• Introduction
  • Validate the .gitlab-ci.yml
  • Unavailable names for jobs
  • Using reserved keywords
• Configuration parameters
• Global parameters
  • Global defaults
    • inherit
  • stages
  • workflow:rules
    • workflow:rules templates
  • include
    • include:local
    • include:file
    • include:remote
    • include:template
    • Nested includes
    • Additional includes examples
• Parameter details
  • image
    • image:name
    • image:entrypoint
    • services
      • services:name
      • services:alias
      • services:entrypoint
      • services:command
  • script
    • before_script and after_script
    • Coloring script output
    • Multiline commands
  • stage
    • Using your own Runners
    • .pre and .post
  • extends
    • Merge details
    • Using extends and include together
  • rules
    • Rules attributes
    • Rules clauses
    • Differences between rules and only/except
      • rules:if
      • Common if clauses for rules
      • rules:changes
      • rules:exists
      • rules:allow_failure
    • Complex rule clauses
  • only/except (basic)
    • Regular expressions
    • Supported only/except regexp syntax
  • only/except (advanced)
    • only:refs/except:refs
    • only:kubernetes/except:kubernetes
    • only:variables/except:variables
    • only:changes/except:changes
      • Using only:changes with pipelines for merge requests
      • Using only:changes without pipelines for merge requests
      • Using only:changes with scheduled pipelines
  • needs
    • Requirements and limitations
      • Changing the needs: job limit
    • Artifact downloads with needs
    • Cross project artifact downloads with needs
      • Artifact downloads between pipelines in the same project
  • tags
  • allow_failure
  • when
    • when:manual
      • Protecting manual jobs
    • when:delayed
  • environment
    • environment:name
    • environment:url
    • environment:on_stop
    • environment:action
    • environment:auto_stop_in
    • environment:kubernetes
    • Dynamic environments
  • cache
    • cache:paths
    • cache:key
      • cache:key:files
      • cache:key:prefix
    • cache:untracked
    • cache:policy
  • artifacts
    • artifacts:paths
    • artifacts:exclude
    • artifacts:expose_as
    • artifacts:name
    • artifacts:untracked
    • artifacts:when
    • artifacts:expire_in
    • artifacts:reports
    • dependencies
      • When a dependent job will fail
  • coverage
  • retry
  • timeout
  • parallel
    • Parallel matrix jobs
  • trigger
    • Simple trigger syntax for multi-project pipelines
    • Complex trigger syntax for multi-project pipelines
    • trigger syntax for child pipeline
      • Trigger child pipeline with generated configuration file
    • Linking pipelines with trigger:strategy
    • Trigger a pipeline by API call
  • interruptible
  • resource_group
  • release
    • release-cli Docker image
    • Script
    • release:tag_name
    • release:name
    • release:description
    • release:ref
    • release:milestones
    • release:released_at
    • Complete example for release
    • releaser-cli command line
  • pages
• variables
  • Git strategy
  • Git submodule strategy
  • Git checkout
  • Git clean flags
  • Git fetch extra flags
  • Job stages attempts
  • Shallow cloning
  • Custom build directories
    • Handling concurrency
    • Nested paths
• Special YAML features
  • Anchors
    • YAML anchors for before_script and after_script
    • YAML anchors for script
    • YAML anchors for variables
  • Hide jobs
• Skip Pipeline
• Processing Git pushes
• Deprecated parameters
  • Globally-defined types
  • Job-defined type
  • Globally-defined image, services, cache, before_script, after_script

GitLab CI/CD pipeline configuration reference

GitLab CI/CD pipelines are configured using a YAML file called .gitlab-ci.yml within each project.

The .gitlab-ci.yml file defines the structure and order of the pipelines and determines:

• What to execute using GitLab Runner.
• What decisions to make when specific conditions are encountered. For example, when a process succeeds or fails.

This topic covers CI/CD pipeline configuration. For other CI/CD configuration information, see:

• GitLab CI/CD Variables, for configuring the environment the pipelines run in.
• GitLab Runner advanced configuration, for configuring GitLab Runner.

We have complete examples of configuring pipelines:

• For a quick introduction to GitLab CI/CD, follow our quick start guide.
• For a collection of examples, see GitLab CI/CD Examples.
• To see a large .gitlab-ci.yml file used in an enterprise, see the .gitlab-ci.yml file for gitlab.

For some additional information about GitLab CI/CD:

•  Watch the CI/CD Ease of configuration video.
• Watch the Making the case for CI/CD in your organization webcast to learn the benefits of CI/CD and how to measure the results of CI/CD automation.
•  Learn how Verizon reduced rebuilds from 30 days to under 8 hours with GitLab.

Introduction

Pipeline configuration begins with jobs. Jobs are the most fundamental element of a .gitlab-ci.yml file.

Jobs are:

• Defined with constraints stating under what conditions they should be executed.
• Top-level elements with an arbitrary name and must contain at least the script clause.
• Not limited in how many can be defined.

For example:

job1:
  script: "execute-script-for-job1"

job2:
  script: "execute-script-for-job2"

The above example is the simplest possible CI/CD configuration with two separate jobs, where each of the jobs executes a different command. Of course a command can execute code directly (./configure;make;make install) or run a script (test.sh) in the repository.

Jobs are picked up by Runners and executed within the environment of the Runner. What is important, is that each job is run independently from each other.

Validate the .gitlab-ci.yml

Each instance of GitLab CI/CD has an embedded debug tool called Lint, which validates the content of your .gitlab-ci.yml files. You can find the Lint under the page ci/lint of your project namespace. For example, https://gitlab.example.com/gitlab-org/project-123/-/ci/lint.

Unavailable names for jobs

Each job must have a unique name, but there are a few reserved keywords that can’t be used as job names:

• image
• services
• stages
• types
• before_script
• after_script
• variables
• cache
• include

Using reserved keywords

If you get validation error when using specific values (for example, true or false), try to:

• Quote them.
• Change them to a different form. For example, /bin/true.

Configuration parameters

A job is defined as a list of parameters that define the job’s behavior.

The following table lists available parameters for jobs:

Global parameters

Some parameters must be defined at a global level, affecting all jobs in the pipeline.

Global defaults

Some parameters can be set globally as the default for all jobs using the default: keyword. Default parameters can then be overridden by job-specific configuration.

The following job parameters can be defined inside a default: block:

• image
• services
• before_script
• after_script
• tags
• cache
• artifacts
• retry
• timeout
• interruptible

In the following example, the ruby:2.5 image is set as the default for all jobs except the rspec 2.6 job, which uses the ruby:2.6 image:

default:
  image: ruby:2.5

rspec:
  script: bundle exec rspec

rspec 2.6:
  image: ruby:2.6
  script: bundle exec rspec

inherit

You can disable inheritance of globally defined defaults and variables with the inherit: parameter.

To enable or disable the inheritance of all variables: or default: parameters, use the following format:

• default: true or default: false
• variables: true or variables: false

To inherit only a subset of default: parameters or variables:, specify what you wish to inherit, and any not listed will not be inherited. Use one of the following formats:

inherit:
  default: [parameter1, parameter2]
  variables: [VARIABLE1, VARIABLE2]

Or:

inherit:
  default:
    - parameter1
    - parameter2
  variables:
    - VARIABLE1
    - VARIABLE2

In the example below:

• rubocop:
  • will inherit: Nothing.
• rspec:
  • will inherit: the default image and the WEBHOOK_URL variable.
  • will not inherit: the default before_script and the DOMAIN variable.
• capybara:
  • will inherit: the default before_script and image.
  • will not inherit: the DOMAIN and WEBHOOK_URL variables.
• karma:
  • will inherit: the default image and before_script, and the DOMAIN variable.
  • will not inherit: WEBHOOK_URL variable.

default:
  image: 'ruby:2.4'
  before_script:
    - echo Hello World

variables:
  DOMAIN: example.com
  WEBHOOK_URL: https://my-webhook.example.com

rubocop:
  inherit:
    default: false
    variables: false
  script: bundle exec rubocop

rspec:
  inherit:
    default: [image]
    variables: [WEBHOOK_URL]
  script: bundle exec rspec

capybara:
  inherit:
    variables: false
  script: bundle exec capybara

karma:
  inherit:
    default: true
    variables: [DOMAIN]
  script: karma

stages

stages is used to define stages that contain jobs and is defined globally for the pipeline.

The specification of stages allows for having flexible multi stage pipelines. The ordering of elements in stages defines the ordering of jobs’ execution:

1. Jobs of the same stage are run in parallel.
2. Jobs of the next stage are run after the jobs from the previous stage complete successfully.

Let’s consider the following example, which defines 3 stages:

stages:
  - build
  - test
  - deploy

1. First, all jobs of build are executed in parallel.
2. If all jobs of build succeed, the test jobs are executed in parallel.
3. If all jobs of test succeed, the deploy jobs are executed in parallel.
4. If all jobs of deploy succeed, the commit is marked as passed.
5. If any of the previous jobs fails, the commit is marked as failed and no jobs of further stage are executed.

There are also two edge cases worth mentioning:

1. If no stages are defined in .gitlab-ci.yml, then the build, test and deploy are allowed to be used as job’s stage by default.
2. If a job does not specify a stage, the job is assigned the test stage.

workflow:rules

The top-level workflow: key applies to the entirety of a pipeline, and will determine whether or not a pipeline is created. It currently accepts a single rules: key that operates similarly to rules: defined within jobs, enabling dynamic configuration of the pipeline.

If you are new to GitLab CI/CD and workflow: rules, you may find the workflow:rules templates useful.

To define your own workflow: rules, the configuration options currently available are:

• if: Define a rule.
• when: May be set to always or never only. If not provided, the default value is always​.

If a pipeline attempts to run but matches no rule, it’s dropped and doesn’t run.

For example, with the following configuration, pipelines run for all push events (changes to branches and new tags) as long as they don’t have -wip in the commit message. Scheduled pipelines and merge request pipelines don’t run, as there’s no rule allowing them.

workflow:
  rules:
    - if: $CI_COMMIT_REF_NAME =~ /-wip$/
      when: never
    - if: '$CI_PIPELINE_SOURCE == "push"'

This example has strict rules, and no other pipelines can run.

Alternatively, you can have loose rules by using only when: never rules, followed by a final when: always rule. This allows all types of pipelines, except for any that match the when: never rules:

workflow:
  rules:
    - if: '$CI_PIPELINE_SOURCE == "schedule"'
      when: never
    - if: '$CI_PIPELINE_SOURCE == "push"'
      when: never
    - when: always

This example never allows pipelines for schedules or push (branches and tags) pipelines, but does allow pipelines in all other cases, including merge request pipelines.

As with rules defined in jobs, be careful not to use a configuration that allows merge request pipelines and branch pipelines to run at the same time, or you could have duplicate pipelines.

Useful workflow rules clauses:

workflow:rules templates

We provide pre-made templates for use with your pipelines that set up workflow: rules for common scenarios. Usage of these will make things easier and prevent duplicate pipelines from running.

The Branch-Pipelines template makes your pipelines run for branches and tags.

Branch pipeline status will be displayed within merge requests that use that branch as a source, but this pipeline type does not support any features offered by Merge Request Pipelines like Pipelines for Merge Results or Merge Trains. Use this template if you are intentionally avoiding those features.

It is included as follows:

include:
  - template: 'Workflows/Branch-Pipelines.gitlab-ci.yml'

The MergeRequest-Pipelines template makes your pipelines run for the default branch (usually master), tags, and all types of merge request pipelines. Use this template if you use any of the the Pipelines for Merge Requests features, as mentioned above.

It is included as follows:

include:
  - template: 'Workflows/MergeRequest-Pipelines.gitlab-ci.yml'

include

Using the include keyword allows the inclusion of external YAML files. This helps to break down the CI/CD configuration into multiple files and increases readability for long configuration files. It’s also possible to have template files stored in a central repository and projects include their configuration files. This helps avoid duplicated configuration, for example, global default variables for all projects.

include requires the external YAML file to have the extensions .yml or .yaml, otherwise the external file won’t be included.

include supports the following inclusion methods:

The include methods do not support variable expansion.

