Writing Tasks

A task defines a sequence of instructions to execute and an execution environment to execute these instructions in. Let's see a line-by-line example of a .cirrus.yml configuration file first:

amd64arm64

test_task:
  container:
    image: openjdk:latest
  test_script: ./gradlew test

test_task:
  arm_container:
    image: openjdk:latest
  test_script: ./gradlew test

The example above defines a single task that will be scheduled and executed on the Linux Cluster using the openjdk:latest Docker image. Only one user-defined script instruction to run ./gradlew test will be executed. Not that complex, right?

Please read the topics below if you want better understand what's going on in a more complex .cirrus.yml configuration file, such as this:

amd64arm64

task:
  container:
    image: node:latest # (1)

  node_modules_cache: # (2)
    folder: node_modules
    fingerprint_script: cat yarn.lock
    populate_script: yarn install

  matrix: # (3)
    - name: Lint
      skip: !changesInclude('.cirrus.yml', '**.{js,ts}') # (4)
      lint_script: yarn run lint
    - name: Test
      container:
        matrix: # (5)
          - image: node:latest
          - image: node:lts
      test_script: yarn run test
    - name: Publish
      depends_on:
        - Lint
        - Test
      only_if: $BRANCH == "master" # (6)
      publish_script: yarn run publish

1. Use any Docker image from public or private registries
2. Use cache instruction to persist folders based on an arbitrary fingerprint_script.
3. Use matrix modification to produce many similar tasks.
4. See what kind of files were changes and skip tasks that are not applicable. See changesInclude and changesIncludeOnly documentation for details.
5. Use nested matrix modification to produce even more tasks.
6. Completely exclude tasks from execution graph by any custom condition.

task:
  arm_container:
    image: node:latest # (1)

  node_modules_cache: # (2)
    folder: node_modules
    fingerprint_script: cat yarn.lock
    populate_script: yarn install

  matrix: # (3)
    - name: Lint
      skip: !changesInclude('.cirrus.yml', '**.{js,ts}') # (4)
      lint_script: yarn run lint
    - name: Test
      arm_container:
        matrix: # (5)
          - image: node:latest
          - image: node:lts
      test_script: yarn run test
    - name: Publish
      depends_on:
        - Lint
        - Test
      only_if: $BRANCH == "master" # (6)
      publish_script: yarn run publish

1. Use any Docker image from public or private registries
2. Use cache instruction to persist folders based on an arbitrary fingerprint_script.
3. Use matrix modification to produce many similar tasks.
4. See what kind of files were changes and skip tasks that are not applicable. See changesInclude and changesIncludeOnly documentation for details.
5. Use nested matrix modification to produce even more tasks.
6. Completely exclude tasks from execution graph by any custom condition.

Task Naming

To name a task one can use the name field. foo_task syntax is a syntactic sugar. Separate name field is very useful when you want to have a rich task name:

task:
  name: Tests (macOS)
  ...

Note: instructions within a task can only be named via a prefix (e.g. test_script).

Visual Task Creation for Beginners

If you are just getting started and prefer a more visual way of creating tasks, there is a third-party Cirrus CI Configuration Builder for generating YAML config that might be helpful.

Execution Environment

In order to specify where to execute a particular task you can choose from a variety of options by defining one of the following fields for a task:

Field Name	Managed by	Description
container	us	Linux Docker Container
arm_container	us	Linux Arm Docker Container
windows_container	us	Windows Docker Container
macos_instance	us	macOS Virtual Machines
freebsd_instance	us	FreeBSD Virtual Machines
compute_engine_instance	us	Full-fledged custom VM
persistent_worker	you	Use any host on any platform and architecture
gce_instance	you	Linux, Windows and FreeBSD Virtual Machines in your GCP project
gke_container	you	Linux Docker Containers on private GKE cluster
ec2_instance	you	Linux Virtual Machines in your AWS
eks_instance	you	Linux Docker Containers on private EKS cluster
azure_container_instance	you	Linux and Windows Docker Container on Azure
oke_instance	you	Linux x86 and Arm Containers on Oracle Cloud

Supported Instructions

Each task is essentially a collection of instructions that are executed sequentially. The following instructions are supported:

• script instruction to execute a script.
• background_script instruction to execute a script in a background.
• cache instruction to persist files between task runs.
• artifacts instruction to store and expose files created via a task.
• file instruction to create a file from an environment variable.

Script Instruction

A script instruction executes commands via shell on Unix or batch on Windows. A script instruction can be named by adding a name as a prefix. For example test_script or my_very_specific_build_step_script. Naming script instructions helps gather more granular information about task execution. Cirrus CI will use it in future to auto-detect performance regressions.

Script commands can be specified as a single string value or a list of string values in a .cirrus.yml configuration file like in the example below:

check_task:
  compile_script: gradle --parallel classes testClasses
  check_script:
    - echo "Here comes more than one script!"
    - printenv
    - gradle check

Note: Each script instruction is executed in a newly created process, therefore environment variables are not preserved between them.

Execution on Windows

When executed on Windows via batch, Cirrus Agent will wrap each line of the script in a call so it's possible to fail fast upon first line exiting with non-zero exit code.

To avoid this "syntactic sugar" just create a script file and execute it.

Background Script Instruction

A background_script instruction is absolutely the same as script instruction but Cirrus CI won't wait for the script to finish and will continue execution of further instructions.

Background scripts can be useful when something needs to be executed in the background. For example, a database or some emulators. Traditionally the same effect is achieved by adding & to a command like $: command &. Problem here is that logs from command will be mixed into regular logs of the following commands. By using background scripts not only logs will be properly saved and displayed, but also command itself will be properly killed in the end of a task.

Here is an example of how background_script instruction can be used to run an android emulator:

android_test_task:
  start_emulator_background_script: emulator -avd test -no-audio -no-window
  wait_for_emulator_to_boot_script: adb wait-for-device
  test_script: gradle test

Cache Instruction

A cache instruction allows to persist a folder and reuse it during the next execution of the task. A cache instruction can be named the same way as script instruction.

