GitLab Matrix Expressions

Matrix expressions allow you to define dependencies between jobs that use parallel:matrix without manually listing every possible combination. This creates a "1:1 mapping" where a downstream job only waits for the specific upstream job that matches its matrix variables (e.g., linux:test waits only for linux:build, not windows:build).

Syntax

The syntax uses the matrix context to reference the current job's matrix identifiers.

'$[[ matrix.IDENTIFIER ]]'

Note: The quotes ' around the expression are required to ensure valid YAML parsing.

Usage Patterns

1. Basic 1:1 Dependency

Use this pattern when you have two jobs (e.g., Build and Test) that run on the same matrix (e.g., multiple providers) and you want them to stay synchronized.

linux:build:
  stage: build
  script: echo "Building..."
  parallel:
    matrix:
      - PROVIDER: [aws, gcp]

linux:test:
  stage: test
  script: echo "Testing..."
  parallel:
    matrix:
      - PROVIDER: [aws, gcp]
  needs:
    - job: linux:build
      parallel:
        matrix:
          # This maps the current job's PROVIDER to the upstream job's PROVIDER
          - PROVIDER: ['$[[ matrix.PROVIDER ]]']

Result: linux:test [aws] only waits for linux:build [aws].

2. Using YAML Anchors (DRY Principle)

For complex matrices, define the matrix once in a hidden job or anchor to avoid repetition and errors.

.build_matrix: &build_matrix
  parallel:
    matrix:
      - OS: ["ubuntu", "alpine"]
        ARCH: ["amd64", "arm64"]

compile_binary:
  <<: *build_matrix
  stage: compile
  script: echo "Compiling..."

integration_test:
  <<: *build_matrix
  stage: test
  needs:
    - job: compile_binary
      parallel:
        matrix:
          - OS: ['$[[ matrix.OS ]]']
            ARCH: ['$[[ matrix.ARCH ]]']

3. Subset Dependencies

You can mix matrix expressions with static values to depend on a specific subset of upstream jobs.

needs:
  - job: prepare_env
    parallel:
      matrix:
        # Match the platform dynamically
        - PLATFORM: ['$[[ matrix.PLATFORM ]]']
          # But only depend on the specific Node version '18'
          VERSION: ["18"] 

Limitations

• Compile-time only: Identifiers are resolved when the pipeline is created. You cannot use runtime variables.
• String replacement only: You cannot perform arithmetic or complex logic inside [[ ... ]].
• Scope: You can only reference identifiers defined in the current job's parallel:matrix section.