The files defined by include are:

• Deep merged with those in .gitlab-ci.yml.
• Always evaluated first and merged with the content of .gitlab-ci.yml, regardless of the position of the include keyword.

include:local

include:local includes a file from the same repository as .gitlab-ci.yml. It’s referenced using full paths relative to the root directory (/).

You can only use files that are currently tracked by Git on the same branch your configuration file is on. In other words, when using a include:local, make sure that both .gitlab-ci.yml and the local file are on the same branch.

All nested includes will be executed in the scope of the same project, so it’s possible to use local, project, remote, or template includes.

Example:

include:
  - local: '/templates/.gitlab-ci-template.yml'

This can be defined as a short local include:

include: '.gitlab-ci-production.yml'

include:file

To include files from another private project under the same GitLab instance, use include:file. This file is referenced using full paths relative to the root directory (/). For example:

include:
  - project: 'my-group/my-project'
    file: '/templates/.gitlab-ci-template.yml'

You can also specify ref, with the default being the HEAD of the project:

include:
  - project: 'my-group/my-project'
    ref: master
    file: '/templates/.gitlab-ci-template.yml'

  - project: 'my-group/my-project'
    ref: v1.0.0
    file: '/templates/.gitlab-ci-template.yml'

  - project: 'my-group/my-project'
    ref: 787123b47f14b552955ca2786bc9542ae66fee5b # Git SHA
    file: '/templates/.gitlab-ci-template.yml'

All nested includes will be executed in the scope of the target project, so it’s possible to use local (relative to target project), project, remote or template includes.

include:remote

include:remote can be used to include a file from a different location, using HTTP/HTTPS, referenced by using the full URL. The remote file must be publicly accessible through a simple GET request as authentication schemas in the remote URL are not supported. For example:

include:
  - remote: 'https://gitlab.com/awesome-project/raw/master/.gitlab-ci-template.yml'

All nested includes will be executed without context as public user, so only another remote or public project, or template, is allowed.

include:template

include:template can be used to include .gitlab-ci.yml templates that are shipped with GitLab.

For example:

# File sourced from GitLab's template collection
include:
  - template: Auto-DevOps.gitlab-ci.yml

Multiple include:template files:

include:
  - template: Android-Fastlane.gitlab-ci.yml
  - template: Auto-DevOps.gitlab-ci.yml

All nested includes will be executed only with the permission of the user, so it’s possible to use project, remote or template includes.

Nested includes

Nested includes allow you to compose a set of includes.

A total of 100 includes is allowed, but duplicate includes are considered a configuration error.

Since GitLab 12.4, the time limit for resolving all files is 30 seconds.

Additional includes examples

There is a list of additional includes examples available.

Parameter details

The following are detailed explanations for parameters used to configure CI/CD pipelines.

image

Used to specify a Docker image to use for the job.

For:

• Simple definition examples, see Define image and services from .gitlab-ci.yml.
• Detailed usage information, refer to Docker integration documentation.

image:name

An extended Docker configuration option.

For more information, see Available settings for image.

image:entrypoint

An extended Docker configuration option.

For more information, see Available settings for image.

services

Used to specify a service Docker image, linked to a base image specified in image.

For:

• Simple definition examples, see Define image and services from .gitlab-ci.yml.
• Detailed usage information, refer to Docker integration documentation.
• For example services, see GitLab CI/CD Services.

services:name

An extended Docker configuration option.

For more information, see Available settings for services.

services:alias

An extended Docker configuration option.

For more information, see Available settings for services.

services:entrypoint

An extended Docker configuration option.

For more information, see Available settings for services.

services:command

An extended Docker configuration option.

For more information, see Available settings for services.

script

script is the only required keyword that a job needs. It’s a shell script which is executed by the Runner. For example:

job:
  script: "bundle exec rspec"

YAML anchors for scripts are available.

This parameter can also contain several commands using an array:

job:
  script:
    - uname -a
    - bundle exec rspec

If any of the script commands return an exit code different from zero, the job will fail and further commands won’t be executed. This behavior can be avoided by storing the exit code in a variable:

job:
  script:
    - false || exit_code=$?
    - if [ $exit_code -ne 0 ]; then echo "Previous command failed"; fi;

before_script and after_script

before_script is used to define a command that should be run before each job, including deploy jobs, but after the restoration of any artifacts. This must be an array.

Scripts specified in before_script are concatenated with any scripts specified in the main script, and executed together in a single shell.

after_script is used to define the command that will be run after each job, including failed ones. This must be an array.

Scripts specified in after_script are executed in a new shell, separate from any before_script or script scripts. As a result, they:

• Have a current working directory set back to the default.
• Have no access to changes done by scripts defined in before_script or script, including:
  • Command aliases and variables exported in script scripts.
  • Changes outside of the working tree (depending on the Runner executor), like software installed by a before_script or script script.
• Have a separate timeout, which is hard coded to 5 minutes. See related issue for details.
• Don’t affect the job’s exit code. If the script section succeeds and the after_script times out or fails, the job will exit with code 0 (Job Succeeded).

It’s possible to overwrite a globally defined before_script or after_script if you set it per-job:

default:
  before_script:
    - global before script

job:
  before_script:
    - execute this instead of global before script
  script:
    - my command
  after_script:
    - execute this after my script

YAML anchors for before_script and after_script are available.

Coloring script output

Script output can be colored using ANSI escape codes, or by running commands or programs that output ANSI escape codes.

For example, using Bash with color codes:

job:
  script:
    - echo -e "\e[31mThis text is red,\e[0m but this text isn't\e[31m however this text is red again."

You can define the color codes in Shell variables, or even custom environment variables, which makes the commands easier to read and reusable.

For example, using the same example as above and variables defined in a before_script:

job:
  before_script:
    - TXT_RED="\e[31m" && TXT_CLEAR="\e[0m"
  script:
    - echo -e "${TXT_RED}This text is red,${TXT_CLEAR} but this part isn't${TXT_RED} however this part is again."
    - echo "This text is not colored"

Or with PowerShell color codes:

job:
  before_script:
    - $esc="$([char]27)"; $TXT_RED="$esc[31m"; $TXT_CLEAR="$esc[0m"
  script:
    - Write-Host $TXT_RED"This text is red,"$TXT_CLEAR" but this text isn't"$TXT_RED" however this text is red again."
    - Write-Host "This text is not colored"

Multiline commands

You can split long commands into multi-line commands to improve readability using | (literal) and > (folded) YAML multiline block scalar indicators.

You can use the | (literal) YAML multiline block scalar indicator to write commands over multiple lines in the script section of a job description. Each line is treated as a separate command. Only the first command is repeated in the job log, but additional commands are still executed:

job:
  script:
    - |
      echo "First command line."
      echo "Second command line."
      echo "Third command line."

The example above renders in the job log as:

$ echo First command line # collapsed multi-line command
First command line
Second command line.
Third command line.

The > (folded) YAML multiline block scalar indicator treats empty lines between sections as the start of a new command:

job:
  script:
    - >
      echo "First command line
      is split over two lines."

      echo "Second command line."

This behaves similarly to writing multiline commands without the > or | block scalar indicators:

job:
  script:
    - echo "First command line
      is split over two lines."

      echo "Second command line."

Both examples above render in the job log as:

$ echo First command line is split over two lines. # collapsed multi-line command
First command line is split over two lines.
Second command line.

When the > or | block scalar indicators are omitted, GitLab will form the command by concatenating non-empty lines, so make sure the lines can run when concatenated.

Shell here documents work with the | and > operators as well. The example below transliterates the lower case letters to upper case:

job:
  script:
    - |
      tr a-z A-Z << END_TEXT
        one two three
        four five six
      END_TEXT

Results in:

$ tr a-z A-Z << END_TEXT # collapsed multi-line command
  ONE TWO THREE
  FOUR FIVE SIX

stage

stage is defined per-job and relies on stages which is defined globally. It allows to group jobs into different stages, and jobs of the same stage are executed in parallel (subject to certain conditions). For example:

stages:
  - build
  - test
  - deploy

job 0:
  stage: .pre
  script: make something useful before build stage

job 1:
  stage: build
  script: make build dependencies

job 2:
  stage: build
  script: make build artifacts

job 3:
  stage: test
  script: make test

job 4:
  stage: deploy
  script: make deploy

job 5:
  stage: .post
  script: make something useful at the end of pipeline

Using your own Runners

When using your own Runners, GitLab Runner runs only one job at a time by default (see the concurrent flag in Runner global settings for more information).

Jobs will run on your own Runners in parallel only if:

• Run on different Runners.
• The Runner’s concurrent setting has been changed.

.pre and .post

The following stages are available to every pipeline:

• .pre, which is guaranteed to always be the first stage in a pipeline.
• .post, which is guaranteed to always be the last stage in a pipeline.

User-defined stages are executed after .pre and before .post.

The order of .pre and .post can’t be changed, even if defined out of order in .gitlab-ci.yml. For example, the following are equivalent configuration:

• Configured in order:

  stages:
    - .pre
    - a
    - b
    - .post
  

• Configured out of order:

  stages:
    - a
    - .pre
    - b
    - .post
  

• Not explicitly configured:

  stages:
    - a
    - b
  

extends

extends defines entry names that a job that uses extends is going to inherit from.

It’s an alternative to using YAML anchors and is a little more flexible and readable:

.tests:
  script: rake test
  stage: test
  only:
    refs:
      - branches

rspec:
  extends: .tests
  script: rake rspec
  only:
    variables:
      - $RSPEC

In the example above, the rspec job inherits from the .tests template job. GitLab will perform a reverse deep merge based on the keys. GitLab will:

• Merge the rspec contents into .tests recursively.
• Not merge the values of the keys.

This results in the following rspec job:

rspec:
  script: rake rspec
  stage: test
  only:
    refs:
      - branches
    variables:
      - $RSPEC

If you do want to include the rake test, see before_script and after_script.

.tests in this example is a hidden job, but it’s possible to inherit from regular jobs as well.

extends supports multi-level inheritance, however it’s not recommended to use more than three levels. The maximum nesting level that is supported is 10. The following example has two levels of inheritance:

.tests:
  only:
    - pushes

.rspec:
  extends: .tests
  script: rake rspec

rspec 1:
  variables:
    RSPEC_SUITE: '1'
  extends: .rspec

rspec 2:
  variables:
    RSPEC_SUITE: '2'
  extends: .rspec

spinach:
  extends: .tests
  script: rake spinach

In GitLab 12.0 and later, it’s also possible to use multiple parents for extends.

Merge details

extends is able to merge hashes but not arrays. The algorithm used for merge is “closest scope wins”, so keys from the last member will always override anything defined on other levels. For example:

.only-important:
  variables:
    URL: "http://my-url.internal"
    IMPORTANT_VAR: "the details"
  only:
    - master
    - stable
  tags:
    - production
  script:
    - echo "Hello world!"

.in-docker:
  variables:
    URL: "http://docker-url.internal"
  tags:
    - docker
  image: alpine

rspec:
  variables:
    GITLAB: "is-awesome"
  extends:
    - .only-important
    - .in-docker
  script:
    - rake rspec

This results in the following rspec job:

rspec:
  variables:
    URL: "http://docker-url.internal"
    IMPORTANT_VAR: "the details"
    GITLAB: "is-awesome"
  only:
    - master
    - stable
  tags:
    - docker
  image: alpine
  script:
    - rake rspec

Note that in the example above:

• variables sections have been merged but that URL: "http://my-url.internal" has been overwritten by URL: "http://docker-url.internal".
• tags: ['production'] has been overwritten by tags: ['docker'].
• script has not been merged but rather script: ['echo "Hello world!"'] has been overwritten by script: ['rake rspec']. Arrays can be merged using YAML anchors.

Using extends and include together

extends works across configuration files combined with include.

For example, if you have a local included.yml file:

.template:
  script:
    - echo Hello!

Then, in .gitlab-ci.yml you can use it like this:

include: included.yml

useTemplate:
  image: alpine
  extends: .template

This will run a job called useTemplate that runs echo Hello! as defined in the .template job, and uses the alpine Docker image as defined in the local job.

rules

The rules keyword can be used to include or exclude jobs in pipelines.