Here is an example:

amd64arm64

test_task:
  container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
    reupload_on_changes: false # since there is a fingerprint script
    fingerprint_script:
      - echo $CIRRUS_OS
      - node --version
      - cat package-lock.json
    populate_script: 
      - npm install
  test_script: npm run test

test_task:
  arm_container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
    reupload_on_changes: false # since there is a fingerprint script
    fingerprint_script:
      - echo $CIRRUS_OS
      - node --version
      - cat package-lock.json
    populate_script: 
      - npm install
  test_script: npm run test

Either folder or a folders field (with a list of folder paths) is required and they tell the agent which folder paths to cache.

Folder paths should be generally relative to the working directory (e.g. node_modules), with the exception of when only a single folder specified. In this case, it can be also an absolute path (/usr/local/bundle).

Folder paths can contain a "glob" pattern to cache multiple files/folders within a working directory (e.g. **/node_modules will cache every node_modules folder within the working directory).

A fingerprint_script and fingerprint_key are optional fields that can specify either:

• a script, the output of which will be hashed and used as a key for the given cache:

node_modules_cache:
  folder: node_modules
  fingerprint_script: cat yarn.lock

• a final cache key:

node_modules_cache:
  folder: node_modules
  fingerprint_key: 2038-01-20

These two fields are mutually exclusive. By default the task name is used as a fingerprint value.

After the last script instruction for the task succeeds, Cirrus CI will calculate checksum of the cached folder (note that it's unrelated to fingerprint_script or fingerprint_key fields) and re-upload the cache if it finds any changes. To avoid a time-costly re-upload, remove volatile files from the cache (for example, in the last script instruction of a task).

populate_script is an optional field that can specify a script that will be executed to populate the cache. populate_script should create the folder if it doesn't exist before the cache instruction. If your dependencies are updated often, please pay attention to fingerprint_script and make sure it will produce different outputs for different versions of your dependency (ideally just print locked versions of dependencies).

reupload_on_changes is an optional field that can specify whether Cirrus Agent should check if contents of cached folder have changed during task execution and re-upload a cache entry in case of any changes. If reupload_on_changes option is not set explicitly then it will be set to false if fingerprint_script or fingerprint_key is presented and true otherwise. Cirrus Agent will detect additions, deletions and modifications of any files under specified folder. All of the detected changes will be logged under Upload '$CACHE_NAME' cache instructions for easier debugging of cache invalidations.

That means the only difference between the example above and below is that yarn install will always be executed in the example below where in the example above only when yarn.lock has changes.

amd64arm64

test_task:
  container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
    fingerprint_script: cat yarn.lock
  install_script: yarn install
  test_script: yarn run test

test_task:
  arm_container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
    fingerprint_script: cat yarn.lock
  install_script: yarn install
  test_script: yarn run test

Caching for Pull Requests

Tasks for PRs upload caches to a separate caching namespace to not interfere with caches used by other tasks. But such PR tasks can read all caches even from the main caching namespace for a repository.

Scope of cached artifacts

Cache artifacts are shared between tasks, so two caches with the same name on e.g. Linux containers and macOS VMs will share the same set of files. This may introduce binary incompatibility between caches. To avoid that, add echo $CIRRUS_OS into fingerprint_script or use $CIRRUS_OS in fingerprint_key, which will distinguish caches based on OS.

Manual cache upload

Normally caches are uploaded at the end of the task execution. However, you can override the default behavior and upload them earlier.

To do this, use the upload_caches instruction, which uploads a list of caches passed to it once executed:

amd64arm64

test_task:
  container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
  upload_caches:
    - node_modules
  install_script: yarn install
  test_script: yarn run test
  pip_cache:
    folder: ~/.cache/pip

test_task:
  arm_container:
    image: node:latest
  node_modules_cache:
    folder: node_modules
  upload_caches:
    - node_modules
  install_script: yarn install
  test_script: yarn run test
  pip_cache:
    folder: ~/.cache/pip

Note that pip cache won't be uploaded in this example: using upload_caches disables the default behavior where all caches are automatically uploaded at the end of the task, so if you want to upload pip cache too, you'll have to either:

• extend the list of uploaded caches in the first upload_caches instruction
• insert a second upload_caches instruction that specifically targets pip cache

Artifacts Instruction

An artifacts instruction allows to store files and expose them in the UI for downloading later. An artifacts instruction can be named the same way as script instruction and has only one required path field which accepts a glob pattern of files relative to $CIRRUS_WORKING_DIR to store. Right now only storing files under $CIRRUS_WORKING_DIR folder as artifacts is supported with a total size limit of 1G for a free task and with no limit on your own infrastructure.

In the example below, Build and Test task produces two artifacts: binaries artifacts with all executables built during a successful task completion and junit artifacts with all test reports regardless of the final task status (more about that you can learn in the next section describing execution behavior).

build_and_test_task:
  # instructions to build and test
  binaries_artifacts:
    path: "build/*"
  always:
    junit_artifacts:
      path: "**/test-results/**.xml"
      format: junit

URLs to the artifacts

Latest build artifacts

It is possible to refer to the latest artifacts directly (artifacts of the latest successful build). Use the following link format to download the latest artifact of a particular task:

https://api.cirrus-ci.com/v1/artifact/github/<USER OR ORGANIZATION>/<REPOSITORY>/<TASK NAME OR ALIAS>/<ARTIFACTS_NAME>/<PATH>

It is possible to also download an archive of all files within an artifact with the following link:

https://api.cirrus-ci.com/v1/artifact/github/<USER OR ORGANIZATION>/<REPOSITORY>/<TASK NAME OR ALIAS>/<ARTIFACTS_NAME>.zip

By default, Cirrus looks up the latest successful build of the default branch for the repository but the branch name can be customized via ?branch=<BRANCH> query parameter.

