• Custom buildpacks
  • Custom buildpacks with Cloud Native Buildpacks
  • Custom buildpacks with Herokuish
  • Multiple buildpacks
• Custom Dockerfile
• Passing arguments to docker build
• Extend Auto DevOps with the API
• Forward CI/CD variables to the build environment
• Custom Helm Chart
• Customize values for Helm Chart
• Customize the helm upgrade command
• Custom Helm chart per environment
• Customizing .gitlab-ci.yml
• Customizing the Kubernetes namespace
• Using components of Auto DevOps
• Use images hosted in a local Docker registry
• PostgreSQL database support
  • Upgrading PostgresSQL
  • Using external PostgreSQL database providers
• CI/CD variables
  • Build and deployment
  • Database
  • Disable jobs
  • Application secret variables
  • Advanced replica variables setup
  • Deploy policy for staging and production environments
  • Deploy policy for canary environments
  • Incremental rollout to production
  • Timed incremental rollout to production
• Auto DevOps banner

Customizing Auto DevOps

While Auto DevOps provides great defaults to get you started, you can customize almost everything to fit your needs. Auto DevOps offers everything from custom buildpacks, to Dockerfiles, and Helm charts. You can even copy the complete CI/CD configuration into your project to enable staging and canary deployments, manage Auto DevOps with GitLab APIs, and more.

Custom buildpacks

If the automatic buildpack detection fails for your project, or if you need more control over your build, you can customize the buildpacks used for the build.

Custom buildpacks with Cloud Native Buildpacks

Specify either:

• The CI/CD variable BUILDPACK_URL according to pack’s specifications.
• A project.toml project descriptor with the buildpacks you would like to include.

Custom buildpacks with Herokuish

Specify either:

• The CI/CD variable BUILDPACK_URL.
• A .buildpacks file at the root of your project, containing one buildpack URL per line.

The buildpack URL can point to either a Git repository URL or a tarball URL. For Git repositories, you can point to a specific Git reference (such as commit SHA, tag name, or branch name) by appending #<ref> to the Git repository URL. For example:

• The tag v142: https://github.com/heroku/heroku-buildpack-ruby.git#v142.
• The branch mybranch: https://github.com/heroku/heroku-buildpack-ruby.git#mybranch.
• The commit SHA f97d8a8ab49: https://github.com/heroku/heroku-buildpack-ruby.git#f97d8a8ab49.

Multiple buildpacks

Using multiple buildpacks is not fully supported by Auto DevOps, because Auto Test can’t use the .buildpacks file. The buildpack heroku-buildpack-multi, used in the backend to parse the .buildpacks file, does not provide the necessary commands bin/test-compile and bin/test.

If your goal is to use only a single custom buildpack, you should provide the project CI/CD variable BUILDPACK_URL instead.

Custom Dockerfile

Support for DOCKERFILE_PATH was added in GitLab 13.2

If your project has a Dockerfile in the root of the project repository, