debian-mirror-gitlab/doc/architecture/blueprints/ci_pipeline_components/index.md

status	creation-date	authors	coach	approvers	owning-stage	participating-stages
ongoing	2022-09-14	@ayufan @fabiopitino @grzesiek	@ayufan @grzesiek	@dhershkovitch @marknuzzo	~devops::verify

CI/CD Catalog

Summary

Goals

The goal of the CI/CD pipeline components catalog is to make the reusing pipeline configurations easier and more efficient. Providing a way to discover, understand and learn how to reuse pipeline constructs allows for a more streamlined experience. Having a CI/CD pipeline components catalog also sets a framework for users to collaborate on pipeline constructs so that they can be evolved and improved over time.

This blueprint defines the architectural guidelines on how to build a CI/CD catalog of pipeline components. This blueprint also defines the long-term direction for iterations and improvements to the solution.

Challenges

• GitLab CI/CD can have a steep learning curve for new users. Users must read the documentation and YAML reference to understand how to configure their pipelines.
• Developers are struggling to reuse existing CI/CD templates with the result of having to reinvent the wheel and write YAML configurations repeatedly.
• GitLab CI templates provide users with scaffolding pipeline or jobs for specific purposes. However versioning them is challenging today due to being shipped with the GitLab instance. See this issue for more information.
• Users of GitLab CI/CD (pipeline authors) today have their own ad-hoc way to organize shared pipeline configurations inside their organization. Those configurations tend to be mostly undocumented.
• The only discoverable configurations are GitLab CI templates. However they don't have any inline documentation so it becomes harder to know what they do and how to use them without copy-pasting the content in the editor and read the actual YAML.
• It's harder to adopt additional GitLab features (CD, security, test, etc.).
• There is no framework for testing reusable CI configurations. Many configurations are not unit tested against single changes.
• Communities, partners, 3rd parties, individual contributors, must go through the GitLab Contribution process to contribute to GitLab managed templates. See this issue for more information.
• GitLab has more than 100 of templates with some of them barely maintained after their addition.

Problems with GitLab CI templates

• GitLab CI Templates have not been designed with deterministic behavior in mind.
• GitLab CI Templates have not been design with reusability in mind.
• Jobs/ templates hard-code the stage: attribute but the user of the template must somehow override or know in advance what stage is needed.
  • The user should be able to import the job inside a given stage or pass the stage names as input parameter when using the component.
  • Failures in mapping the correct stage can result in confusing errors.
• Some templates are designed to work with AutoDevOps but are not generic enough (example).
• Many CI templates, especially those language specific are tutorial/scaffolding-style templates.
  • They are meant to show the user how a typical pipeline would look like but it requires high customization from the user perspective.
  • They require a different UX: copy-paste in the position of the Pipeline Editor cursor.
• Some templates like SAST.latest.gitlab-ci.yml add multiple jobs conditionally to the same pipeline.
  • Ideally these jobs could run as a child pipeline and make the reports available to the parent pipeline.
  • This epic is necessary for Parent-child pipelines to be used.
• Some templates incorrectly use variables, image and other top-level keywords but that defines them in all pipeline jobs, not just those defined in the template.
  • This technique introduces inheritance issues when a template modifies jobs unnecessarily.

Opportunities

• Having a catalog of pipeline constructs where users can search and find what they need can greatly lower the bar for new users.
• Customers are already trying to rollout their ad-hoc catalog of shared configurations. We could provide a standardized way to write, package and share pipeline constructs directly in the product.
• As we implement new pipeline constructs (for example, reusable job steps) they could be items of the catalog. The catalog can boost the adoption of new constructs.
• The catalog can be a place where we strengthen our relationship with partners, having components offered and maintained by our partners.
• With discoverability and better versioning mechanism we can have more improvements and better collaboration.
• Competitive landscape is showing the need for such feature
  • R2DevOps implements a catalog of CI templates for GitLab pipelines.
  • GitHub Actions provides an extensive catalog of reusable job steps.
  • CircleCI Orbs provide reusable YAML configuration packages.