Current build artifacts

It is possible to refer to the artifacts of the current build directly:

https://api.cirrus-ci.com/v1/artifact/build/<CIRRUS_BUILD_ID>/<ARTIFACTS_NAME>.zip

Note that if several tasks are uploading artifacts with the same name then the ZIP archive from the above link will contain merged content of all artifacts. It's also possible to refer to an artifact of a particular task within a build by name:

https://api.cirrus-ci.com/v1/artifact/build/<CIRRUS_BUILD_ID>/<TASK NAME OR ALIAS>/<ARTIFACTS_NAME>.zip

It is also possible to download artifacts given a task id directly:

https://api.cirrus-ci.com/v1/artifact/task/<CIRRUS_TASK_ID>/<ARTIFACTS_NAME>.zip

It's also possible to download a particular file of an artifact and not the whole archive by using <ARTIFACTS_NAME>/<PATH> instead of <ARTIFACTS_NAME>.zip.

Artifact Type

By default, Cirrus CI will try to guess mimetype of files in artifacts by looking at their extensions. In case when artifacts don't have extensions, it's possible to explicitly set the Content-Type via type field:

  my_task:
    my_dotjar_artifacts:
      path: build/*.jar
      type: application/java-archive

A list of some of the basic types supported can be found here.

Artifact Parsing

Cirrus CI supports parsing artifacts in order to extract information that can be presented in the UI for a better user experience. Use the format field of an artifact instruction to specify artifact's format (mimetypes):

junit_artifacts:
  path: "**/test-results/**.xml"
  type: text/xml
  format: junit

Currently, Cirrus CI supports:

• Android Lint Report format
• GolangCI Lint's JSON format
• JUnit's XML format
  • Python's Unittest format
• XCLogParser
• JetBrains Qodana
• Buf CLI for Protocol Buffers

Please let us know what kind of formats Cirrus CI should support next!

File Instruction

A file instruction allows to create a file from either an environment variable or directly from the configuration file. It is especially useful for situations when execution environment doesn't have proper shell to use echo ... >> ... syntax, for example, within scratch Docker containers.

Here is an example of how to populate Docker config from an encrypted environment variable:

task:
  environment:
    DOCKER_CONFIG_JSON: ENCRYPTED[qwerty]
  docker_config_file:
    path: /root/.docker/config.json
    variable_name: DOCKER_CONFIG_JSON

You can also populate a file directly from the .cirrus.yml configuration file:

task:
  git_config_file:
    path: /root/.gitconfig
    from_contents: |
      [user]
        name = John Doe
        email = john@example.com

Execution Behavior of Instructions

By default, Cirrus CI executes instructions one after another and stops the overall task execution on the first failure. Sometimes there might be situations when some scripts should always be executed or some debug information needs to be saved on a failure. For such situations the always and on_failure keywords can be used to group instructions.

task:
  test_script: ./run_tests.sh
  on_failure:
    debug_script: ./print_additional_debug_info.sh
  cleanup_script: ./cleanup.sh # failure here will not trigger `on_failure` instruction above
  always:
    test_reports_script: ./print_test_reports.sh

In the example above, print_additional_debug_info.sh script will be executed only on failures of test_script to output some additional debug information. print_test_reports.sh on the other hand will be executed both on successful and failed runs to print test reports (test reports are always useful! ).

Sometimes, a complex task might exceed the pre-defined timeout, and it might not be clear why. In this case, the on_timeout execution behavior, which has an extra time budget of 5 minutes might be useful:

task:
  integration_test_script: ./run_integration_tests.sh
  on_timeout:
    cleanup_script: ./cleanup.sh

Environment Variables

Environment variables can be configured under the env or environment keywords in .cirrus.yml files. Here is an example:

echo_task:
  env:
    FOO: Bar
  echo_script: echo $FOO

You can reference other environment variables using $VAR, ${VAR} or %VAR% syntax:

custom_path_task:
  env:
    SDK_ROOT: ${HOME}/sdk
    PATH: ${SDK_ROOT}/bin:${PATH}
  custom_script: sdktool install

Environment variables may also be set at the root level of .cirrus.yml. In that case, they will be merged with each task's individual environment variables, but the task level variables always take precedence. For example:

env:
  PATH: /sdk/bin:${PATH}

echo_task:
  env:
    PATH: /opt/bin:${PATH}
  echo_script: echo $PATH

Will output /opt/bin:/usr/local/bin:/usr/bin or similar, but will not include /sdk/bin because this root level setting is ignored.

Also some default environment variables are pre-defined:

Name	Value / Description
CI	true
CIRRUS_CI	true
CI_NODE_INDEX	Index of the current task within CI_NODE_TOTAL tasks
CI_NODE_TOTAL	Total amount of unique tasks for a given CIRRUS_BUILD_ID build
CONTINUOUS_INTEGRATION	true
CIRRUS_API_CREATED	true if the current build was created through the API.
CIRRUS_BASE_BRANCH	Base branch name if current build was triggered by a PR. For example master
CIRRUS_BASE_SHA	Base SHA if current build was triggered by a PR
CIRRUS_BRANCH	Branch name. For example my-feature
CIRRUS_BUILD_ID	Unique build ID
CIRRUS_CHANGE_IN_REPO	Git SHA
CIRRUS_CHANGE_MESSAGE	Commit message or PR title and description, depending on trigger event (Non-PRs or PRs respectively).
CIRRUS_CHANGE_TITLE	First line of CIRRUS_CHANGE_MESSAGE
CIRRUS_CPU	Amount of CPUs requested by the task. CIRRUS_CPU value is integer and rounded up for tasks that requested non-interger amount of CPUs.
CIRRUS_CRON	Cron Build name configured in the repository settings if this build was triggered by Cron. For example, nightly.
CIRRUS_DEFAULT_BRANCH	Default repository branch name. For example master
CIRRUS_DOCKER_CONTEXT	Docker build's context directory to use for Dockerfile as a CI environment. Defaults to project's root directory.
CIRRUS_LAST_GREEN_BUILD_ID	The build id of the last successful build on the same branch at the time of the current build creation.
CIRRUS_LAST_GREEN_CHANGE	Corresponding to CIRRUS_LAST_GREEN_BUILD_ID SHA (used in changesInclude and changesIncludeOnly functions).
CIRRUS_PR	PR number if current build was triggered by a PR. For example 239.
CIRRUS_PR_DRAFT	true if current build was triggered by a Draft PR.
CIRRUS_PR_TITLE	Title of a corresponding PR if any.
CIRRUS_PR_BODY	Body of a corresponding PR if any.
CIRRUS_PR_LABELS	comma separated list of PR's labels if current build was triggered by a PR.
CIRRUS_TAG	Tag name if current build was triggered by a new tag. For example v1.0
CIRRUS_OIDC_TOKEN	OpenID Connect Token issued by https://oidc.cirrus-ci.com with audience set to https://cirrus-ci.com/github/$CIRRUS_REPO_OWNER (can be changed via $CIRRUS_OIDC_TOKEN_AUDIENCE). Please refer to a dedicated section below for in-depth details.
CIRRUS_OS, OS	Host OS. Either linux, windows or darwin.
CIRRUS_TASK_NAME	Task name
CIRRUS_TASK_NAME_ALIAS	Task name alias if any.
CIRRUS_TASK_ID	Unique task ID
CIRRUS_RELEASE	GitHub Release id if current tag was created for a release. Handy for uploading release assets.
CIRRUS_REPO_CLONE_TOKEN	Temporary GitHub access token to perform a clone.
CIRRUS_REPO_NAME	Repository name. For example my-project
CIRRUS_REPO_OWNER	Repository owner (an organization or a user). For example my-organization
CIRRUS_REPO_FULL_NAME	Repository full name/slug. For example my-organization/my-project
CIRRUS_REPO_CLONE_URL	URL used for cloning. For example https://github.com/my-organization/my-project.git
CIRRUS_USER_COLLABORATOR	true if a user initialized a build is already a contributor to the repository. false otherwise.
CIRRUS_USER_PERMISSION	admin, write, read or none.
CIRRUS_HTTP_CACHE_HOST	Host and port number on which local HTTP cache can be accessed on.
GITHUB_CHECK_SUITE_ID	Monotonically increasing id of a corresponding GitHub Check Suite which caused the Cirrus CI build.
CIRRUS_ENV	Path to a file, by writing to which you can set task-wide environment variables.
CIRRUS_ENV_SENSITIVE	Set to true to mask all variable values written to the CIRRUS_ENV file in the console output

Behavioral Environment Variables

And some environment variables can be set to control behavior of the Cirrus CI Agent:

Name	Default Value	Description
CIRRUS_AGENT_EXPOSE_SCRIPTS_OUTPUTS	not set	If set, instructs Cirrus Agent to stream scripts outputs to the console as well as Cirrus API. Useful in case your Kubernetes cluster has logging collection enabled.
CIRRUS_CLI_VERSION	not set	Cirrus CLI version to use when running the agent. If not set, the latest release will be used.
CIRRUS_CLONE_DEPTH	0 which will reflect in a full clone of a single branch	Clone depth.
CIRRUS_CLONE_TAGS	false which will not fetch any tags	Whether to fetch git tags.
CIRRUS_CLONE_SUBMODULES	false	Set to true to clone submodules recursively.
CIRRUS_LOG_TIMESTAMP	false	Indicate Cirrus Agent to prepend timestamp to each line of logs.
CIRRUS_OIDC_TOKEN_AUDIENCE	not set	Allows to override aud claim for CIRRUS_OIDC_TOKEN.
CIRRUS_SHELL	sh on Linux/macOS/FreeBSD and cmd.exe on Windows. Set to direct to execute each script directly without wrapping the commands in a shell script.	Shell that Cirrus CI uses to execute scripts. By default sh is used.
CIRRUS_VOLUME	/tmp	Defines a path for a temporary volume to be mounted into instances running in a Kubernetes cluster. This volume is mounted into all additional containers and is persisted between steps of a pipe.
CIRRUS_WORKING_DIR	cirrus-ci-build folder inside of a system's temporary folder	Working directory where Cirrus CI executes builds. Default to cirrus-ci-build folder inside of a system's temporary folder.
CIRRUS_ESCAPING_PROCESSES	not set	Set this variable to prevent the agent from terminating the processes spawned in each non-background instruction after that instruction ends. By default, the agent tries it's best to garbage collect these processes and their standard input/output streams. It's generally better to use a Background Script Instruction instead of this variable to achieve the same effect.
CIRRUS_WINDOWS_ERROR_MODE	not set	Set this value to force all processes spawned by the agent to call the equivalent of SetErrorMode() with the provided value (for example, 0x8001) before beginning their execution.
CIRRUS_VAULT_URL	not set	Address of the Vault server expressed as a URL and port (for example, https://vault.example.com:8200/), see HashiCorp Vault Support.
CIRRUS_VAULT_NAMESPACE	not set	A Vault Enterprise Namespace to use when authenticating and reading secrets from Vault.
CIRRUS_VAULT_AUTH_PATH	jwt	Alternative auth method mount point, in case it was mounted to a non-default path.
CIRRUS_VAULT_ROLE	not set

Internals of OpenID Connect tokens

OpenID Connect is a very powerful mechanism that allows two independent systems establish trust without sharing any secrets. In the core of OpenID Connect is a simple JWT token that is signed by a trusted party (in our case it's Cirrus CI). Then the second system can be configured to trust such CIRRUS_OIDC_TOKENs signed by Cirrus CI. For examples please check Vault Integration, Google Cloud Integration and AWS Integration.

Once such external system receives a request authenticated with CIRRUS_OIDC_TOKEN it can verify the signature of the token via publicly available keys. Then it can extract claims from the token to make necessary assertions. Properly configuring assertions of such claims is crucial for secure integration with OIDC. Let's take a closer look at claims that are available through a payload of a CIRRUS_OIDC_TOKEN:

task:
  container:
    image: alpine:latest
  install_script: apk add jq
  payload_script: echo $CIRRUS_OIDC_TOKEN | jq -R 'split(".") | .[1] | @base64d | fromjson'

The above task will print out payload of a CIRRUS_OIDC_TOKEN that contains claims from the configuration that can be used for assertions.

