GitLab Runner feature flags

caution
Data corruption, stability degradation, performance degradation, and security issues may occur if you enable a feature that’s disabled by default. Before you enable feature flags, you should be aware of the risks involved. For more information, see Risks when enabling features still in development.

Feature flags are toggles that allow you to enable or disable specific features. These flags are typically used:

  • For beta features that are made available for volunteers to test, but that are not ready to be enabled for all users.

    Beta features are sometimes incomplete or need further testing. A user who wants to use a beta feature can choose to accept the risk and explicitly enable the feature with a feature flag. Other users who do not need the feature or who are not willing to accept the risk on their system have the feature disabled by default and are not impacted by possible bugs and regressions.

  • For breaking changes that result in functionality deprecation or feature removal in the near future.

    As the product evolves, features are sometimes changed or removed entirely. Known bugs are often fixed, but in some cases, users have already found a workaround for a bug that affected them; forcing users to adopt the standardized bug fix might cause other problems with their customized configurations.

    In such cases, the feature flag is used to switch from the old behavior to the new one on demand. This allows users to adopt new versions of the product while giving them time to plan for a smooth, permanent transition from the old behavior to the new behavior.

Feature flags are toggled using environment variables. To:

  • Activate a feature flag, set the corresponding environment variable to "true" or 1.
  • Deactivate a feature flag, set the corresponding environment variable to "false" or 0.