Glossary

This section defines some terms that are used throughout this document. With these terms we are only identifying abstract concepts and are subject to changes as we refine the design by discovering new insights.

• Component Is the reusable unit of pipeline configuration.
• Components repository represents a collection of CI components stored in the same project.
• Project is the GitLab project attached to a single components repository.
• Catalog is a collection of resources like components repositories.
• Catalog resource is the single item displayed in the catalog. A components repository is a catalog resource.
• Version is a specific revision of catalog resource. It maps to the released tag in the project, which allows components to be pinned to a specific revision.
• Steps is a collection of instructions for how jobs can be executed.

Definition of pipeline component

A pipeline component is a reusable single-purpose building block that abstracts away a single pipeline configuration unit. Components are used to compose a part or entire pipeline configuration. It can optionally take input parameters and set output data to be adaptable and reusable in different pipeline contexts, while encapsulating and isolating implementation details.

Components allow a pipeline to be assembled by using abstractions instead of having all the details defined in one place. When using a component in a pipeline, a user shouldn't need to know the implementation details of the component and should only rely on the provided interface.

A pipeline component defines its type which indicates in which context of the pipeline configuration the component can be used. For example, a component of type X can only be used according to the type X use-case.

For best experience with any systems made of components it's fundamental that components:

• Single purpose: a component must focus on a single goal and the scope be as small as possible.
• Isolated: when a component is used in a pipeline, its implementation details should not leak outside the component itself and into the main pipeline.
• Reusable: a component is designed to be used in different pipelines. Depending on the assumptions it's built on a component can be more or less generic. Generic components are more reusable but may require more customization.
• Versioned: when using a component we must specify the version we are interested in. The version identifies the exact interface and behavior of the component.
• Resolvable: when a component depends on another component, this dependency must be explicit and trackable.

Predictable components

Eventually, we want to make CI Catalog Components predictable. Including a component by its path, using a fixed @ version, should always return the same configuration, regardless of a context from which it is getting included from. The resulting configuration should be the same for a given component version and the set of inputs passed using include:inputs keyword, therefore it should be deterministic.

A component should not produce side effects by being included and should be referentially transparent.

Making components predictable is a process, and we may not be able to achieve this without significantly redesigning CI templates, what could be disruptive for users and customers right now. We initially considered restricting some top-level keywords, like include: remote: to make components more deterministic, but eventually agreed that we first need to iterate on the MVP to better understand the design that is required to make components more predictable. The predictability, determinism, referential transparency and making CI components predictable is still important for us, but we may be unable to achieve it early iterations.

Structure of a component

A pipeline component is identified by a unique address in the form <fqdn>/<component-path>@<version> containing:

• FQDN (Fully Qualified Domain Name).
• The path to a repository or directory that defines it.
• A specific version

For example: gitlab.com/gitlab-org/dast@1.0.

The FQDN

Initially we support only component addresses that point to the same GitLab instance, meaning that the FQDN matches the GitLab host.

The component path

The directory identified by the component path must contain at least the component YAML and optionally a related README.md documentation file.

The component path can be:

• A path to a project: gitlab.com/gitlab-org/dast. The default component is processed.
• A path to an explicit component: gitlab.com/gitlab-org/dast/api-scan. In this case the explicit api-scan component is processed.
• A relative path to a local directory: ./path/to/component. This path must contain the component YAML that defines the component. The path must start with ./ or ../ to indicate a path relative to the current file's path.

