wanted!
⚠️ Had we not specified
installa hidden job earlier, we would have had one extra
installjob declared in our pipeline that runs in the root directory - where there is no
package.json. This will fail the pipeline.
📝 To prevent side effects from including external jobs, it's good practice to declare all template jobs hidden and
extend them when needed.
We can also keep any other reusable snippets of configuration in the template library. I'd like to call them
Some mixin examples:
- Bash scripts
- Pipeline configuration partials
Here's one good example of what I meant by pipeline configuration partial:
- docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY
- docker logout $CI_REGISTRY
This mixin logs the user into GitLab Registry
before_script and logs them out
after_script. However, it does not have
script declared, which means that it cannot be run as a
scriptis required for job definition.
To use the mixin, all we need to do is
extend it the same way we do with regular templates and, most importantly, define what's in
script in the new job:
- docker build $MY_APP_IMAGE
- docker push $MY_APP_IMAGE
Here's what we've learned today:
- Properties of a job template: generic, importable and customizable
- Use environment variables to parameterize templates
- Hide a job to prevent side effects
- Use mixins to further simplify the pipeline
I'm sure there are many other techniques in building a CI template not listed here. Please let me know in the comments what you think of my approach and any suggestions/recommendations for further optimization!
I hope you enjoyed this article 💚
Cover photo by Pankaj Patel on Unsplash