Rules are evaluated in order until the first match. When matched, the job is either included or excluded from the pipeline, depending on the configuration. If included, the job also has certain attributes added to it.

Rules attributes

The job attributes allowed by rules are:

• when: If not defined, defaults to when: on_success.
  • If used as when: delayed, start_in is also required.
• allow_failure: If not defined, defaults to allow_failure: false.

If a rule evaluates to true, and when has any value except never, the job is included in the pipeline.

For example:

docker build:
  script: docker build -t my-image:$CI_COMMIT_REF_SLUG .
  rules:
    - if: '$CI_COMMIT_BRANCH == "master"'
      when: delayed
      start_in: '3 hours'
      allow_failure: true

Additional job configuration may be added to rules in the future. If something useful is not available, please open an issue.

Rules clauses

Available rule clauses are:

Rules are evaluated in order until a match is found. If a match is found, the attributes are checked to see if the job should be added to the pipeline. If no attributes are defined, the defaults are:

• when: on_success
• allow_failure: false

The job is added to the pipeline:

• If a rule matches and has when: on_success, when: delayed or when: always.
• If no rules match, but the last clause is when: on_success, when: delayed or when: always (with no rule).

The job is not added to the pipeline:

• If no rules match, and there is no standalone when: on_success, when: delayed or when: always.
• If a rule matches, and has when: never as the attribute.

For example, using if clauses to strictly limit when jobs run:

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
      when: manual
      allow_failure: true
    - if: '$CI_PIPELINE_SOURCE == "schedule"'

In this example:

• If the pipeline is for a merge request, the first rule matches, and the job is added to the merge request pipeline with attributes of:
  • when: manual (manual job)
  • allow_failure: true (allows the pipeline to continue running even if the manual job is not run)
• If the pipeline is not for a merge request, the first rule doesn’t match, and the second rule is evaluated.
• If the pipeline is a scheduled pipeline, the second rule matches, and the job is added to the scheduled pipeline. Since no attributes were defined, it is added with:
  • when: on_success (default)
  • allow_failure: false (default)
• In all other cases, no rules match, so the job is not added to any other pipeline.

Alternatively, you can define a set of rules to exclude jobs in a few cases, but run them in all other cases:

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
      when: never
    - if: '$CI_PIPELINE_SOURCE == "schedule"'
      when: never
    - when: on_success

• If the pipeline is for a merge request, the job is not be added to the pipeline.
• If the pipeline is a scheduled pipeline, the job is not be added to the pipeline.
• In all other cases, the job is added to the pipeline, with when: on_success.

Differences between rules and only/except

Jobs defined with only/except do not trigger merge request pipelines by default. You must explicitly add only: merge_requests.

Jobs defined with rules can trigger all types of pipelines. You do not have to explicitly configure each type.

For example:

job:
  script: "echo This creates double pipelines!"
  rules:
    - if: '$CUSTOM_VARIABLE == "false"'
      when: never
    - when: always

This job does not run when $CUSTOM_VARIABLE is false, but it does run in all other pipelines, including both push (branch) and merge request pipelines. With this configuration, every push to an open merge request’s source branch causes duplicated pipelines. Explicitly allowing both push and merge request pipelines in the same job could have the same effect.

We recommend using workflow: rules to limit which types of pipelines are permitted. Allowing only merge request pipelines, or only branch pipelines, eliminates duplicated pipelines. Alternatively, you can rewrite the rules to be stricter, or avoid using a final when (always, on_success or delayed).

It is not possible to run a job for branch pipelines first, then only for merge request pipelines after the merge request is created (skipping the duplicate branch pipeline). See the related issue for more details.

Also, we don’t recommend mixing only/except jobs with rules jobs in the same pipeline. It may not cause YAML errors, but debugging the exact execution behavior can be complex due to the different default behaviors of only/except and rules.

rules:if

rules:if clauses determine whether or not jobs are added to a pipeline by evaluating a simple if statement. If the if statement is true, the job is either included or excluded from a pipeline. In plain English, if rules can be interpreted as one of:

• “If this rule evaluates to true, add the job” (default).
• “If this rule evaluates to true, do not add the job” (by adding when: never).

rules:if differs slightly from only:variables by accepting only a single expression string per rule, rather than an array of them. Any set of expressions to be evaluated can be conjoined into a single expression by using && or ||, and use the variable matching syntax.

if: clauses are evaluated based on the values of predefined environment variables or custom environment variables.

For example:

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_MERGE_REQUEST_SOURCE_BRANCH_NAME =~ /^feature/ && $CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "master"'
      when: always
    - if: '$CI_MERGE_REQUEST_SOURCE_BRANCH_NAME =~ /^feature/'
      when: manual
      allow_failure: true
    - if: '$CI_MERGE_REQUEST_SOURCE_BRANCH_NAME' # Checking for the presence of a variable is possible

Some details regarding the logic that determines the when for the job:

• If none of the provided rules match, the job is set to when: never and is not included in the pipeline.
• A rule without any conditional clause, such as a when or allow_failure rule without if or changes, always matches, and is always used if reached.
• If a rule matches and has no when defined, the rule uses the when defined for the job, which defaults to on_success if not defined.
• You can define when once per rule, or once at the job-level, which applies to all rules. You can’t mix when at the job-level with when in rules.

Common if clauses for rules

For behavior similar to the only/except keywords, you can check the value of the $CI_PIPELINE_SOURCE variable:

For example:

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_PIPELINE_SOURCE == "schedule"'
      when: manual
      allow_failure: true
    - if: '$CI_PIPELINE_SOURCE == "push"'

This example runs the job as a manual job in scheduled pipelines or in push pipelines (to branches or tags), with when: on_success (default). It does not add the job to any other pipeline type.

Another example:

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
    - if: '$CI_PIPELINE_SOURCE == "schedule"'

This example runs the job as a when: on_success job in merge request pipelines and scheduled pipelines. It does not run in any other pipeline type.

Other commonly used variables for if clauses:

• if: $CI_COMMIT_TAG: If changes are pushed for a tag.
• if: $CI_COMMIT_BRANCH: If changes are pushed to any branch.
• if: '$CI_COMMIT_BRANCH == "master"': If changes are pushed to master.
• if: '$CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH': If changes are pushed to the default branch (usually master). Useful if reusing the same configuration in multiple projects with potentially different default branches.
• if: '$CI_COMMIT_BRANCH =~ /regex-expression/': If the commit branch matches a regular expression.
• if: '$CUSTOM_VARIABLE !~ /regex-expression/': If the custom variable CUSTOM_VARIABLE does not match a regular expression.
• if: '$CUSTOM_VARIABLE == "value1"': If the custom variable CUSTOM_VARIABLE is exactly value1.

To avoid running pipelines when a branch is created without any changes, check the value of $CI_COMMIT_BEFORE_SHA. It has a value of 0000000000000000000000000000000000000000:

• In branches with no commits.
• In tag pipelines and scheduled pipelines. You should define rules very narrowly if you don’t want to skip these.

To skip pipelines on all empty branches, but also tags and schedules:

rules:
  - if: $CI_COMMIT_BEFORE_SHA == '0000000000000000000000000000000000000000'
    when: never

To skip branch pipelines when the branch is empty:

rules:
  - if: $CI_COMMIT_BRANCH && $CI_COMMIT_BEFORE_SHA != '0000000000000000000000000000000000000000'

rules:changes

To determine if jobs should be added to a pipeline, rules: changes clauses check the files changed by Git push events.

rules: changes works exactly the same way as only: changes and except: changes, accepting an array of paths. Similarly, it always returns true if there is no Git push event. It should only be used for branch pipelines or merge request pipelines.

For example:

workflow:
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'

docker build:
  script: docker build -t my-image:$CI_COMMIT_REF_SLUG .
  rules:
    - changes:
        - Dockerfile
      when: manual
      allow_failure: true

In this example:

• workflow: rules allows only pipelines for merge requests for all jobs.
• If Dockerfile has changed, add the job to the pipeline as a manual job, and allow the pipeline to continue running even if the job is not triggered (allow_failure: true).
• If Dockerfile has not changed, do not add job to any pipeline (same as when: never).

To implement a rule similar to except: changes, use when: never.

rules:exists

exists accepts an array of paths and will match if any of these paths exist as files in the repository.

For example:

job:
  script: docker build -t my-image:$CI_COMMIT_REF_SLUG .
  rules:
    - exists:
        - Dockerfile

You can also use glob patterns to match multiple files in any directory within the repository.

For example:

job:
  script: bundle exec rspec
  rules:
    - exists:
        - spec/**.rb

rules:allow_failure

You can use allow_failure: true within rules: to allow a job to fail, or a manual job to wait for action, without stopping the pipeline itself. All jobs using rules: default to allow_failure: false if allow_failure: is not defined.

The rule-level rules:allow_failure option overrides the job-level allow_failure option, and is only applied when the job is triggered by the particular rule.

job:
  script: "echo Hello, Rules!"
  rules:
    - if: '$CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "master"'
      when: manual
      allow_failure: true

In this example, if the first rule matches, then the job will have when: manual and allow_failure: true.

Complex rule clauses

To conjoin if, changes, and exists clauses with an AND, use them in the same rule.

In the following example:

• We run the job manually if Dockerfile or any file in docker/scripts/ has changed AND $VAR == "string value".
• Otherwise, the job won’t be included in the pipeline.

docker build:
  script: docker build -t my-image:$CI_COMMIT_REF_SLUG .
  rules:
    - if: '$VAR == "string value"'
      changes: # Will include the job and set to when:manual if any of the follow paths match a modified file.
        - Dockerfile
        - docker/scripts/*
      when: manual
  # - when: never would be redundant here, this is implied any time rules are listed.

Keywords such as branches or refs that are currently available for only/except are not yet available in rules as they are being individually considered for their usage and behavior in this context. Future keyword improvements are being discussed in our epic for improving rules, where anyone can add suggestions or requests.

only/except (basic)

only and except are two parameters that set a job policy to limit when jobs are created:

1. only defines the names of branches and tags for which the job will run.
2. except defines the names of branches and tags for which the job will not run.

There are a few rules that apply to the usage of job policy:

• only and except are inclusive. If both only and except are defined in a job specification, the ref is filtered by only and except.
• only and except allow the use of regular expressions (supported regexp syntax).
• only and except allow to specify a repository path to filter jobs for forks.

In addition, only and except allow the use of special keywords:

In the example below, job will run only for refs that start with issue-, whereas all branches will be skipped:

job:
  # use regexp
  only:
    - /^issue-.*$/
  # use special keyword
  except:
    - branches

Pattern matching is case-sensitive by default. Use i flag modifier, like /pattern/i to make a pattern case-insensitive:

job:
  # use regexp
  only:
    - /^issue-.*$/i
  # use special keyword
  except:
    - branches

In this example, job will run only for refs that are tagged, or if a build is explicitly requested via an API trigger or a Pipeline Schedule:

job:
  # use special keywords
  only:
    - tags
    - triggers
    - schedules

The repository path can be used to have jobs executed only for the parent repository and not forks:

job:
  only:
    - branches@gitlab-org/gitlab
  except:
    - master@gitlab-org/gitlab
    - /^release/.*$/@gitlab-org/gitlab

The above example will run job for all branches on gitlab-org/gitlab, except master and those with names prefixed with release/.

If a job does not have an only rule, only: ['branches', 'tags'] is set by default. If it does not have an except rule, it’s empty.

For example,

job:
  script: echo 'test'

is translated to:

job:
  script: echo 'test'
  only: ['branches', 'tags']

Regular expressions

Because @ is used to denote the beginning of a ref’s repository path, matching a ref name containing the @ character in a regular expression requires the use of the hex character code match \x40.

Only the tag or branch name can be matched by a regular expression. The repository path, if given, is always matched literally.

If a regular expression shall be used to match the tag or branch name, the entire ref name part of the pattern has to be a regular expression, and must be surrounded by /. (With regular expression flags appended after the closing /.) So issue-/.*/ won’t work to match all tag names or branch names that begin with issue-.

Supported only/except regexp syntax

In GitLab 11.9.4, GitLab begun internally converting regexp used in only and except parameters to RE2.