Relative local paths are a abbreviated form of the full component address, meaning that ./path/to/component called from a file mydir/file.yml in gitlab-org/dast project would be expanded to:

gitlab.com/gitlab-org/dast/mydir/path/to/component@<CURRENT_SHA>

The component YAML file follows the filename convention <type>.yml where component type is one of:

Component type	Context
template	For components used under include: keyword

Based on the context where the component is used we fetch the correct YAML file. For example:

• if we are including a component gitlab.com/gitlab-org/dast@1.0 we expect a YAML file named template.yml in the root directory of gitlab-org/dast repository.
• if we are including a component gitlab.com/gitlab-org/dast/api-scan@1.0 we expect a YAML file named template.yml inside a directory api-scan of gitlab-org/dast repository.

A component YAML file:

• Must have a name to be referenced to.
• Must specify its type in the filename, which defines how it can be used (raw configuration to be included, child pipeline workflow, job step).
• Must define its content based on the type.
• Must specify input parameters that it accepts. Components should depend on input parameters for dynamic values and not environment variables.
• Can optionally define output data that it returns.
• Should be validated statically (for example: using JSON schema validators).

spec:
  inputs:
    website:
    environment:
      default: test
    test_run:
      options:
        - unit
        - integration
        - system
---
# content of the component

The component version

The version of the component can be (in order of highest priority first):

1. A commit SHA - For example: gitlab.com/gitlab-org/dast@e3262fdd0914fa823210cdb79a8c421e2cef79d8
2. A tag - For example: gitlab.com/gitlab-org/dast@1.0
3. A special moving target version that points to the most recent released tag. The target project must be explicitly marked as a catalog resource - For example: gitlab.com/gitlab-org/dast@~latest
4. A branch name - For example: gitlab.com/gitlab-org/dast@master

If a tag and branch exist with the same name, the tag takes precedence over the branch. Similarly, if a tag is named e3262fdd0914fa823210cdb79a8c421e2cef79d8, a commit SHA (if exists) takes precedence over the tag.

As we want to be able to reference any revisions (even those not released), a component must be defined in a Git repository.

When referencing a component by local path (for example ./path/to/component), its version is implicit and matches the commit SHA of the current pipeline context.

Components repository

A components repository is a GitLab project/repository that exclusively hosts one or more pipeline components.

A components repository can be a catalog resource. For a components repository it's highly recommended to set an appropriate avatar and project description to improve discoverability in the catalog.

Components repositories that are released in the catalog must have a README.md file at the root directory of the repository. The README.md represents the documentation of the components repository, hence it's recommended even when not listing the repository in the catalog.

Structure of a components repository

A components repository can host one or more components. The author can decide whether to define a single component per repository or include multiple cohesive components in the same repository.

A components repository is identified by the project full path.

Let's imagine we are developing a component that runs RSpec tests for a Rails app. We create a project called myorg/rails-rspec.

The following directory structure would support 1 component per repository:

.
├── template.yml
├── README.md
└── .gitlab-ci.yml

The .gitlab-ci.yml is recommended for the project to ensure changes are verified accordingly.

The component is now identified by the path gitlab.com/myorg/rails-rspec which also maps to the project path. We expect a template.yml file and README.md to be located in the root directory of the repository.

The following directory structure would support multiple components per repository:

.
├── .gitlab-ci.yml
├── README.md
├── unit/
│   └── template.yml
├── integration/
│   └── template.yml
└── feature/
    └── template.yml

In this example we are defining multiple test profiles that are executed with RSpec. The user could choose to use one or more of these.

Each of these components are identified by their path gitlab.com/myorg/rails-rspec/unit, gitlab.com/myorg/rails-rspec/integration and gitlab.com/myorg/rails-rspec/feature.

This directory structure could also support both strategies:

.
├── template.yml       # myorg/rails-rspec
├── README.md
├── LICENSE
├── .gitlab-ci.yml
├── unit/
│   └── template.yml   # myorg/rails-rspec/unit
├── integration/
│   └── template.yml   # myorg/rails-rspec/integration
└── feature/
    └── template.yml   # myorg/rails-rspec/feature

With the above structure we could have a top-level component that can be used as the default component. For example, myorg/rails-rspec could run all the test profiles together. However, more specific test profiles could be used separately (for example myorg/rails-rspec/integration).