Available feature flags

Feature flag Default value Deprecated To be removed with Description
FF_NETWORK_PER_BUILD false No   Enables creation of a Docker network per build with the docker executor
FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY false No   When set to false disables execution of remote Kubernetes commands through exec in favor of attach to solve problems like #4119
FF_USE_DIRECT_DOWNLOAD true No   When set to true Runner tries to direct-download all artifacts instead of proxying through GitLab on a first try. Enabling might result in a download failures due to problem validating TLS certificate of Object Storage if it is enabled by GitLab. See Self-signed certificates or custom Certification Authorities
FF_SKIP_NOOP_BUILD_STAGES true No   When set to false all build stages are executed even if running them has no effect
FF_USE_FASTZIP false No   Fastzip is a performant archiver for cache/artifact archiving and extraction
FF_DISABLE_UMASK_FOR_DOCKER_EXECUTOR false No   If enabled will remove the usage of umask 0000 call for jobs executed with docker executor. Instead Runner will try to discover the UID and GID of the user configured for the image used by the build container and will change the ownership of the working directory and files by running the chmod command in the predefined container (after updating sources, restoring cache and downloading artifacts). POSIX utility id must be installed and operational in the build image for this feature flag. Runner will execute id with options -u and -g to retrieve the UID and GID.
FF_ENABLE_BASH_EXIT_CODE_CHECK false No   If enabled, bash scripts don’t rely solely on set -e, but check for a non-zero exit code after each script command is executed.
FF_USE_WINDOWS_LEGACY_PROCESS_STRATEGY false No   In GitLab Runner 16.10 and later, the default is false. In GitLab Runner 16.9 and earlier, the default is true. When disabled, processes that Runner creates on Windows (shell and custom executor) will be created with additional setup that should improve process termination. When set to true, legacy process setup is used. To successfully and gracefully drain a Windows Runner, this feature flag should be set to false.
FF_USE_NEW_BASH_EVAL_STRATEGY false No   When set to true, the Bash eval call is executed in a subshell to help with proper exit code detection of the script executed.
FF_USE_POWERSHELL_PATH_RESOLVER false No   When enabled, PowerShell resolves pathnames rather than Runner using OS-specific filepath functions that are specific to where Runner is hosted.
FF_USE_DYNAMIC_TRACE_FORCE_SEND_INTERVAL false No   When enabled, the trace force send interval for logs is dynamically adjusted based on the trace update interval.
FF_SCRIPT_SECTIONS false No   When enabled, each script line from the .gitlab-ci.yml file is in a collapsible section in the job output, and shows the duration of each line. When the command spans multiple lines, the complete command is displayed within the job log output terminal.
FF_ENABLE_JOB_CLEANUP false No   When enabled, the project directory will be cleaned up at the end of the build. If GIT_CLONE is used, the whole project directory will be deleted. If GIT_FETCH is used, a series of Git clean commands will be issued.
FF_KUBERNETES_HONOR_ENTRYPOINT false No   When enabled, the Docker entrypoint of an image will be honored if FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY is not set to true
FF_POSIXLY_CORRECT_ESCAPES false No   When enabled, POSIX shell escapes are used rather than bash-style ANSI-C quoting. This should be enabled if the job environment uses a POSIX-compliant shell.
FF_RESOLVE_FULL_TLS_CHAIN false No   In GitLab Runner 16.4 and later, the default is false. In GitLab Runner 16.3 and earlier, the default is true. When enabled, the runner resolves a full TLS chain all the way down to a self-signed root certificate for CI_SERVER_TLS_CA_FILE. This was previously required to make Git HTTPS clones work for a Git client built with libcurl prior to v7.68.0 and OpenSSL. However, the process to resolve certificates might fail on some operating systems, such as macOS, that reject root certificates signed with older signature algorithms. If certificate resolution fails, you might need to disable this feature. This feature flag can only be disabled in the [runners.feature_flags] configuration.
FF_DISABLE_POWERSHELL_STDIN false No   When enabled, PowerShell scripts for shell and custom executors are passed by file, rather than passed and executed via stdin. This is required for jobs’ allow_failure:exit_codes keywords to work correctly.
FF_USE_POD_ACTIVE_DEADLINE_SECONDS true No   When enabled, the pod activeDeadlineSeconds is set to the CI/CD job timeout. This flag affects the pod’s lifecycle.
FF_USE_ADVANCED_POD_SPEC_CONFIGURATION false No   When enabled, the user can set an entire whole pod specification in the config.toml file. For more information, see Overwrite generated pod specifications (Experiment).
FF_SET_PERMISSIONS_BEFORE_CLEANUP true No   When enabled, permissions on directories and files in the project directory are set first, to ensure that deletions during cleanup are successful.
FF_SECRET_RESOLVING_FAILS_IF_MISSING true No   When enabled, secret resolving fails if the value cannot be found.
FF_RETRIEVE_POD_WARNING_EVENTS true Yes   When enabled, all warning events associated with the pod are retrieved when the job fails. This flag was deprecated in GitLab Runner 17.2 and is planned for removal in 18.0. In GitLab Runner 17.2 and later, pod warning events are retrieved and logged unconditionally. However, no failure occurs even if RBAC permissions are missing.
FF_PRINT_POD_EVENTS false No   When enabled, all events associated with the build pod will be printed until it’s started.
FF_USE_GIT_BUNDLE_URIS true No   When enabled, the Git transfer.bundleURI configuration option is set to true. This FF is enabled by default. Set to false to disable Git bundle support.
FF_USE_DUMB_INIT_WITH_KUBERNETES_EXECUTOR false No   When enabled, dumb-init is used to execute all the scripts. This allows dumb-init to run as the first process in the helper and build container.
FF_USE_INIT_WITH_DOCKER_EXECUTOR false No   When enabled, the Docker executor starts the service and build containers with the --init option, which runs tini-init as PID 1.
FF_LOG_IMAGES_CONFIGURED_FOR_JOB false No   When enabled, the runner logs names of the image and service images defined for each received job.
FF_USE_DOCKER_AUTOSCALER_DIAL_STDIO true No   When enabled (the default), docker system stdio is used to tunnel to the remote Docker daemon. When disabled, for SSH connections a native SSH tunnel is used, and for WinRM connections a ‘fleeting-proxy’ helper binary is first deployed.
FF_CLEAN_UP_FAILED_CACHE_EXTRACT false No   When enabled, commands are inserted into build scripts to detect a failed cache extraction and clean up partial cache contents left behind.
FF_USE_WINDOWS_JOB_OBJECT false No   When enabled, a job object is created for each process that the runner creates on Windows with the shell and custom executors. To force-kill the processes, the runner closes the job object. This should improve the termination of difficult-to-kill processes.
FF_TIMESTAMPS false No   When enabled timestamps are added to the beginning of each log trace line.
FF_DISABLE_AUTOMATIC_TOKEN_ROTATION false No   When enabled, it restricts automatic token rotation and logs a warning when the token is about to expire.
FF_USE_LEGACY_GCS_CACHE_ADAPTER false No   When enabled, the legacy GCS Cache adapter is used. When disabled (default), a newer GCS Cache adapter is used which uses Google Cloud Storage’s SDK for authentication. This should resolve authentication problems in environments that the legacy adapter struggled with, such as workload identity configurations in GKE.
FF_DISABLE_UMASK_FOR_KUBERNETES_EXECUTOR false No   When enabled, removes the umask 0000 call for jobs executed with the Kubernetes executor. Instead, the runner tries to discover the user ID (UID) and group ID (GID) of the user the build container runs as. The runner also changes the ownership of the working directory and files by running the chown command in the predefined container (after updating sources, restoring cache, and downloading artifacts).

Enable feature flag in pipeline configuration

You can use CI variables to enable feature flags:

  • For all jobs in the pipeline (globally):

    variables:
      FEATURE_FLAG_NAME: 1
    
  • For a single job:

    job:
      stage: test
      variables:
        FEATURE_FLAG_NAME: 1
      script:
      - echo "Hello"
    

Enable feature flag in runner environment variables

To enable the feature for every job a Runner runs, specify the feature flag as an environment variable in the Runner configuration:

[[runners]]
  name = "ruby-2.7-docker"
  url = "https://CI/"
  token = "TOKEN"
  limit = 0
  executor = "docker"
  builds_dir = ""
  shell = ""
  environment = ["FEATURE_FLAG_NAME=1"]

Enable feature flag in runner configuration

History

You can enable feature flags by specifying them under [runners.feature_flags]. This setting prevents any job from overriding the feature flag values.

Some feature flags are also only usable when you configure this setting, because they don’t deal with how the job is executed.

[[runners]]
  name = "ruby-2.7-docker"
  url = "https://CI/"
  token = "TOKEN"
  executor = "docker"
  [runners.feature_flags]
    FF_USE_DIRECT_DOWNLOAD = true