This means that only subset of features provided by Ruby Regexp is supported. RE2 limits the set of features provided due to computational complexity, which means some features became unavailable in GitLab 11.9.4. For example, negative lookaheads.

For GitLab versions from 11.9.7 and up to GitLab 12.0, GitLab provides a feature flag that can be enabled by administrators that allows users to use unsafe regexp syntax. This brings compatibility with previously allowed syntax version and allows users to gracefully migrate to the new syntax.

Feature.enable(:allow_unsafe_ruby_regexp)

only/except (advanced)

GitLab supports both simple and complex strategies, so it’s possible to use an array and a hash configuration scheme.

Four keys are available:

• refs
• variables
• changes
• kubernetes

If you use multiple keys under only or except, the keys will be evaluated as a single conjoined expression. That is:

• only: means “include this job if all of the conditions match”.
• except: means “exclude this job if any of the conditions match”.

With only, individual keys are logically joined by an AND:

(any of refs) AND (any of variables) AND (any of changes) AND (if Kubernetes is active)

In the example below, the test job will only be created when all of the following are true:

• The pipeline has been scheduled or runs for master.
• The variables keyword matches.
• The kubernetes service is active on the project.

test:
  script: npm run test
  only:
    refs:
      - master
      - schedules
    variables:
      - $CI_COMMIT_MESSAGE =~ /run-end-to-end-tests/
    kubernetes: active

except is implemented as a negation of this complete expression:

NOT((any of refs) AND (any of variables) AND (any of changes) AND (if Kubernetes is active))

This means the keys are treated as if joined by an OR. This relationship could be described as:

(any of refs) OR (any of variables) OR (any of changes) OR (if Kubernetes is active)

In the example below, the test job will not be created when any of the following are true:

• The pipeline runs for the master.
• There are changes to the README.md file in the root directory of the repository.

test:
  script: npm run test
  except:
    refs:
      - master
    changes:
      - "README.md"

only:refs/except:refs

The refs strategy can take the same values as the simplified only/except configuration.

In the example below, the deploy job is going to be created only when the pipeline has been scheduled or runs for the master branch:

deploy:
  only:
    refs:
      - master
      - schedules

only:kubernetes/except:kubernetes

The kubernetes strategy accepts only the active keyword.

In the example below, the deploy job is going to be created only when the Kubernetes service is active in the project:

deploy:
  only:
    kubernetes: active

only:variables/except:variables

The variables keyword is used to define variables expressions. In other words, you can use predefined variables / project / group or environment-scoped variables to define an expression GitLab is going to evaluate in order to decide whether a job should be created or not.

Examples of using variables expressions:

deploy:
  script: cap staging deploy
  only:
    refs:
      - branches
    variables:
      - $RELEASE == "staging"
      - $STAGING

Another use case is excluding jobs depending on a commit message:

end-to-end:
  script: rake test:end-to-end
  except:
    variables:
      - $CI_COMMIT_MESSAGE =~ /skip-end-to-end-tests/

Learn more about variables expressions.

only:changes/except:changes

Using the changes keyword with only or except makes it possible to define if a job should be created based on files modified by a Git push event.

This means the only:changes policy is useful for pipelines where:

• $CI_PIPELINE_SOURCE == 'push'
• $CI_PIPELINE_SOURCE == 'merge_request_event'
• $CI_PIPELINE_SOURCE == 'external_pull_request_event'

If there is no Git push event, such as for pipelines with sources other than the three above, changes can’t determine if a given file is new or old, and will always return true.

A basic example of using only: changes:

docker build:
  script: docker build -t my-image:$CI_COMMIT_REF_SLUG .
  only:
    changes:
      - Dockerfile
      - docker/scripts/*
      - dockerfiles/**/*
      - more_scripts/*.{rb,py,sh}

In the scenario above, when pushing commits to an existing branch in GitLab, it creates and triggers the docker build job, provided that one of the commits contains changes to any of the following:

• The Dockerfile file.
• Any of the files inside docker/scripts/ directory.
• Any of the files and subdirectories inside the dockerfiles directory.
• Any of the files with rb, py, sh extensions inside the more_scripts directory.

You can also use glob patterns to match multiple files in either the root directory of the repository, or in any directory within the repository, but they must be wrapped in double quotes or GitLab will fail to parse the .gitlab-ci.yml. For example:

test:
  script: npm run test
  only:
    changes:
      - "*.json"
      - "**/*.sql"

The following example will skip the build job if a change is detected in any file in the root directory of the repository with a .md extension:

build:
  script: npm run build
  except:
    changes:
      - "*.md"

Using only:changes with pipelines for merge requests

With pipelines for merge requests, it’s possible to define a job to be created based on files modified in a merge request.

In order to deduce the correct base SHA of the source branch, we recommend combining this keyword with only: [merge_requests]. This way, file differences are correctly calculated from any further commits, thus all changes in the merge requests are properly tested in pipelines.

For example:

docker build service one:
  script: docker build -t my-service-one-image:$CI_COMMIT_REF_SLUG .
  only:
    refs:
      - merge_requests
    changes:
      - Dockerfile
      - service-one/**/*

In the scenario above, if a merge request is created or updated that changes either files in service-one directory or the Dockerfile, GitLab creates and triggers the docker build service one job.

Note that if pipelines for merge requests is combined with only: [change], but only: [merge_requests] is omitted, there could be unwanted behavior.

For example:

docker build service one:
  script: docker build -t my-service-one-image:$CI_COMMIT_REF_SLUG .
  only:
    changes:
      - Dockerfile
      - service-one/**/*

In the example above, a pipeline could fail due to changes to a file in service-one/**/*. A later commit could then be pushed that does not include any changes to this file, but includes changes to the Dockerfile, and this pipeline could pass because it’s only testing the changes to the Dockerfile. GitLab checks the most recent pipeline, that passed, and will show the merge request as mergeable, despite the earlier failed pipeline caused by a change that was not yet corrected.

With this configuration, care must be taken to check that the most recent pipeline properly corrected any failures from previous pipelines.

Using only:changes without pipelines for merge requests

Without pipelines for merge requests, pipelines run on branches or tags that don’t have an explicit association with a merge request. In this case, a previous SHA is used to calculate the diff, which equivalent to git diff HEAD~. This could result in some unexpected behavior, including:

• When pushing a new branch or a new tag to GitLab, the policy always evaluates to true.
• When pushing a new commit, the changed files are calculated using the previous commit as the base SHA.

Using only:changes with scheduled pipelines

only:changes always evaluates as “true” in Scheduled pipelines. All files are considered to have “changed” when a scheduled pipeline runs.

needs

The needs: keyword enables executing jobs out-of-order, allowing you to implement a directed acyclic graph in your .gitlab-ci.yml.

This lets you run some jobs without waiting for other ones, disregarding stage ordering so you can have multiple stages running concurrently.

Let’s consider the following example:

linux:build:
  stage: build

mac:build:
  stage: build

lint:
  stage: test
  needs: []

linux:rspec:
  stage: test
  needs: ["linux:build"]

linux:rubocop:
  stage: test
  needs: ["linux:build"]

mac:rspec:
  stage: test
  needs: ["mac:build"]

mac:rubocop:
  stage: test
  needs: ["mac:build"]

production:
  stage: deploy

This example creates four paths of execution:

• Linter: the lint job will run immediately without waiting for the build stage to complete because it has no needs (needs: []).

• Linux path: the linux:rspec and linux:rubocop jobs will be run as soon as the linux:build job finishes without waiting for mac:build to finish.

• macOS path: the mac:rspec and mac:rubocop jobs will be run as soon as the mac:build job finishes, without waiting for linux:build to finish.

• The production job will be executed as soon as all previous jobs finish; in this case: linux:build, linux:rspec, linux:rubocop, mac:build, mac:rspec, mac:rubocop.

Requirements and limitations

• If needs: is set to point to a job that is not instantiated because of only/except rules or otherwise does not exist, the pipeline will be created with YAML error.
• The maximum number of jobs that a single job can need in the needs: array is limited:
  • For GitLab.com, the limit is ten. For more information, see our infrastructure issue.
  • For self-managed instances, the limit is:
    • 10, if the ci_plan_needs_size_limit feature flag is disabled (default).
    • 50, if the ci_plan_needs_size_limit feature flag is enabled. This limit can be changed.
• If needs: refers to a job that is marked as parallel:. the current job will depend on all parallel jobs created.
• needs: is similar to dependencies: in that it needs to use jobs from prior stages, meaning it’s impossible to create circular dependencies. Depending on jobs in the current stage is not possible either, but support is planned.
• Related to the above, stages must be explicitly defined for all jobs that have the keyword needs: or are referred to by one.

Changing the needs: job limit

The maximum number of jobs that can be defined within needs: defaults to 10.

To change this limit to 50 on a self-managed installation, a GitLab administrator with access to the GitLab Rails console can enable the :ci_plan_needs_size_limit feature flag:

Feature::enable(:ci_plan_needs_size_limit)

After the feature flag is enabled, you can choose a custom limit. For example, to set the limit to 100:

Plan.default.actual_limits.update!(ci_needs_size_limit: 100)

Artifact downloads with needs

When using needs, artifact downloads are controlled with artifacts: true (default) or artifacts: false.

Since GitLab 12.6, you can’t combine the dependencies keyword with needs to control artifact downloads in jobs. dependencies is still valid in jobs that do not use needs.

In the example below, the rspec job will download the build_job artifacts, while the rubocop job won’t:

build_job:
  stage: build
  artifacts:
    paths:
      - binaries/

rspec:
  stage: test
  needs:
    - job: build_job
      artifacts: true

rubocop:
  stage: test
  needs:
    - job: build_job
      artifacts: false

Additionally, in the three syntax examples below, the rspec job will download the artifacts from all three build_jobs, as artifacts is true for build_job_1, and will default to true for both build_job_2 and build_job_3.

rspec:
  needs:
    - job: build_job_1
      artifacts: true
    - job: build_job_2
    - build_job_3

Cross project artifact downloads with needs

needs can be used to download artifacts from up to five jobs in pipelines on other refs in the same project, or pipelines in different projects, groups and namespaces:

build_job:
  stage: build
  script:
    - ls -lhR
  needs:
    - project: namespace/group/project-name
      job: build-1
      ref: master
      artifacts: true

build_job will download the artifacts from the latest successful build-1 job on the master branch in the group/project-name project. If the project is in the same group or namespace, you can omit them from the project: key. For example, project: group/project-name or project: project-name.

The user running the pipeline must have at least reporter access to the group or project, or the group/project must have public visibility.

Artifact downloads between pipelines in the same project

needs can be used to download artifacts from different pipelines in the current project by setting the project keyword as the current project’s name, and specifying a ref. In the example below, build_job will download the artifacts for the latest successful build-1 job with the other-ref ref:

build_job:
  stage: build
  script:
    - ls -lhR
  needs:
    - project: group/same-project-name
      job: build-1
      ref: other-ref
      artifacts: true

tags

tags is used to select specific Runners from the list of all Runners that are allowed to run this project.

During the registration of a Runner, you can specify the Runner’s tags, for example ruby, postgres, development.

tags allow you to run jobs with Runners that have the specified tags assigned to them:

job:
  tags:
    - ruby
    - postgres

The specification above, will make sure that job is built by a Runner that has both ruby AND postgres tags defined.

Tags are also a great way to run different jobs on different platforms, for example, given an OS X Runner with tag osx and Windows Runner with tag windows, the following jobs run on respective platforms:

windows job:
  stage:
    - build
  tags:
    - windows
  script:
    - echo Hello, %USERNAME%!

osx job:
  stage:
    - build
  tags:
    - osx
  script:
    - echo "Hello, $USER!"

allow_failure

allow_failure allows a job to fail without impacting the rest of the CI suite. The default value is false, except for manual jobs using the when: manual syntax, unless using rules: syntax, where all jobs default to false, including when: manual jobs.

When enabled and the job fails, the job will show an orange warning in the UI. However, the logical flow of the pipeline will consider the job a success/passed, and is not blocked.

Assuming all other jobs are successful, the job’s stage and its pipeline will show the same orange warning. However, the associated commit will be marked “passed”, without warnings.