NOTE: Nesting of components is not permitted. This limitation encourages cohesion at project level and keeps complexity low.

spec:inputs: parameters

If the component takes any input parameters they must be specified according to the following schema:

spec:
  inputs:
    website: # by default all declared inputs are mandatory.
    environment:
      default: test # apply default if not provided. This makes the input optional.
    flags:
      default: null # make an input entirely optional with no value by default.
    test_run:
      options: # a choice must be made from the list since there is no default value.
        - unit
        - integration
        - system
---
# content of the component
my-job:
  script: echo

The YAML in this case contains 2 documents. The first document represents the specifications while the second document represents the content.

When using the component we pass the input parameters as follows:

include:
  - component: gitlab.com/org/my-component@1.0
    inputs:
      website: ${MY_WEBSITE} # variables expansion
      test_run: system
      environment: $[[ inputs.environment ]] # interpolation of upstream inputs

Variables expansion must be supported for include:inputs syntax as well as interpolation of possible inputs provided upstream.

Input parameters are validated as soon as possible:

1. Read the file gitlab-template.yml inside org/my-component project.
2. Parse spec:inputs from the specifications and validate the parameters against this schema.
3. If successfully validated, proceed with parsing the content. Return an error otherwise.
4. Interpolate input parameters inside the component's content.

spec:
  inputs:
    environment:
      options: [test, staging, production]
---
"run-tests-$[[ inputs.environment ]]":
  script: ./run-test

scan-website:
  script: ./scan-website $[[ inputs.environment ]]
  rules:
    - if: $[[ inputs.environment ]] == 'staging'
    - if: $[[ inputs.environment ]] == 'production'

With $[[ inputs.XXX ]] inputs are interpolated immediately after parsing the content.

CI configuration interpolation perspectives and limitations

With spec:inputs users will be able to define input arguments for CI configuration. With include:inputs, they will pass these arguments to CI components.

inputs in $[[ inputs.something ]] is going to be an initial "object" or "container" that we will provide, to allow users to access their arguments in the interpolation block. This, however, can evolve into myriads of directions, for example:

1. We could provide variables or env object, for users to access their environment variables easier.
2. We can extend the block evaluation to easier navigate JSON or YAML objects passed from elsewhere.
3. We can provide access to the repository files, snippets or issues from there too.

The CI configuration interpolation is a relative compute-intensive technology, especially because we foresee this mechanism being used frequently on GitLab.com. In order to ensure that users are using this responsibly, we have introduced various limits, required to keep our production system safe. The limits should not impact users, because there are application limits available on a different level (maximum YAML size supported, timeout on parsing YAML files etc); the interpolation limits we've introduced are typically much higher then these. Some of them are:

1. An interpolation block should not be larger than 1 kilobyte.
2. A YAML value with interpolation in it can't be larger than 1 megabyte.
3. YAML configuration can't consist of more than half million entries.

Why input parameters and not environment variables?

Until today we have been leveraging environment variables to pass information around. For example, we use environment variables to pass information from an upstream pipeline to a downstream pipeline.

Using environment variables for passing information to a component is like declaring global variables in programming languages. The more variables we declare the more we risk variable conflicts and increase variables scope.

Input parameters are like variables passed to the component which exist inside a specific scope and they don't leak to the outside. Inputs are not inherited from upstream includes. They must be passed explicitly.

This paradigm allows to build more robust and isolated components as well as declare and enforce contracts.

