Pipeline Architecture

Pipelines are the fundamental building blocks for CI/CD in GitLab. This page documents some of the important concepts related to them.

There are three main ways to structure your pipelines, each with their own advantages. These methods can be mixed and matched if needed:

• Basic: Good for straightforward projects where all the configuration is in one easy to find place.
• Directed Acyclic Graph: Good for large, complex projects that need efficient execution.
• Child/Parent Pipelines: Good for monorepos and projects with lots of independently defined components.

For more details about any of the keywords used below, check out our CI YAML reference for details.

Basic Pipelines

This is the simplest pipeline in GitLab. It will run everything in the build stage concurrently, and once all of those finish, it will run everything in the test stage the same way, and so on. It's not the most efficient, and if you have lots of steps it can grow quite complex, but it's easier to maintain:

graph LR
  subgraph deploy stage
    deploy --> deploy_a
    deploy --> deploy_b
  end
  subgraph test stage
    test --> test_a
    test --> test_b
  end
  subgraph build stage
    build --> build_a
    build --> build_b
  end
  build_a -.-> test
  build_b -.-> test
  test_a -.-> deploy
  test_b -.-> deploy

Example basic /.gitlab-ci.yml pipeline configuration matching the diagram:

stages:
  - build
  - test
  - deploy

image: alpine

build_a:
  stage: build
  script:
    - echo "This job builds something."

build_b:
  stage: build
  script:
    - echo "This job builds something else."

test_a:
  stage: test
  script:
    - echo "This job tests something. It will only run when all jobs in the"
    - echo "build stage are complete."

test_b:
  stage: test
  script:
    - echo "This job tests something else. It will only run when all jobs in the"
    - echo "build stage are complete too. It will start at about the same time as test_a."

deploy_a:
  stage: deploy
  script:
    - echo "This job deploys something. It will only run when all jobs in the"
    - echo "test stage complete."

deploy_b:
  stage: deploy
  script:
    - echo "This job deploys something else. It will only run when all jobs in the"
    - echo "test stage complete. It will start at about the same time as deploy_a."

Directed Acyclic Graph Pipelines

If efficiency is important to you and you want everything to run as quickly as possible, you can use Directed Acyclic Graphs (DAG). Use the needs keyword to define dependency relationships between your jobs. When GitLab knows the relationships between your jobs, it can run everything as fast as possible, and even skips into subsequent stages when possible.

In the example below, if build_a and test_a are much faster than build_b and test_b, GitLab will start deploy_a even if build_b is still running.

graph LR
  subgraph Pipeline using DAG
    build_a --> test_a --> deploy_a
    build_b --> test_b --> deploy_b
  end

Example DAG /.gitlab-ci.yml configuration matching the diagram:

stages:
  - build
  - test
  - deploy

image: alpine

build_a:
  stage: build
  script:
    - echo "This job builds something quickly."

build_b:
  stage: build
  script:
    - echo "This job builds something else slowly."

test_a:
  stage: test
  needs: build_a
  script:
    - echo "This test job will start as soon as build_a finishes."
    - echo "It will not wait for build_b, or other jobs in the build stage, to finish."

test_b:
  stage: test
  needs: build_b
  script:
    - echo "This test job will start as soon as build_b finishes."
    - echo "It will not wait for other jobs in the build stage to finish."

deploy_a:
  stage: deploy
  needs: test_a
  script:
    - echo "Since build_a and test_a run quickly, this deploy job can run much earlier."
    - echo "It does not need to wait for build_b or test_b."

deploy_b:
  stage: deploy
  needs: test_b
  script:
    - echo "Since build_b and test_b run slowly, this deploy job will run much later."

Child / Parent Pipelines

In the examples above, it's clear we've got two types of things that could be built independently. This is an ideal case for using Child / Parent Pipelines) via the trigger keyword. It will separate out the configuration into multiple files, keeping things very simple. You can also combine this with:

• The rules keyword: For example, have the child pipelines triggered only when there are changes to that area.
• The include keyword: Bring in common behaviors, ensuring you are not repeating yourself.
• DAG pipelines inside of child pipelines, achieving the benefits of both.

graph LR
  subgraph Parent pipeline
    trigger_a -.-> build_a
  trigger_b -.-> build_b
    subgraph child pipeline B
    build_b --> test_b --> deploy_b
    end

    subgraph child pipeline A
      build_a --> test_a --> deploy_a
    end
  end

Example /.gitlab-ci.yml configuration for the parent pipeline matching the diagram:

stages:
  - triggers

trigger_a:
  stage: triggers
  trigger:
    include: a/.gitlab-ci.yml
  rules:
    - changes:
      - a/*

trigger_b:
  stage: triggers
  trigger:
    include: b/.gitlab-ci.yml
  rules:
    - changes:
      - b/*

Example child a pipeline configuration, located in /a/.gitlab-ci.yml, making use of the DAG needs: keyword:

stages:
  - build
  - test
  - deploy

image: alpine

build_a:
  stage: build
  script:
    - echo "This job builds something."

test_a:
  stage: test
  needs: build_a
  script:
    - echo "This job tests something."

deploy_a:
  stage: deploy
  needs: test_a
  script:
    - echo "This job deploys something."

Example child b pipeline configuration, located in /b/.gitlab-ci.yml, making use of the DAG needs: keyword:

stages:
  - build
  - test
  - deploy

image: alpine

build_b:
  stage: build
  script:
    - echo "This job builds something else."

test_b:
  stage: test
  needs: build_b
  script:
    - echo "This job tests something else."

deploy_b:
  stage: deploy
  needs: test_b
  script:
    - echo "This job deploys something else."

It's also possible to set jobs to run before or after triggering child pipelines, for example if you have common setup steps or a unified deployment at the end.