In the example below, job1 and job2 will run in parallel, but if job1 fails, it won’t stop the next stage from running, since it’s marked with allow_failure: true:

job1:
  stage: test
  script:
    - execute_script_that_will_fail
  allow_failure: true

job2:
  stage: test
  script:
    - execute_script_that_will_succeed

job3:
  stage: deploy
  script:
    - deploy_to_staging

when

when is used to implement jobs that are run in case of failure or despite the failure.

when can be set to one of the following values:

1. on_success - execute job only when all jobs from prior stages succeed (or are considered succeeding because they are marked allow_failure). This is the default.
2. on_failure - execute job only when at least one job from prior stages fails.
3. always - execute job regardless of the status of jobs from prior stages.
4. manual - execute job manually (added in GitLab 8.10). Read about manual actions below.
5. delayed - execute job after a certain period (added in GitLab 11.14). Read about delayed actions below.
6. never:
   • With rules, don’t execute job.
   • With workflow:rules, don’t run pipeline.

For example:

stages:
  - build
  - cleanup_build
  - test
  - deploy
  - cleanup

build_job:
  stage: build
  script:
    - make build

cleanup_build_job:
  stage: cleanup_build
  script:
    - cleanup build when failed
  when: on_failure

test_job:
  stage: test
  script:
    - make test

deploy_job:
  stage: deploy
  script:
    - make deploy
  when: manual

cleanup_job:
  stage: cleanup
  script:
    - cleanup after jobs
  when: always

The above script will:

1. Execute cleanup_build_job only when build_job fails.
2. Always execute cleanup_job as the last step in pipeline regardless of success or failure.
3. Allow you to manually execute deploy_job from GitLab’s UI.

when:manual

Manual actions are a special type of job that are not executed automatically, they need to be explicitly started by a user. An example usage of manual actions would be a deployment to a production environment. Manual actions can be started from the pipeline, job, environment, and deployment views. Read more at the environments documentation.

Manual actions can be either optional or blocking. Blocking manual actions will block the execution of the pipeline at the stage this action is defined in. It’s possible to resume execution of the pipeline when someone executes a blocking manual action by clicking a play button.

When a pipeline is blocked, it won’t be merged if Merge When Pipeline Succeeds is set. Blocked pipelines also do have a special status, called manual. When the when:manual syntax is used, manual actions are non-blocking by default. If you want to make manual action blocking, it’s necessary to add allow_failure: false to the job’s definition in .gitlab-ci.yml.

Optional manual actions have allow_failure: true set by default and their Statuses don’t contribute to the overall pipeline status. So, if a manual action fails, the pipeline will eventually succeed.

Manual actions are considered to be write actions, so permissions for protected branches are used when a user wants to trigger an action. In other words, in order to trigger a manual action assigned to a branch that the pipeline is running for, the user needs to have the ability to merge to this branch. It’s possible to use protected environments to more strictly protect manual deployments from being run by unauthorized users.

Protecting manual jobs

It’s possible to use protected environments to define a precise list of users authorized to run a manual job. By allowing only users associated with a protected environment to trigger manual jobs, it’s possible to implement some special use cases, such as:

• More precisely limiting who can deploy to an environment.
• Enabling a pipeline to be blocked until an approved user “approves” it.

To do this, you must:

1. Add an environment to the job. For example:

   deploy_prod:
     stage: deploy
     script:
       - echo "Deploy to production server"
     environment:
       name: production
       url: https://example.com
     when: manual
     only:
       - master
   

2. In the protected environments settings, select the environment (production in the example above) and add the users, roles or groups that are authorized to trigger the manual job to the Allowed to Deploy list. Only those in this list will be able to trigger this manual job, as well as GitLab administrators who are always able to use protected environments.

Additionally, if a manual job is defined as blocking by adding allow_failure: false, the next stages of the pipeline won’t run until the manual job is triggered. This can be used as a way to have a defined list of users allowed to “approve” later pipeline stages by triggering the blocking manual job.

when:delayed

Delayed job are for executing scripts after a certain period. This is useful if you want to avoid jobs entering pending state immediately.

You can set the period with start_in key. The value of start_in key is an elapsed time in seconds, unless a unit is provided. start_in key must be less than or equal to one week. Examples of valid values include:

• '5'
• 10 seconds
• 30 minutes
• 1 day
• 1 week

When there is a delayed job in a stage, the pipeline won’t progress until the delayed job has finished. This means this keyword can also be used for inserting delays between different stages.

The timer of a delayed job starts immediately after the previous stage has completed. Similar to other types of jobs, a delayed job’s timer won’t start unless the previous stage passed.

The following example creates a job named timed rollout 10% that is executed 30 minutes after the previous stage has completed:

timed rollout 10%:
  stage: deploy
  script: echo 'Rolling out 10% ...'
  when: delayed
  start_in: 30 minutes

You can stop the active timer of a delayed job by clicking the (Unschedule) button. This job will never be executed in the future unless you execute the job manually.

You can start a delayed job immediately by clicking the Play button. GitLab Runner will pick your job soon and start the job.

environment

environment is used to define that a job deploys to a specific environment. If environment is specified and no environment under that name exists, a new one will be created automatically.

In its simplest form, the environment keyword can be defined like:

deploy to production:
  stage: deploy
  script: git push production HEAD:master
  environment: production

In the above example, the deploy to production job will be marked as doing a deployment to the production environment.

environment:name

The environment name can contain:

• letters
• digits
• spaces
• -
• _
• /
• $
• {
• }

Common names are qa, staging, and production, but you can use whatever name works with your workflow.

Instead of defining the name of the environment right after the environment keyword, it’s also possible to define it as a separate value. For that, use the name keyword under environment:

deploy to production:
  stage: deploy
  script: git push production HEAD:master
  environment:
    name: production

environment:url

This is an optional value that when set, it exposes buttons in various places in GitLab which when clicked take you to the defined URL.

In the example below, if the job finishes successfully, it will create buttons in the merge requests and in the environments/deployments pages which will point to https://prod.example.com.

deploy to production:
  stage: deploy
  script: git push production HEAD:master
  environment:
    name: production
    url: https://prod.example.com

environment:on_stop

Closing (stopping) environments can be achieved with the on_stop keyword defined under environment. It declares a different job that runs in order to close the environment.

Read the environment:action section for an example.

environment:action

The action keyword can be used to specify jobs that prepare, start, or stop environments.

Take for instance:

review_app:
  stage: deploy
  script: make deploy-app
  environment:
    name: review/$CI_COMMIT_REF_NAME
    url: https://$CI_ENVIRONMENT_SLUG.example.com
    on_stop: stop_review_app

stop_review_app:
  stage: deploy
  variables:
    GIT_STRATEGY: none
  script: make delete-app
  when: manual
  environment:
    name: review/$CI_COMMIT_REF_NAME
    action: stop

In the above example we set up the review_app job to deploy to the review environment, and we also defined a new stop_review_app job under on_stop. Once the review_app job is successfully finished, it will trigger the stop_review_app job based on what is defined under when. In this case we set it up to manual so it will need a manual action via GitLab’s web interface in order to run.

Also in the example, GIT_STRATEGY is set to none so that GitLab Runner won’t try to check out the code after the branch is deleted when the stop_review_app job is automatically triggered.

The stop_review_app job is required to have the following keywords defined:

• when - reference
• environment:name
• environment:action

Additionally, both jobs should have matching rules or only/except configuration. In the example above, if the configuration is not identical, the stop_review_app job might not be included in all pipelines that include the review_app job, and it will not be possible to trigger the action: stop to stop the environment automatically.

environment:auto_stop_in

The auto_stop_in keyword is for specifying life period of the environment, that when expired, GitLab automatically stops them.

For example,

review_app:
  script: deploy-review-app
  environment:
    name: review/$CI_COMMIT_REF_NAME
    auto_stop_in: 1 day

When review_app job is executed and a review app is created, a life period of the environment is set to 1 day.

For more information, see the environments auto-stop documentation

environment:kubernetes

The kubernetes block is used to configure deployments to a Kubernetes cluster that is associated with your project.

For example:

deploy:
  stage: deploy
  script: make deploy-app
  environment:
    name: production
    kubernetes:
      namespace: production

This will set up the deploy job to deploy to the production environment, using the production Kubernetes namespace.

For more information, see Available settings for kubernetes.

Dynamic environments

For example:

deploy as review app:
  stage: deploy
  script: make deploy
  environment:
    name: review/$CI_COMMIT_REF_NAME
    url: https://$CI_ENVIRONMENT_SLUG.example.com/

The deploy as review app job will be marked as deployment to dynamically create the review/$CI_COMMIT_REF_NAME environment, where $CI_COMMIT_REF_NAME is an environment variable set by the Runner. The $CI_ENVIRONMENT_SLUG variable is based on the environment name, but suitable for inclusion in URLs. In this case, if the deploy as review app job was run in a branch named pow, this environment would be accessible with an URL like https://review-pow.example.com/.

This of course implies that the underlying server which hosts the application is properly configured.

The common use case is to create dynamic environments for branches and use them as Review Apps. You can see a simple example using Review Apps at https://gitlab.com/gitlab-examples/review-apps-nginx/.

cache

cache is used to specify a list of files and directories which should be cached between jobs. You can only use paths that are within the local working copy.

If cache is defined outside the scope of jobs, it means it’s set globally and all jobs will use that definition.

cache:paths

Use the paths directive to choose which files or directories will be cached. Paths are relative to the project directory ($CI_PROJECT_DIR) and can’t directly link outside it. Wildcards can be used that follow the glob patterns and:

• In GitLab Runner 13.0 and later, doublestar.Glob.
• In GitLab Runner 12.10 and earlier, filepath.Match.

Cache all files in binaries that end in .apk and the .config file:

rspec:
  script: test
  cache:
    paths:
      - binaries/*.apk
      - .config

Locally defined cache overrides globally defined options. The following rspec job will cache only binaries/:

cache:
  paths:
    - my/files

rspec:
  script: test
  cache:
    key: rspec
    paths:
      - binaries/

Note that since cache is shared between jobs, if you’re using different paths for different jobs, you should also set a different cache:key otherwise cache content can be overwritten.

cache:key

Since the cache is shared between jobs, if you’re using different paths for different jobs, you should also set a different cache:key otherwise cache content can be overwritten.

The key directive allows you to define the affinity of caching between jobs, allowing to have a single cache for all jobs, cache per-job, cache per-branch or any other way that fits your workflow. This way, you can fine tune caching, allowing you to cache data between different jobs or even different branches.

The cache:key variable can use any of the predefined variables, and the default key, if not set, is just literal default which means everything is shared between pipelines and jobs by default, starting from GitLab 9.0.

For example, to enable per-branch caching:

cache:
  key: "$CI_COMMIT_REF_SLUG"
  paths:
    - binaries/

If you use Windows Batch to run your shell scripts you need to replace $ with %:

cache:
  key: "%CI_COMMIT_REF_SLUG%"
  paths:
    - binaries/

cache:key:files

The cache:key:files keyword extends the cache:key functionality by making it easier to reuse some caches, and rebuild them less often, which will speed up subsequent pipeline runs.

When you include cache:key:files, you must also list the project files that will be used to generate the key, up to a maximum of two files. The cache key will be a SHA checksum computed from the most recent commits (up to two, if two files are listed) that changed the given files. If neither file was changed in any commits, the fallback key will be default.

cache:
  key:
    files:
      - Gemfile.lock
      - package.json
  paths:
    - vendor/ruby
    - node_modules

In this example we’re creating a cache for Ruby and Node.js dependencies that is tied to current versions of the Gemfile.lock and package.json files. Whenever one of these files changes, a new cache key is computed and a new cache is created. Any future job runs using the same Gemfile.lock and package.json with cache:key:files will use the new cache, instead of rebuilding the dependencies.

cache:key:prefix

The prefix parameter adds extra functionality to key:files by allowing the key to be composed of the given prefix combined with the SHA computed for cache:key:files. For example, adding a prefix of test, will cause keys to look like: test-feef9576d21ee9b6a32e30c5c79d0a0ceb68d1e5. If neither file was changed in any commits, the prefix is added to default, so the key in the example would be test-default.

Like cache:key, prefix can use any of the predefined variables, but the following are not allowed:

• the / character (or the equivalent URI-encoded %2F)
• a value made only of . (or the equivalent URI-encoded %2E)

cache:
  key:
    files:
      - Gemfile.lock
    prefix: ${CI_JOB_NAME}
  paths:
    - vendor/ruby

rspec:
  script:
    - bundle exec rspec

For example, adding a prefix of $CI_JOB_NAME will cause the key to look like: rspec-feef9576d21ee9b6a32e30c5c79d0a0ceb68d1e5 and the job cache is shared across different branches. If a branch changes Gemfile.lock, that branch will have a new SHA checksum for cache:key:files. A new cache key will be generated, and a new cache will be created for that key. If Gemfile.lock is not found, the prefix is added to default, so the key in the example would be rspec-default.

cache:untracked

Set untracked: true to cache all files that are untracked in your Git repository:

rspec:
  script: test
  cache:
    untracked: true

Cache all Git untracked files and files in binaries:

rspec:
  script: test
  cache:
    untracked: true
    paths:
      - binaries/

cache:policy

The default behavior of a caching job is to download the files at the start of execution, and to re-upload them at the end. This allows any changes made by the job to be persisted for future runs, and is known as the pull-push cache policy.

If you know the job does not alter the cached files, you can skip the upload step by setting policy: pull in the job specification. Typically, this would be twinned with an ordinary cache job at an earlier stage to ensure the cache is updated from time to time:

stages:
  - setup
  - test

prepare:
  stage: setup
  cache:
    key: gems
    paths:
      - vendor/bundle
  script:
    - bundle install --deployment

rspec:
  stage: test
  cache:
    key: gems
    paths:
      - vendor/bundle
    policy: pull
  script:
    - bundle exec rspec ...

This helps to speed up job execution and reduce load on the cache server, especially when you have a large number of cache-using jobs executing in parallel.

Additionally, if you have a job that unconditionally recreates the cache without reference to its previous contents, you can use policy: push in that job to skip the download step.

artifacts

artifacts is used to specify a list of files and directories which should be attached to the job when it succeeds, fails, or always.

The artifacts will be sent to GitLab after the job finishes and will be available for download in the GitLab UI.

Read more about artifacts.

artifacts:paths

Paths are relative to the project directory ($CI_PROJECT_DIR) and can’t directly link outside it. Wildcards can be used that follow the glob patterns and:

• In GitLab Runner 13.0 and later, doublestar.Glob.
• In GitLab Runner 12.10 and earlier, filepath.Match.

To restrict which jobs a specific job will fetch artifacts from, see dependencies.

Send all files in binaries and .config:

artifacts:
  paths:
    - binaries/
    - .config

To disable artifact passing, define the job with empty dependencies:

job:
  stage: build
  script: make build
  dependencies: []

You may want to create artifacts only for tagged releases to avoid filling the build server storage with temporary build artifacts.

Create artifacts only for tags (default-job won’t create artifacts):

default-job:
  script:
    - mvn test -U
  except:
    - tags

release-job:
  script:
    - mvn package -U
  artifacts:
    paths:
      - target/*.war
  only:
    - tags

You can use wildcards for directories too. For example, if you want to get all the files inside the directories that end with xyz:

job:
  artifacts:
    paths:
      - path/*xyz/*

artifacts:exclude

exclude makes it possible to prevent files from being added to an artifacts archive.

Similar to artifacts:paths, exclude paths are relative to the project directory. Wildcards can be used that follow the glob patterns and filepath.Match.

For example, to store all files in binaries/, but not *.o files located in subdirectories of binaries/:

artifacts:
  paths:
    - binaries/
  exclude:
    - binaries/**/*.o

Files matched by artifacts:untracked can be excluded using artifacts:exclude too.

artifacts:expose_as

The expose_as keyword can be used to expose job artifacts in the merge request UI.

For example, to match a single file:

test:
  script: [ "echo 'test' > file.txt" ]
  artifacts:
    expose_as: 'artifact 1'
    paths: ['file.txt']

With this configuration, GitLab will add a link artifact 1 to the relevant merge request that points to file1.txt.

An example that will match an entire directory:

test:
  script: [ "mkdir test && echo 'test' > test/file.txt" ]
  artifacts:
    expose_as: 'artifact 1'
    paths: ['test/']

Note the following:

• Artifacts do not display in the merge request UI when using variables to define the artifacts:paths.
• A maximum of 10 job artifacts per merge request can be exposed.
• Glob patterns are unsupported.
• If a directory is specified, the link will be to the job artifacts browser if there is more than one file in the directory.
• For exposed single file artifacts with .html, .htm, .txt, .json, .xml, and .log extensions, if GitLab Pages is:
  • Enabled, GitLab will automatically render the artifact.
  • Not enabled, you will see the file in the artifacts browser.

artifacts:name

The name directive allows you to define the name of the created artifacts archive. That way, you can have a unique name for every archive which could be useful when you’d like to download the archive from GitLab. The artifacts:name variable can make use of any of the predefined variables. The default name is artifacts, which becomes artifacts.zip when downloaded.

To create an archive with a name of the current job:

job:
  artifacts:
    name: "$CI_JOB_NAME"
    paths:
      - binaries/

To create an archive with a name of the current branch or tag including only the binaries directory:

job:
  artifacts:
    name: "$CI_COMMIT_REF_NAME"
    paths:
      - binaries/

To create an archive with a name of the current job and the current branch or tag including only the binaries directory:

job:
  artifacts:
    name: "$CI_JOB_NAME-$CI_COMMIT_REF_NAME"
    paths:
      - binaries/

To create an archive with a name of the current stage and branch name:

job:
  artifacts:
    name: "$CI_JOB_STAGE-$CI_COMMIT_REF_NAME"
    paths:
      - binaries/

If you use Windows Batch to run your shell scripts you need to replace $ with %:

job:
  artifacts:
    name: "%CI_JOB_STAGE%-%CI_COMMIT_REF_NAME%"
    paths:
      - binaries/

If you use Windows PowerShell to run your shell scripts you need to replace $ with $env::

job:
  artifacts:
    name: "$env:CI_JOB_STAGE-$env:CI_COMMIT_REF_NAME"
    paths:
      - binaries/

artifacts:untracked

artifacts:untracked is used to add all Git untracked files as artifacts (along to the paths defined in artifacts:paths).

Send all Git untracked files:

artifacts:
  untracked: true

Send all Git untracked files and files in binaries:

artifacts:
  untracked: true
  paths:
    - binaries/

Send all untracked files but exclude *.txt:

artifacts:
  untracked: true
  exclude:
    - *.txt

artifacts:when

artifacts:when is used to upload artifacts on job failure or despite the failure.

artifacts:when can be set to one of the following values:

1. on_success - upload artifacts only when the job succeeds. This is the default.
2. on_failure - upload artifacts only when the job fails.
3. always - upload artifacts regardless of the job status.

To upload artifacts only when job fails:

job:
  artifacts:
    when: on_failure

artifacts:expire_in

expire_in allows you to specify how long artifacts should live before they expire and are therefore deleted, counting from the time they are uploaded and stored on GitLab. If the expiry time is not defined, it defaults to the instance wide setting (30 days by default).

To override the expiration time and keep artifacts forever:

• Use the Keep button on the job page.
• Set the value of expire_in to never. Available in GitLab 13.3 and later.

After their expiry, artifacts are deleted hourly by default (via a cron job), and are not accessible anymore.

The value of expire_in is an elapsed time in seconds, unless a unit is provided. Examples of valid values:

• 42
• 3 mins 4 sec
• 2 hrs 20 min
• 2h20min
• 6 mos 1 day
• 47 yrs 6 mos and 4d
• 3 weeks and 2 days
• never

To expire artifacts 1 week after being uploaded:

job:
  artifacts:
    expire_in: 1 week

artifacts:reports

The artifacts:reports keyword is used for collecting test reports, code quality reports, and security reports from jobs. It also exposes these reports in GitLab’s UI (merge requests, pipeline views, and security dashboards).

These are the available report types:

dependencies

By default, all artifacts from all previous stages are passed, but you can use the dependencies parameter to define a limited list of jobs (or no jobs) to fetch artifacts from.

To use this feature, define dependencies in context of the job and pass a list of all previous jobs from which the artifacts should be downloaded. You can only define jobs from stages that are executed before the current one. An error will be shown if you define jobs from the current stage or next ones. Defining an empty array will skip downloading any artifacts for that job. The status of the previous job is not considered when using dependencies, so if it failed or it’s a manual job that was not run, no error occurs.

In the following example, we define two jobs with artifacts, build:osx and build:linux. When the test:osx is executed, the artifacts from build:osx will be downloaded and extracted in the context of the build. The same happens for test:linux and artifacts from build:linux.

The job deploy will download artifacts from all previous jobs because of the stage precedence:

build:osx:
  stage: build
  script: make build:osx
  artifacts:
    paths:
      - binaries/

build:linux:
  stage: build
  script: make build:linux
  artifacts:
    paths:
      - binaries/

test:osx:
  stage: test
  script: make test:osx
  dependencies:
    - build:osx

test:linux:
  stage: test
  script: make test:linux
  dependencies:
    - build:linux

deploy:
  stage: deploy
  script: make deploy

When a dependent job will fail

If the artifacts of the job that is set as a dependency have been expired or erased, then the dependent job will fail.

coverage

coverage allows you to configure how code coverage will be extracted from the job output.

Regular expressions are the only valid kind of value expected here. So, using surrounding / is mandatory in order to consistently and explicitly represent a regular expression string. You must escape special characters if you want to match them literally.

A simple example:

job1:
  script: rspec
  coverage: '/Code coverage: \d+\.\d+/'

retry

retry allows you to configure how many times a job is going to be retried in case of a failure.

When a job fails and has retry configured, it’s going to be processed again up to the amount of times specified by the retry keyword.

If retry is set to 2, and a job succeeds in a second run (first retry), it won’t be retried again. retry value has to be a positive integer, equal or larger than 0, but lower or equal to 2 (two retries maximum, three runs in total).

A simple example to retry in all failure cases:

test:
  script: rspec
  retry: 2

By default, a job will be retried on all failure cases. To have a better control on which failures to retry, retry can be a hash with the following keys:

• max: The maximum number of retries.
• when: The failure cases to retry.

To retry only runner system failures at maximum two times:

test:
  script: rspec
  retry:
    max: 2
    when: runner_system_failure

If there is another failure, other than a runner system failure, the job will not be retried.

To retry on multiple failure cases, when can also be an array of failures:

test:
  script: rspec
  retry:
    max: 2
    when:
      - runner_system_failure
      - stuck_or_timeout_failure

Possible values for when are:

• always: Retry on any failure (default).
• unknown_failure: Retry when the failure reason is unknown.
• script_failure: Retry when the script failed.
• api_failure: Retry on API failure.
• stuck_or_timeout_failure: Retry when the job got stuck or timed out.
• runner_system_failure: Retry if there was a runner system failure (for example, job setup failed).
• missing_dependency_failure: Retry if a dependency was missing.
• runner_unsupported: Retry if the runner was unsupported.
• stale_schedule: Retry if a delayed job could not be executed.
• job_execution_timeout: Retry if the script exceeded the maximum execution time set for the job.
• archived_failure: Retry if the job is archived and can’t be run.
• unmet_prerequisites: Retry if the job failed to complete prerequisite tasks.
• scheduler_failure: Retry if the scheduler failed to assign the job to a runner.
• data_integrity_failure: Retry if there was a structural integrity problem detected.

You can specify the number of retry attempts for certain stages of job execution using variables.

timeout

timeout allows you to configure a timeout for a specific job. For example:

build:
  script: build.sh
  timeout: 3 hours 30 minutes

test:
  script: rspec
  timeout: 3h 30m

The job-level timeout can exceed the project-level timeout but can’t exceed the Runner-specific timeout.

parallel

parallel allows you to configure how many instances of a job to run in parallel. This value has to be greater than or equal to two (2) and less than or equal to 50.

This creates N instances of the same job that run in parallel. They are named sequentially from job_name 1/N to job_name N/N.