{
  // Reserved Claims https://openid.net/specs/draft-jones-json-web-token-07.html#rfc.section.4.1 
  "iss": "https://oidc.cirrus-ci.com",
  "aud": "https://cirrus-ci.com/github/cirruslabs", // can be changed via $CIRRUS_OIDC_TOKEN_AUDIENCE
  "sub": "repo:github:cirruslabs/cirrus-ci-docs",
  "nbf": ...,
  "exp": ...,
  "iat": ...,
  "jti": "...",
  // Cirrus Added Claims
  "platform": "github", // Currently only GitHub is supported but more platforms will be added in the future
  "owner": "cirruslabs", // Unique organization or username on the platform
  "owner_id": "29414678", // Internal ID of the organization or user on the platform
  "repository": "cirrus-ci-docs", // Repository name
  "repository_visibility": "public", // either public or private
  "repository_id": "5730634941071360", // Internal Cirrus CI ID of the repository
  "build_id": "1234567890", // Internal Cirrus CI ID of the build. Same as $CIRRUS_BUILD_ID
  "branch": "fkorotkov-patch-2", // Git branch name. Same as $CIRRUS_BRANCH
  "change_in_repo": "e6e989d4792a678b697a9f17a787761bfefb52d0", // Git commit SHA. Same as $CIRRUS_CHANGE_IN_REPO
  "pr": "123", // Pull request number if a build was triggered by a PR. Same as $CIRRUS_PR
  "pr_draft": "false", // Whether the pull request is a draft. Same as $CIRRUS_PR_DRAFT
  "pr_labels": "", // Comma-separated list of labels of the pull request. Same as $CIRRUS_PR_LABELS
  "tag": "1.0.0", // Git tag name if a build was triggered by a tag creation. Same as $CIRRUS_TAG
  "task_id": "987654321", // Internal Cirrus CI ID of the task. Same as $CIRRUS_TASK_ID
  "task_name": "main", // Name of the task. Same as $CIRRUS_TASK_NAME
  "task_name_alias": "main", // Optional name alias of the task. Same as $CIRRUS_TASK_NAME_ALIAS
  "user_collaborator": "true", // Whether the user is a collaborator of the repository. Same as $CIRRUS_USER_COLLABORATOR
  "user_permission": "admin", // Permission level of the user in the repository. Same as $CIRRUS_USER_PERMISSION
}

Please use the above claims to configure assertions in your external system. For example, you can assert that only tokens for specific branches can retrieve secrets for deploying to production.

Encrypted Variables

It is possible to add encrypted variables to a .cirrus.yml file. These variables are decrypted only in builds for commits and pull requests that are made by users with write permission or approved by them.

In order to encrypt a variable go to repository's settings page via clicking settings icon on a repository's main page (for example https://cirrus-ci.com/github/my-organization/my-repository) and follow instructions.

Warning

Only users with WRITE permissions can add encrypted variables to a repository.

Multi-Line Encrypted Variables

Please avoid using encrypted variables that consist of multiple lines as they won't be masked (replaced by HIDDEN-BY-CIRRUS-CI) correctly due to the line-buffered nature of the task's instructions output.

An encrypted variable will be presented in a form like ENCRYPTED[qwerty239abc] which can be safely committed to .cirrus.yml file:

publish_task:
  environment:
    AUTH_TOKEN: ENCRYPTED[qwerty239abc]
  script: ./publish.sh

Cirrus CI encrypts variables with a unique per-repository 256-bit encryption key, so that forks and even repositories within the same organization cannot re-use them. qwerty239abc from the example above is NOT the content of your encrypted variable, it's just an internal ID. No one can brute force your secrets from such ID. In addition, Cirrus CI doesn't know a relation between an encrypted variable and a repository for which the encrypted variable was created.

Organization Level Encrypted Variables

Sometimes there might be secrets that are used in almost all repositories of an organization. For example, credentials to a compute service where tasks will be executed. In order to create such sharable encrypted variable go to organization's settings page via clicking settings icon on an organization's main page (for example https://cirrus-ci.com/github/my-organization) and follow instructions in Organization Level Encrypted Variables section.

Encrypted Variable for Cloud Credentials

In case you use integration with one of supported computing services, an encrypted variable used to store credentials that Cirrus is using to communicate with the computing service won't be decrypted if used in environment variables. These credentials have too many permissions for most of the cases, please create separate credentials with the minimum needed permissions for your specific case.

gcp_credentials: SECURED[!qwerty]

env:
  CREDENTIALS: SECURED[!qwerty] # won't be decrypted in any case

Skipping Task in Forked Repository

In forked repository the decryption of variable fails, which causes failure of task depending on it. To avoid this by default, make the sensitive task conditional:

task:
  name: Task requiring decrypted variables
  only_if: $CIRRUS_REPO_OWNER == 'my-organization'
  ...

Owner of forked repository can re-enable the task, if they have the required sensitive data, by encrypting the variable by themselves and editing both the encrypted variable and repo-owner condition in the .cirrus.yml file.

HashiCorp Vault support

In addition to using Cirrus CI for managing secrets, it is possible to retrieve secrets from HashiCorp Vault.

You will need to configure a JWT authentication method and point it to the Cirrus CI's OIDC discovery URL: https://oidc.cirrus-ci.com.

This ensures that a cryptographic JWT token (CIRRUS_OIDC_TOKEN) that each Cirrus CI's task get assigned will be verified by your Vault installation.

From the Cirrus CI's side, use the CIRRUS_VAULT_URL environment variable to point Cirrus Agent at your vault and configure other Vault-specific variables, if needed. Note that it's not required for CIRRUS_VAULT_URL to be publicly available since Cirrus CI can orchestrate tasks on your infrastructure. Only Cirrus Agent executing a task from within an execution environment needs access to your Vault.

Once done, you will be able to use the VAULT[path/to/secret selector] syntax to retrieve a version 2 secret, for example:

publish_task:
  environment:
    AUTH_TOKEN: VAULT[secret/data/github data.token]
  script: ./publish.sh

The path is exactly the one you are familiar from invoking Vault CLI like vault read ..., and the selector is a simply dot-delimited list of fields to query in the output.

Caching of Vault secrets

Note that all VAULT[...] invocations cache the retrieved secrets on a per-path basis by default. Caching happens within a single task execution and is not shared between several tasks using the same secret.