Input parameters for existing include: syntax

Because we are adding input parameters to components used via include:component we have an opportunity to extend it to other include: types support inputs through inputs: syntax:

include:
  - component: gitlab.com/org/my-component@1.0
    inputs:
      foo: bar
  - local: path/to/file.yml
    inputs:
      foo: bar
  - project: org/another
    file: .gitlab-ci.yml
    inputs:
      foo: bar
  - remote: http://example.com/ci/config
    inputs:
      foo: bar
  - template: Auto-DevOps.gitlab-ci.yml
    inputs:
      foo: bar

Then the configuration being included must specify the inputs by defining a specification section in the YAML:

spec:
  inputs:
    foo:
---
# rest of the configuration

If a YAML includes content using include:inputs but the including YAML doesn't define spec:inputs in the specifications, an error should be raised.

include:inputs	spec:inputs	result
specified		raise error
specified	specified	validate inputs
	specified	use defaults
		legacy include: without input passing

Input parameters for pipelines

Inputs can also be used to pass parameters to a pipeline when triggered and benefit from immediate validation.

Today we have different use cases where using explicit input parameters would be beneficial:

1. Run Pipeline UI form.
   • Problem today: We are using top-level variables with variables:*:description to surface environment variables to the UI. The problem with this is the mix of responsibilities as well as the jump in precedence that a variable gets (from a YAML variable to a pipeline variable). Building validation and features on top of this solution is challenging and complex.
2. Trigger a pipeline via API. For example POST /projects/:id/pipelines/trigger with { inputs: { provider: 'aws' } }
3. Trigger a pipeline via trigger: syntax.

deploy-app:
  trigger:
    project: org/deployer
    inputs:
      provider: aws
      deploy_environment: staging

To solve the problem of Run Pipeline UI form we could fully leverage the inputs specifications:

spec:
  inputs:
    concurrency:
      default: 10    # displayed as default value in the input box
    provider: # can enforce `required` in the form validation
      description: Deployment provider # optional: render as input label.
    deploy_environment:
      options: # render a selectbox with options in order of how they are defined below
        - staging    # 1st option
        - canary     # 2nd option
        - production # 3rd option
      default: staging # selected by default in the UI.
                      # if `default:` is not specified, the user must explicitly select
                      # an option.
      description: Deployment environment # optional: render as input label.
---
# rest of the pipeline config

CI Catalog

The CI Catalog is an index of resources that users can leverage in CI/CD. It initially contains a list of components repositories that users can discover and use in their pipelines.

In the future, the Catalog could contain also other types of resources (for example: integrations, project templates, etc.).

To list a components repository in the Catalog we need to mark the project as being a catalog resource. We do that initially with an API endpoint, similar to changing a project setting.

Once a project is marked as a "catalog resource" it can be displayed in the Catalog.

We could create a database record when the API endpoint is used and remove the record when the same is disabled/removed.

Catalog resource

Upon publishing, a catalog resource should have at least following attributes:

• path: to be uniquely identified.
• name: for components repository this could be the project name.
• documentation: we would use the README.md file which would be mandatory.
• versions: one or more releases of the resource.

Other properties of a catalog resource:

• description: for components repository this could be the project description.
• avatar image: we could use the project avatar.
• indicators of popularity (stars, forks).
• categorization: user should select a category and or define search tags

As soon as a components repository is marked as being a "catalog resource" we should be seeing the resource listed in the Catalog.

Initially for the resource, the project may not have any released tags. Users would be able to use the components repository by specifying a branch name or commit SHA for the version. However, these types of version qualifiers should not be listed in the catalog resource's page for various reasons:

• The list of branches and tags can get very large.
• Branches and tags may not be meaningful for the end-user.
• Branches and tags don't communicate versioning thoroughly.

Releasing new resource versions to the Catalog

The versions that should be displayed for the resource should be the project releases. Creating project releases is an official act of versioning a resource.

A resource page would have:

• The latest release in evidence (for example: the default version).
• The ability to inspect and use past releases of the resource.
• The documentation represented by the README.md.

Users should be able to release new versions of the resource in a CI pipeline job, similar to how software is being deployed following Continuous Delivery principles.

To ensure that the components repository and the including components meet quality standards, users can test them before releasing new versions in the CI Catalog.