For every job, CI_NODE_INDEX and CI_NODE_TOTAL environment variables are set.

Marking a job to be run in parallel requires adding parallel to your configuration file. For example:

test:
  script: rspec
  parallel: 5

A simple example using Semaphore Test Boosters and RSpec to run some Ruby tests:

# Gemfile
source 'https://rubygems.org'

gem 'rspec'
gem 'semaphore_test_boosters'

test:
  parallel: 3
  script:
    - bundle
    - bundle exec rspec_booster --job $CI_NODE_INDEX/$CI_NODE_TOTAL

You can then navigate to the Jobs tab of a new pipeline build and see your RSpec job split into three separate jobs.

Parallel matrix jobs

matrix: allows you to configure different variables for jobs that are running in parallel. There can be from 2 to 50 jobs.

Every job gets the same CI_NODE_TOTAL environment variable value, and a unique CI_NODE_INDEX value.

deploystacks:
  stage: deploy
  script:
    - bin/deploy
  parallel:
    matrix:
      - PROVIDER: aws
        STACK:
          - monitoring
          - app1
          - app2
      - PROVIDER: ovh
        STACK: [monitoring, backup, app]
      - PROVIDER: [gcp, vultr]
        STACK: [data, processing]

This generates 10 parallel deploystacks jobs, each with different values for PROVIDER and STACK:

deploystacks 1/10 with PROVIDER=aws and STACK=monitoring
deploystacks 2/10 with PROVIDER=aws and STACK=app1
deploystacks 3/10 with PROVIDER=aws and STACK=app2
deploystacks 4/10 with PROVIDER=ovh and STACK=monitoring
deploystacks 5/10 with PROVIDER=ovh and STACK=backup
deploystacks 6/10 with PROVIDER=ovh and STACK=app
deploystacks 7/10 with PROVIDER=gcp and STACK=data
deploystacks 8/10 with PROVIDER=gcp and STACK=processing
deploystacks 9/10 with PROVIDER=vultr and STACK=data
deploystacks 10/10 with PROVIDER=vultr and STACK=processing

trigger

trigger allows you to define downstream pipeline trigger. When a job created from trigger definition is started by GitLab, a downstream pipeline gets created.

This keyword allows the creation of two different types of downstream pipelines:

• Multi-project pipelines
• Child pipelines

Since GitLab 13.2, you can see which job triggered a downstream pipeline by hovering your mouse cursor over the downstream pipeline job in the pipeline graph.

Simple trigger syntax for multi-project pipelines

The simplest way to configure a downstream trigger is to use trigger keyword with a full path to a downstream project:

rspec:
  stage: test
  script: bundle exec rspec

staging:
  stage: deploy
  trigger: my/deployment

Complex trigger syntax for multi-project pipelines

It’s possible to configure a branch name that GitLab will use to create a downstream pipeline with:

rspec:
  stage: test
  script: bundle exec rspec

staging:
  stage: deploy
  trigger:
    project: my/deployment
    branch: stable

It’s possible to mirror the status from a triggered pipeline:

trigger_job:
  trigger:
    project: my/project
    strategy: depend

It’s possible to mirror the status from an upstream pipeline:

upstream_bridge:
  stage: test
  needs:
    pipeline: other/project

trigger syntax for child pipeline

To create a child pipeline, specify the path to the YAML file containing the CI config of the child pipeline:

trigger_job:
  trigger:
    include: path/to/child-pipeline.yml

Similar to multi-project pipelines, it’s possible to mirror the status from a triggered pipeline:

trigger_job:
  trigger:
    include:
      - local: path/to/child-pipeline.yml
    strategy: depend

Trigger child pipeline with generated configuration file

You can also trigger a child pipeline from a dynamically generated configuration file:

generate-config:
  stage: build
  script: generate-ci-config > generated-config.yml
  artifacts:
    paths:
      - generated-config.yml

child-pipeline:
  stage: test
  trigger:
    include:
      - artifact: generated-config.yml
        job: generate-config

The generated-config.yml is extracted from the artifacts and used as the configuration for triggering the child pipeline.

Linking pipelines with trigger:strategy

By default, the trigger job completes with the success status as soon as the downstream pipeline is created.

To force the trigger job to wait for the downstream (multi-project or child) pipeline to complete, use strategy: depend. This will make the trigger job wait with a “running” status until the triggered pipeline completes. At that point, the trigger job will complete and display the same status as the downstream job.

trigger_job:
  trigger:
    include: path/to/child-pipeline.yml
    strategy: depend

This can help keep your pipeline execution linear. In the example above, jobs from subsequent stages will wait for the triggered pipeline to successfully complete before starting, at the cost of reduced parallelization.

Trigger a pipeline by API call

Triggers can be used to force a rebuild of a specific branch, tag or commit, with an API call when a pipeline gets created using a trigger token.

Not to be confused with the trigger parameter.

Read more in the triggers documentation.

interruptible

interruptible is used to indicate that a job should be canceled if made redundant by a newer pipeline run. Defaults to false. This value will only be used if the automatic cancellation of redundant pipelines feature is enabled.

When enabled, a pipeline on the same branch will be canceled when:

• It’s made redundant by a newer pipeline run.
• Either all jobs are set as interruptible, or any uninterruptible jobs haven’t started.

Pending jobs are always considered interruptible.

Here is a simple example:

stages:
  - stage1
  - stage2
  - stage3

step-1:
  stage: stage1
  script:
    - echo "Can be canceled."
  interruptible: true

step-2:
  stage: stage2
  script:
    - echo "Can not be canceled."

step-3:
  stage: stage3
  script:
    - echo "Because step-2 can not be canceled, this step will never be canceled, even though set as interruptible."
  interruptible: true

In the example above, a new pipeline run will cause an existing running pipeline to be:

• Canceled, if only step-1 is running or pending.
• Not canceled, once step-2 starts running.

resource_group

Sometimes running multiple jobs or pipelines at the same time in an environment can lead to errors during the deployment.

To avoid these errors, the resource_group attribute can be used to ensure that the Runner won’t run certain jobs simultaneously.

When the resource_group key is defined for a job in .gitlab-ci.yml, job executions are mutually exclusive across different pipelines for the same project. If multiple jobs belonging to the same resource group are enqueued simultaneously, only one of the jobs will be picked by the Runner, and the other jobs will wait until the resource_group is free.

Here is a simple example:

deploy-to-production:
  script: deploy
  resource_group: production

In this case, if a deploy-to-production job is running in a pipeline, and a new deploy-to-production job is created in a different pipeline, it won’t run until the currently running/pending deploy-to-production job is finished. As a result, you can ensure that concurrent deployments will never happen to the production environment.

There can be multiple resource_groups defined per environment. A good use case for this is when deploying to physical devices. You may have more than one physical device, and each one can be deployed to, but there can be only one deployment per device at any given time.

For more information, see Deployments Safety.

release

release indicates that the job creates a Release.

These methods are supported:

• tag_name
• name (optional)
• description (optional)
• ref (optional)
• milestones (optional)
• released_at (optional)

The Release is created only if the job processes without error. If the Rails API returns an error during Release creation, the release job fails.

release-cli Docker image

The Docker image to use for the release-cli must be specified, using the following directive:

image: registry.gitlab.com/gitlab-org/release-cli:latest

Script

All jobs require a script tag at a minimum. A :release job can use the output of a :script tag, but if this is not necessary, a placeholder script can be used, for example:

script:
  - echo 'release job'

An issue exists to remove this requirement in an upcoming version of GitLab.

A pipeline can have multiple release jobs, for example:

ios-release:
  script:
    - echo 'iOS release job'
  release:
     tag_name: v1.0.0-ios
     description: 'iOS release v1.0.0'

android-release:
  script:
    - echo 'Android release job'
  release:
     tag_name: v1.0.0-android
     description: 'Android release v1.0.0'

release:tag_name

The tag_name must be specified. It can refer to an existing Git tag or can be specified by the user.

When the specified tag doesn’t exist in the repository, a new tag is created from the associated SHA of the pipeline.

For example, when creating a Release from a Git tag:

job:
  release:
    tag_name: $CI_COMMIT_TAG
    description: changelog.txt

It is also possible to create any unique tag, in which case only: tags is not mandatory. A semantic versioning example:

job:
  release:
    tag_name: ${MAJOR}_${MINOR}_${REVISION}
    description: changelog.txt

• The Release is created only if the job’s main script succeeds.
• If the Release already exists, it is not updated and the job with the release keyword fails.
• The release section executes after the script tag and before the after_script.

release:name

The Release name. If omitted, it is populated with the value of release: tag_name.

release:description

Specifies the longer description of the Release.

release:ref

If the release: tag_name doesn’t exist yet, the release is created from ref. ref can be a commit SHA, another tag name, or a branch name.

release:milestones

The title of each milestone the release is associated with.

release:released_at

The date and time when the release is ready. Defaults to the current date and time if not defined. Expected in ISO 8601 format (2019-03-15T08:00:00Z).

Complete example for release

Combining the individual examples given above for release results in the following code snippets. There are two options, depending on how you generate the tags. These options cannot be used together, so choose one:

• To create a release when you push a Git tag, or when you add a Git tag in the UI by going to Repository > Tags:

  release_job:
    stage: release
    image: registry.gitlab.com/gitlab-org/release-cli:latest
    rules:
      - if: $CI_COMMIT_TAG                  # Run this job when a tag is created manually
    script:
      - echo 'running release_job'
    release:
       name: 'Release $CI_COMMIT_TAG'
       description: 'Created using the release-cli $EXTRA_DESCRIPTION' # $EXTRA_DESCRIPTION must be defined
       tag_name: '$CI_COMMIT_TAG'                                      # elsewhere in the pipeline.
       ref: '$CI_COMMIT_TAG'
       milestones:
         - 'm1'
         - 'm2'
         - 'm3'
       released_at: '2020-07-15T08:00:00Z'  # Optional, will auto generate if not defined,
                                            # or can use a variable.
  

• To create a release automatically when commits are pushed or merged to the default branch, using a new Git tag that is defined with variables:

  prepare_job:
    stage: prepare                                              # This stage must run before the release stage
    rules:
      - if: $CI_COMMIT_TAG
        when: never                                             # Do not run this job when a tag is created manually
      - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH             # Run this job when commits are pushed or merged to the default branch
    script:
      - echo "EXTRA_DESCRIPTION=some message" >> variables.env  # Generate the EXTRA_DESCRIPTION and TAG environment variables
      - echo "TAG=v$(cat VERSION)" >> variables.env             # and append to the variables.env file
    artifacts:
      reports:
        dotenv: variables.env                                   # Use artifacts:reports:dotenv to expose the variables to other jobs

  release_job:
    stage: release
    image: registry.gitlab.com/gitlab-org/release-cli:latest
    needs:
    - job: prepare_job
      artifacts: true
    rules:
      - if: $CI_COMMIT_TAG
        when: never                                 # Do not run this job when a tag is created manually
      - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH # Run this job when commits are pushed or merged to the default branch
    script:
      - echo 'running release_job for $TAG'
    release:
       name: 'Release $TAG'
       description: 'Created using the release-cli $EXTRA_DESCRIPTION' # $EXTRA_DESCRIPTION and the $TAG
       tag_name: '$TAG'                                                # variables must be defined elsewhere
       ref: '$CI_COMMIT_SHA'                                           # in the pipeline. For example, in the
       milestones:                                                     # prepare_job
         - 'm1'
         - 'm2'
         - 'm3'
       released_at: '2020-07-15T08:00:00Z'          # Optional, will auto generate if not defined,
                                                    # or can use a variable.

releaser-cli command line

The entries under the :release node are transformed into a bash command line and sent to the Docker container, which contains the release-cli. You can also call the release-cli directly from a script entry.

The YAML described above would be translated into a CLI command like this:

release-cli create --name "Release $CI_COMMIT_SHA" --description "Created using the release-cli $EXTRA_DESCRIPTION" --tag-name "v${MAJOR}.${MINOR}.${REVISION}" --ref "$CI_COMMIT_SHA" --released-at "2020-07-15T08:00:00Z" --milestone "m1" --milestone "m2" --milestone "m3"

pages

pages is a special job that is used to upload static content to GitLab that can be used to serve your website. It has a special syntax, so the two requirements below must be met:

• Any static content must be placed under a public/ directory.
• artifacts with a path to the public/ directory must be defined.