To disable caching, use VAULT_NOCACHE[...] instead of VAULT[...].

Multi-Line Vault Secrets

Please avoid referencing Vault secrets that consist of multiple lines as they won't be masked (replaced by HIDDEN-BY-CIRRUS-CI) correctly due to the line-buffered nature of the task's instructions output.

Mixing of VAULT[...] and VAULT_NOCACHE[...] on the same path

Using both VAULT[...] and VAULT_NOCACHE[...] on the same path is not recommended because the order in which these invocations are processed is not deterministic.

Cron Builds

It is possible to configure invocations of re-occurring builds via the well-known Cron expressions. Cron builds can be configured on a repository's settings page (not in .cirrus.yml).

It's possible to configure several cron builds with unique names which will be available via CIRRUS_CRON environment variable. Each cron build should specify branch to trigger new builds for and a cron expression compatible with Quartz. You can use this generator to generate/validate your expressions.

Note: Cron Builds are timed with the UTC timezone.

Matrix Modification

Sometimes it's useful to run the same task against different software versions. Or run different batches of tests based on an environment variable. For cases like these, the matrix modifier comes very handy. It's possible to use matrix keyword only inside of a particular task to have multiple tasks based on the original one. Each new task will be created from the original task by replacing the whole matrix YAML node with each matrix's children separately.

Let check an example of a .cirrus.yml:

amd64arm64

test_task:
  container:
    matrix:
      - image: node:latest
      - image: node:lts
  test_script: yarn run test

test_task:
  arm_container:
    matrix:
      - image: node:latest
      - image: node:lts
  test_script: yarn run test

Which will be expanded into:

amd64arm64

test_task:
  container:
    image: node:latest
  test_script: yarn run test

test_task:
  container:
    image: node:lts
  test_script: yarn run test

test_task:
  arm_container:
    image: node:latest
  test_script: yarn run test

test_task:
  arm_container:
    image: node:lts
  test_script: yarn run test

Tip

The matrix modifier can be used multiple times within a task.

The matrix modification makes it easy to create some pretty complex testing scenarios like this:

amd64arm64

task:
  container:
    matrix:
      - image: node:latest
      - image: node:lts
  node_modules_cache:
    folder: node_modules
    fingerprint_script:
      - node --version
      - cat yarn.lock
    populate_script: yarn install
  matrix:
    - name: Build
      build_script: yarn build
    - name: Test
      test_script: yarn run test

task:
  arm_container:
    matrix:
      - image: node:latest
      - image: node:lts
  node_modules_cache:
    folder: node_modules
    fingerprint_script:
      - node --version
      - cat yarn.lock
    populate_script: yarn install
  matrix:
    - name: Build
      build_script: yarn build
    - name: Test
      test_script: yarn run test

Another example showing how to create 2 tasks with different environment variables:

task:
  name: matrixdemo
  container:
    image: alpine:3.20
  env:
    matrix:
      FIRST_VAR: foo
      FIRST_VAR: bar
      FIRST_VAR: baz
    matrix:
      SECOND_VAR: alpha
      SECOND_VAR: bravo
  print_script:
    - echo "FIRST_VAR value is $FIRST_VAR"
    - echo "SECOND_VAR value is $SECOND_VAR"

The above will generate 6 tasks.

And yet another example, showing how to granularly control the environment variables for each matrix:

task:
  name: matrixdemo
  container:
    image: ubuntu:24.04
  env:
    # Android Command-line tools were obtained with https://stackoverflow.com/a/78890086/7009800
    matrix:
      - ANDROID_CLT_VERSION: 9123335 # v8, latest compatible with JDK 8+
        JDK_PACKAGE: openjdk-8-jdk
      - ANDROID_CLT_VERSION: 9862592 # v10, latest compatible with JDK 11+
        JDK_PACKAGE: openjdk-11-jdk
      - ANDROID_CLT_VERSION: 11479570 # v13, latest compatible with JDK 17+
        JDK_PACKAGE: openjdk-17-jdk
      - ANDROID_CLT_VERSION: 11479570 # v13, latest compatible with JDK 17+
        JDK_PACKAGE: openjdk-21-jdk
      info_script:
        - echo "Building Android SDK with Android Command-line tools $ANDROID_CLT_VERSION and JDK $JDK_PACKAGE"

The above will generate 4 tasks.

Task Execution Dependencies

Sometimes it might be very handy to execute some tasks only after successful execution of other tasks. For such cases it is possible to specify task names that a particular task depends. Use depends_on keyword to define dependencies:

amd64arm64

container:
  image: node:latest

lint_task:
  script: yarn run lint

test_task:
  script: yarn run test

publish_task:
  depends_on:
    - test
    - lint
  script: yarn run publish

arm_container:
  image: node:latest

lint_task:
  script: yarn run lint

test_task:
  script: yarn run test

publish_task:
  depends_on:
    - test
    - lint
  script: yarn run publish

Task Names and Aliases

It is possible to specify the task's name via the name field. lint_task syntax is a syntactic sugar that will be expanded into:

task:
  name: lint
  ...

Names can be also pretty complex:

task:
  name: Test Shard $TESTS_SPLIT
  env:
    matrix:
      TESTS_SPLIT: 1/3
      TESTS_SPLIT: 2/2
      TESTS_SPLIT: 3/3
  tests_script: ./.ci/tests.sh

deploy_task:
  only_if: $CIRRUS_BRANCH == 'master'
  depends_on:
    - Test Shard 1/3
    - Test Shard 2/3
    - Test Shard 3/3
  script: ./.ci/deploy.sh
  ...

Complex task names make it difficult to list and maintain all of such task names in your depends_on field. In order to make it simpler you can use the alias field to have a short simplified name for several tasks to use in depends_on.