Some examples of checks we can run during the release of a new resource version:

• Ensure the project contains a README.md in the root directory.
• Ensure the project description exists.
• If an index of available components is present for a components repository, ensure each component has valid YAML.

Once a new release for the project gets created we index the resource's metadata. We want to initially index as much metadata as possible, to gain more flexibility in how we design CI Catalog's main page. We don't want to be constrained by the lack of data available to properly visualize resources in the CI Catalog. To do that, we may need to find all resources that are being released and index their data and metadata. For example: index the content of spec: section for CI components.

See an example of development workflow for a components repository.

Note about future resource types

In the future, to support multiple types of resources in the Catalog we could require a file catalog-resource.yml to be defined in the root directory of the project:

name: DAST
description: Scan a web endpoint to find vulnerabilities
category: security
tags: [dynamic analysis, security scanner]
type: components_repository

This file could also be used for indexing metadata about the content of the resource. For example, users could list the components in the repository and we can index further data for search purpose:

name: DAST
description: Scan a web endpoint to find vulnerabilities
category: security
tags: [dynamic analysis, security scanner]
type: components_repository
metadata:
  components:
    - all-scans
    - scan-x
    - scan-y

Implementation guidelines

• Start with the smallest user base. Dogfood the feature for gitlab-org and gitlab-com groups. Involve the Engineering Productivity and other groups authoring pipeline configurations to test and validate our solutions.
• Ensure we can integrate all the feedback gathered, even if that means changing the technical design or UX. Until we make the feature GA we should have clear expectations with early adopters.
• Reuse existing functionality as much as possible. Don't reinvent the wheel on the initial iterations. For example: reuse project features like title, description, avatar to build a catalog.
• Leverage GitLab features for the development lifecycle of the components (testing via .gitlab-ci.yml, release management, Pipeline Editor, etc.).
• Design the catalog with self-managed support in mind.
• Allow the catalog and the workflow to support future types of pipeline constructs and new ways of using them.
• Design components and catalog following industry best practice related to building deterministic package managers.

Iterations

1. Experimentation phase
   • Build an MVC behind a feature flag with namespace actor.
   • Enable the feature flag only for gitlab-com and gitlab-org namespaces to initiate the dogfooding.
   • Refine the solution and UX based on feedback.
   • Find customers to be early adopters of this feature and iterate on their feedback.
2. Design new pipeline constructs (in parallel with other phases)
   • Start the technical and design process to work on proposals for new pipeline constructs (steps, workflows, templates).
   • Implement new constructs. The catalog must be compatible with them.
   • Dogfood new constructs and iterate on feedback.
   • Release new constructs on private catalogs.
3. Release the private catalog for groups on Ultimate plan.
   • Iterate on feedback.
4. Release the public catalog for all GitLab users (prospect feature)
   • Publish new versions of GitLab CI templates as components using the new constructs whenever possible.
   • Allow self-managed administrators to populate their self-managed catalog by importing/updating components from GitLab.com or from repository exports.
   • Iterate on feedback.

Limits

Any MVC that exposes a feature should be added with limitations from the beginning. It's safer to add new features with restrictions than trying to limit a feature after it's being used. We can always soften the restrictions later depending on user demand.

Some limits we could consider adding:

• number of components that a single project can contain/export
• number of imports that a .gitlab-ci.yml file can use
• number of imports that a component can declare/use
• max level of nested imports
• max length of the exported component name

Who

Proposal:

Role	Who
Author	Fabio Pitino
Engineering Leaders	Cheryl Li, Mark Nuzzo
Product Manager	Dov Hershkovitch
Architecture Evolution Coaches	Kamil Trzciński, Grzegorz Bizon

DRIs:

Role	Who
Leadership	Mark Nuzzo
Product	Dov Hershkovitch
Engineering	Fabio Pitino
UX	Kevin Comoli (interim), Sunjung Park

Domain experts:

Area	Who
Verify / Pipeline authoring	Avielle Wolfe
Verify / Pipeline authoring	Laura Montemayor