The example below simply moves all files from the root of the project to the public/ directory. The .public workaround is so cp does not also copy public/ to itself in an infinite loop:

pages:
  stage: deploy
  script:
    - mkdir .public
    - cp -r * .public
    - mv .public public
  artifacts:
    paths:
      - public
  only:
    - master

Read more on GitLab Pages user documentation.

variables

GitLab CI/CD allows you to define variables inside .gitlab-ci.yml that are then passed in the job environment. They can be set globally and per-job. When the variables keyword is used on a job level, it will override the global YAML variables and predefined ones of the same name.

They are stored in the Git repository and are meant to store non-sensitive project configuration, for example:

variables:
  DATABASE_URL: "postgres://postgres@postgres/my_database"

These variables can be later used in all executed commands and scripts. The YAML-defined variables are also set to all created service containers, thus allowing to fine tune them.

Except for the user defined variables, there are also the ones set up by the Runner itself. One example would be CI_COMMIT_REF_NAME which has the value of the branch or tag name for which project is built. Apart from the variables you can set in .gitlab-ci.yml, there are also the so called Variables which can be set in GitLab’s UI.

YAML anchors for variables are available.

Learn more about variables and their priority.

Git strategy

You can set the GIT_STRATEGY used for getting recent application code, either globally or per-job in the variables section. If left unspecified, the default from project settings will be used.

There are three possible values: clone, fetch, and none.

clone is the slowest option. It clones the repository from scratch for every job, ensuring that the local working copy is always pristine.

variables:
  GIT_STRATEGY: clone

fetch is faster as it re-uses the local working copy (falling back to clone if it does not exist). git clean is used to undo any changes made by the last job, and git fetch is used to retrieve commits made since the last job ran.

variables:
  GIT_STRATEGY: fetch

none also re-uses the local working copy, but skips all Git operations (including GitLab Runner’s pre-clone script, if present). It’s mostly useful for jobs that operate exclusively on artifacts (for example, deploy). Git repository data may be present, but it’s certain to be out of date, so you should only rely on files brought into the local working copy from cache or artifacts.

variables:
  GIT_STRATEGY: none

Git submodule strategy

Requires GitLab Runner v1.10+.

The GIT_SUBMODULE_STRATEGY variable is used to control if / how Git submodules are included when fetching the code before a build. You can set them globally or per-job in the variables section.

There are three possible values: none, normal, and recursive:

• none means that submodules won’t be included when fetching the project code. This is the default, which matches the pre-v1.10 behavior.

• normal means that only the top-level submodules will be included. It’s equivalent to:

  git submodule sync
  git submodule update --init
  

• recursive means that all submodules (including submodules of submodules) will be included. This feature needs Git v1.8.1 and later. When using a GitLab Runner with an executor not based on Docker, make sure the Git version meets that requirement. It’s equivalent to:

  git submodule sync --recursive
  git submodule update --init --recursive
  

Note that for this feature to work correctly, the submodules must be configured (in .gitmodules) with either:

• the HTTP(S) URL of a publicly-accessible repository, or
• a relative path to another repository on the same GitLab server. See the Git submodules documentation.

Git checkout

The GIT_CHECKOUT variable can be used when the GIT_STRATEGY is set to either clone or fetch to specify whether a git checkout should be run. If not specified, it defaults to true. You can set them globally or per-job in the variables section.

If set to false, the Runner will:

• when doing fetch - update the repository and leave working copy on the current revision,
• when doing clone - clone the repository and leave working copy on the default branch.

Having this setting set to true will mean that for both clone and fetch strategies the Runner will checkout the working copy to a revision related to the CI pipeline:

variables:
  GIT_STRATEGY: clone
  GIT_CHECKOUT: "false"
script:
  - git checkout -B master origin/master
  - git merge $CI_COMMIT_SHA

Git clean flags

The GIT_CLEAN_FLAGS variable is used to control the default behavior of git clean after checking out the sources. You can set it globally or per-job in the variables section.

GIT_CLEAN_FLAGS accepts all possible options of the git clean command.

git clean is disabled if GIT_CHECKOUT: "false" is specified.

If GIT_CLEAN_FLAGS is:

• Not specified, git clean flags default to -ffdx.
• Given the value none, git clean is not executed.

For example:

variables:
  GIT_CLEAN_FLAGS: -ffdx -e cache/
script:
  - ls -al cache/

Git fetch extra flags

The GIT_FETCH_EXTRA_FLAGS variable is used to control the behavior of git fetch. You can set it globally or per-job in the variables section.

GIT_FETCH_EXTRA_FLAGS accepts all possible options of the git fetch command, but please note that GIT_FETCH_EXTRA_FLAGS flags will be appended after the default flags that can’t be modified.

The default flags are:

• GIT_DEPTH.
• The list of refspecs.
• A remote called origin.

If GIT_FETCH_EXTRA_FLAGS is:

• Not specified, git fetch flags default to --prune --quiet along with the default flags.
• Given the value none, git fetch is executed only with the default flags.

For example, the default flags are --prune --quiet, so you can make git fetch more verbose by overriding this with just --prune:

variables:
  GIT_FETCH_EXTRA_FLAGS: --prune
script:
  - ls -al cache/

The configurtion above will result in git fetch being called this way:

git fetch origin $REFSPECS --depth 50  --prune

Where $REFSPECS is a value provided to the Runner internally by GitLab.

Job stages attempts

You can set the number for attempts the running job will try to execute each of the following stages:

The default is one single attempt.

Example:

variables:
  GET_SOURCES_ATTEMPTS: 3

You can set them globally or per-job in the variables section.

Shallow cloning

You can specify the depth of fetching and cloning using GIT_DEPTH. This allows shallow cloning of the repository which can significantly speed up cloning for repositories with a large number of commits or old, large binaries. The value is passed to git fetch and git clone.

Since Git fetching and cloning is based on a ref, such as a branch name, Runners can’t clone a specific commit SHA. If there are multiple jobs in the queue, or you’re retrying an old job, the commit to be tested needs to be within the Git history that is cloned. Setting too small a value for GIT_DEPTH can make it impossible to run these old commits. You will see unresolved reference in job logs. You should then reconsider changing GIT_DEPTH to a higher value.

Jobs that rely on git describe may not work correctly when GIT_DEPTH is set since only part of the Git history is present.

To fetch or clone only the last 3 commits:

variables:
  GIT_DEPTH: "3"

You can set it globally or per-job in the variables section.

Custom build directories

By default, GitLab Runner clones the repository in a unique subpath of the $CI_BUILDS_DIR directory. However, your project might require the code in a specific directory (Go projects, for example). In that case, you can specify the GIT_CLONE_PATH variable to tell the Runner in which directory to clone the repository:

variables:
  GIT_CLONE_PATH: $CI_BUILDS_DIR/project-name

test:
  script:
    - pwd

The GIT_CLONE_PATH has to always be within $CI_BUILDS_DIR. The directory set in $CI_BUILDS_DIR is dependent on executor and configuration of runners.builds_dir setting.

Handling concurrency

An executor using a concurrency greater than 1 might lead to failures because multiple jobs might be working on the same directory if the builds_dir is shared between jobs. GitLab Runner does not try to prevent this situation. It’s up to the administrator and developers to comply with the requirements of Runner configuration.

To avoid this scenario, you can use a unique path within $CI_BUILDS_DIR, because Runner exposes two additional variables that provide a unique ID of concurrency:

• $CI_CONCURRENT_ID: Unique ID for all jobs running within the given executor.
• $CI_CONCURRENT_PROJECT_ID: Unique ID for all jobs running within the given executor and project.

The most stable configuration that should work well in any scenario and on any executor is to use $CI_CONCURRENT_ID in the GIT_CLONE_PATH. For example:

variables:
  GIT_CLONE_PATH: $CI_BUILDS_DIR/$CI_CONCURRENT_ID/project-name

test:
  script:
    - pwd

The $CI_CONCURRENT_PROJECT_ID should be used in conjunction with $CI_PROJECT_PATH as the $CI_PROJECT_PATH provides a path of a repository. That is, group/subgroup/project. For example:

variables:
  GIT_CLONE_PATH: $CI_BUILDS_DIR/$CI_CONCURRENT_ID/$CI_PROJECT_PATH

test:
  script:
    - pwd

Nested paths

The value of GIT_CLONE_PATH is expanded once and nesting variables within is not supported.

For example, you define both the variables below in your .gitlab-ci.yml file:

variables:
  GOPATH: $CI_BUILDS_DIR/go
  GIT_CLONE_PATH: $GOPATH/src/namespace/project

The value of GIT_CLONE_PATH is expanded once into $CI_BUILDS_DIR/go/src/namespace/project, and results in failure because $CI_BUILDS_DIR is not expanded.

Special YAML features

It’s possible to use special YAML features like anchors (&), aliases (*) and map merging (<<), which will allow you to greatly reduce the complexity of .gitlab-ci.yml.

Read more about the various YAML features.

In most cases, the extends keyword is more user friendly and should be used over these special YAML features. YAML anchors may still need to be used to merge arrays.

Anchors

YAML has a handy feature called ‘anchors’, which lets you easily duplicate content across your document. Anchors can be used to duplicate/inherit properties, and is a perfect example to be used with hidden jobs to provide templates for your jobs. When there is duplicate keys, GitLab will perform a reverse deep merge based on the keys.

The following example uses anchors and map merging. It will create two jobs, test1 and test2, that will inherit the parameters of .job_template, each having their own custom script defined:

.job_template: &job_definition  # Hidden key that defines an anchor named 'job_definition'
  image: ruby:2.6
  services:
    - postgres
    - redis

test1:
  <<: *job_definition           # Merge the contents of the 'job_definition' alias
  script:
    - test1 project

test2:
  <<: *job_definition           # Merge the contents of the 'job_definition' alias
  script:
    - test2 project

& sets up the name of the anchor (job_definition), << means “merge the given hash into the current one”, and * includes the named anchor (job_definition again). The expanded version looks like this:

.job_template:
  image: ruby:2.6
  services:
    - postgres
    - redis

test1:
  image: ruby:2.6
  services:
    - postgres
    - redis
  script:
    - test1 project

test2:
  image: ruby:2.6
  services:
    - postgres
    - redis
  script:
    - test2 project

Let’s see another one example. This time we will use anchors to define two sets of services. This will create two jobs, test:postgres and test:mysql, that will share the script directive defined in .job_template, and the services directive defined in .postgres_services and .mysql_services respectively:

.job_template: &job_definition
  script:
    - test project
  tags:
    - dev

.postgres_services:
  services: &postgres_definition
    - postgres
    - ruby

.mysql_services:
  services: &mysql_definition
    - mysql
    - ruby

test:postgres:
  <<: *job_definition
  services: *postgres_definition
  tags:
    - postgres

test:mysql:
  <<: *job_definition
  services: *mysql_definition

The expanded version looks like this:

.job_template:
  script:
    - test project
  tags:
    - dev

.postgres_services:
  services:
    - postgres
    - ruby

.mysql_services:
  services:
    - mysql
    - ruby

test:postgres:
  script:
    - test project
  services:
    - postgres
    - ruby
  tags:
    - postgres

test:mysql:
  script:
    - test project
  services:
    - mysql
    - ruby
  tags:
    - dev

You can see that the hidden jobs are conveniently used as templates.

YAML anchors for before_script and after_script

You can use YAML anchors with before_script and after_script, which makes it possible to include a predefined list of commands in multiple jobs.

Example:

.something_before: &something_before
  - echo 'something before'

.something_after: &something_after
  - echo 'something after'
  - echo 'another thing after'

job_name:
  before_script:
    - *something_before
  script:
    - echo 'this is the script'
  after_script:
    - *something_after

YAML anchors for script

You can use YAML anchors with scripts, which makes it possible to include a predefined list of commands in multiple jobs.

For example:

.something: &something
  - echo 'something'

job_name:
  script:
    - *something
    - echo 'this is the script'