Here is a modified version of an example above that leverages the alias field:

task:
  name: Test Shard $TESTS_SPLIT
  alias: Tests
  env:
    matrix:
      TESTS_SPLIT: 1/3
      TESTS_SPLIT: 2/2
      TESTS_SPLIT: 3/3
  tests_script: ./.ci/tests.sh

deploy_task:
  only_if: $CIRRUS_BRANCH == 'master'
  depends_on: Tests
  script: ./.ci/deploy.sh

Conditional Task Execution

Some tasks are meant to be created only if a certain condition is met. And some tasks can be skipped in some cases. Cirrus CI supports the only_if and skip keywords in order to provide such flexibility:

• The only_if keyword controls whether or not a task will be created. For example, you may want to publish only changes committed to the master branch.

  publish_task:
    only_if: $CIRRUS_BRANCH == 'master'
    script: yarn run publish
  

• The skip keyword allows to skip execution of a task and mark it as successful. For example, you may want to skip linting if no source files have changed since the last successful run.

  lint_task:
    skip: "!changesInclude('.cirrus.yml', '**.{js,ts}')"
    script: yarn run lint
  

Skip CI Completely

Just include [skip ci] or [skip cirrus] in the first line or last line of your commit message in order to skip CI execution for a commit completely.

If you push multiple commits at the same time, only the last commit message will be checked for [skip ci] or [ci skip].

If you open a PR, PR title will be checked for [skip ci] or [ci skip] instead of the last commit message on the PR branch.

Supported Operators

Currently only basic operators like ==, !=, =~, !=~, &&, || and unary ! are supported in only_if and skip expressions. Environment variables can also be used as usual.

Pattern Matching Example

Use =~ operator for pattern matching.

check_aggreement_task:
  only_if: $CIRRUS_BRANCH =~ 'pull/.*'

Note that =~ operator can match against multiline values (dotall mode) and therefore looking for the exact occurrence of the regular expression so don't forget to use .* around your term for matching it at any position (for example, $CIRRUS_CHANGE_TITLE =~ '.*\[docs\].*').

Supported Functions

Currently two functions are supported in the only_if and skip expressions:

• changesInclude function allows to check which files were changed
• changesIncludeOnly is a more strict version of changesInclude, i.e. it won't evaluate to true if there are changed files other than the ones covered by patterns

These two functions behave differently for PR builds and regular builds:

• For PR builds, functions check the list of files affected by the PR.
• For regular builds, the CIRRUS_LAST_GREEN_CHANGE environment variable will be used to determine list of affected files between CIRRUS_LAST_GREEN_CHANGE and CIRRUS_CHANGE_IN_REPO. In case CIRRUS_LAST_GREEN_CHANGE is not available (either it's a new branch or there were no passing builds before), list of files affected by a commit associated with CIRRUS_CHANGE_IN_REPO environment variable will be used instead.

changesInclude function can be very useful for skipping some tasks when no changes to sources have been made since the last successful Cirrus CI build.

lint_task:
  skip: "!changesInclude('.cirrus.yml', '**.{js,ts}')"
  script: yarn run lint

changesIncludeOnly function can be used to skip running a heavyweight task if only documentation was changed, for example:

build_task:
  skip: "changesIncludeOnly('doc/*')"

Auto-Cancellation of Tasks

Cirrus CI can automatically cancel tasks in case of new pushes to the same branch. By default, Cirrus CI auto-cancels all tasks for non default branch (for most repositories master branch) but this behavior can be changed by specifying auto_cancellation field:

task:
  auto_cancellation: $CIRRUS_BRANCH != 'master' && $CIRRUS_BRANCH !=~ 'release/.*'
  ...

Stateful Tasks

It's possible to tell Cirrus CI that a certain task is stateful and Cirrus CI will use a slightly different scheduling algorithm to minimize chances of such tasks being interrupted. Stateful tasks are intended to use low CPU count. Scheduling times of such stateful tasks might be a bit longer then usual especially for tasks with high CPU requirements.

By default, Cirrus CI marks a task as stateful if its name contains one of the following terms: deploy, push, publish, upload or release. Otherwise, you can explicitly mark a task as stateful via stateful field:

task:
  name: Propagate to Production
  stateful: true
  ...

Failure Toleration

Sometimes tasks can play a role of sanity checks. For example, a task can check that your library is working with the latest nightly version of some dependency package. It will be great to be notified about such failures but it's not necessary to fail the whole build when a failure occurs. Cirrus CI has the allow_failures keyword which will make a task to not affect the overall status of a build.

test_nightly_task:
  allow_failures: $SOME_PACKAGE_DEPENDENCY_VERSION == 'nightly'

Skipping Notifications

You can also skip posting red statuses to GitHub via skip_notifications field.

skip_notifications: $SOME_PACKAGE_DEPENDENCY_VERSION == 'nightly'

It can help to track potential issues overtime without distracting the main workflow.

Manual tasks

By default a Cirrus CI task is automatically triggered when all its dependency tasks finished successfully. Sometimes though, it can be very handy to trigger some tasks manually, for example, perform a deployment to staging for manual testing upon all automation checks have succeeded. In order change the default behavior please use trigger_type field like this:

task:
  name: "Staging Deploy"
  trigger_type: manual
  depends_on:
    - Tests (Unit)
    - Tests (Ingegration)
    - Lint

You'll be able to manually trigger such paused tasks via Cirrus CI Web UI or directly from GitHub Checks page.

Task Execution Lock

Some CI tasks perform external operations which are required to be executed one at a time. For example, parallel deploys to the same environment is usually a bad idea. In order to restrict parallel execution of a certain task within a repository, you can use execution_lock to specify a task's lock key, a unique string that will be used to make sure that any tasks with the same execution_lock string are executed one at a time. Here is an example of how to make sure deployments on a specific branch can not run in parallel:

task:
  name: "Automatic Staging Deploy"
  execution_lock: $CIRRUS_BRANCH

You'll be able to manually trigger such paused tasks via the Cirrus CI Web Dashboard or directly from the commit's checks page on GitHub.

Required PR Labels

Similar to manual tasks Cirrus CI can pause execution of tasks until a corresponding PR gets labeled. This can be particular useful when you'd like to do an initial review before running all unit and integration tests on every supported platform. Use the required_pr_labels field to specify a list of labels a PR requires to have in order to trigger a task. Here is a simple example of .cirrus.yml config that automatically runs a linting tool but requires initial-review label being presented in order to run tests:

lint_task:
  # ...

test_task:
  required_pr_labels: initial-review
  # ...

Note: required_pr_labels has no effect on tasks created for non-PR builds.

You can also require multiple labels to continue executing the task for even more flexibility:

deploy_task:
  required_pr_labels: 
    - initial-review
    - ready-for-staging
  depends_on: build
  # ...

In the example above both initial-review and ready-for-staging labels should be presented on a PR in order to perform a deployment via deploy task.

HTTP Cache

For the most cases regular caching mechanism where Cirrus CI caches a folder is more than enough. But modern build systems like Gradle, Bazel and Pants can take advantage of remote caching. Remote caching is when a build system uploads and downloads intermediate results of a build execution while the build itself is still executing.

Cirrus CI agent starts a local caching server and exposes it via CIRRUS_HTTP_CACHE_HOST environments variable. Caching server supports GET, POST, HEAD and DELETE requests to upload, download, check presence and delete artifacts.

Info

If port 12321 is available CIRRUS_HTTP_CACHE_HOST will be equal to localhost:12321.

For example running the following command:

curl -s -X POST --data-binary @myfolder.tar.gz http://$CIRRUS_HTTP_CACHE_HOST/name-key

...has the same effect as the following caching instruction:

name_cache:
  folder: myfolder
  fingerprint_key: key

Info

To see how HTTP Cache can be used with Gradle's Build Cache please check this example.

Additional Containers

Sometimes one container is not enough to run a CI build. For example, your application might use a MySQL database as a storage. In this case you most likely want a MySQL instance running for your tests.

One option here is to pre-install MySQL and use a background_script to start it. This approach has some inconveniences like the need to pre-install MySQL by building a custom Docker container.

For such use cases Cirrus CI allows to run additional containers in parallel with the main container that executes a task. Each additional container is defined under additional_containers keyword in .cirrus.yml. Each additional container should have a unique name and specify at least a container image.

Normally, you would also specify a port (or ports, if there are many) to instruct the Cirrus CI to configure the networking between the containers and wait for the ports to be available before running the task. Additional containers do not inherit environment variables because they are started before the main task receives it's environment variables.

In the example below we use an official MySQL Docker image that exposes the standard MySQL port (3306). Tests will be able to access MySQL instance via localhost:3306.

amd64arm64

container:
  image: golang:latest
  additional_containers:
    - name: mysql
      image: mysql:latest
      port: 3306
      cpu: 1.0
      memory: 512Mi
      env:
        MYSQL_ROOT_PASSWORD: ""
        MYSQL_ALLOW_EMPTY_PASSWORD: "yes"

arm_container:
  image: golang:latest
  additional_containers:
    - name: mysql
      image: mysql:latest
      port: 3306
      cpu: 1.0
      memory: 512Mi
      env:
        MYSQL_ROOT_PASSWORD: ""
        MYSQL_ALLOW_EMPTY_PASSWORD: "yes"

Additional container can be very handy in many scenarios. Please check Cirrus CI catalog of examples for more details.

Default Resources

By default, each additional container will get 0.5 CPU and 512Mi of memory. These values can be configured as usual via cpu and memory fields.

Port Mapping

It's also possible to map ports of additional containers by using <HOST_PORT>:<CONTAINER_PORT> format for the port field. For example, port: 80:8080 will map port 8080 of the container to be available on local port 80 within a task.

Note: don't use port mapping unless absolutely necessary. A perfect use case is when you have several additional containers which start the service on the same port and there's no easy way to change that. Port mapping limits the number of places the container can be scheduled and will affect how fast such tasks are scheduled.

To specify multiple mappings use the ports field, instead of the port:

ports:
  - 8080
  - 3306

Overriding Default Command

It's also possible to override the default CMD of an additional container via command field:

amd64arm64

container:
  image: golang:latest
  additional_containers:
    - name: mysql
      image: mysql:latest
      port: 7777
      command: mysqld --port 7777
      env:
        MYSQL_ROOT_PASSWORD: ""
        MYSQL_ALLOW_EMPTY_PASSWORD: "yes"

arm_container:
  image: golang:latest
  additional_containers:
    - name: mysql
      image: mysql:latest
      port: 7777
      command: mysqld --port 7777
      env:
        MYSQL_ROOT_PASSWORD: ""
        MYSQL_ALLOW_EMPTY_PASSWORD: "yes"

Warning

Note that additional_containers can be used only with the Linux Clusters, a GKE cluster or a EKS cluster.

Embedded Badges

Cirrus CI provides a way to embed a badge that can represent status of your builds into a ReadMe file or a website.

For example, this is a badge for cirruslabs/cirrus-ci-web repository that contains Cirrus CI's front end:

In order to embed such a check into a "read-me" file or your website, just use a URL to a badge that looks like this:

https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>.svg

If you want a badge for a particular branch, use the ?branch=<BRANCH NAME> query parameter (at the end of the URL) like this:

https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>.svg?branch=<BRANCH NAME>

By default, Cirrus picks the latest build in a final state for the repository or a particular branch if branch parameter is specified. It's also possible to explicitly set a concrete build to use with ?buildId=<BUILD ID> query parameter.

If you want a badge for a particular task within the latest finished build, use the ?task=<TASK NAME> query parameter (at the end of the URL) like this:

https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>.svg?task=tests

You can even pick a specific script instruction within the task with an additional script=<SCRIPT NAME> parameter:

https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>.svg?task=build&script=lint

Badges in Markdown

Here is how Cirrus CI's badge can be embeded in a Markdown file:

[![Build Status](https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>.svg)](https://cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>)

CCTray XML

Cirrus CI supports exporting information about the latest repository builds via the CCTray XML format, using the following URL format:

https://api.cirrus-ci.com/github/<USER OR ORGANIZATION>/<REPOSITORY>/cctray.xml

Some tools with support of CCtray are:

• CCMenu (macOS Native build status monitor).
• Barklarm (Open Source multiplatform alarm and build status monitor).
• Nevergreen (Build radiation service).

Note: for private repositories you'll need to configure access token.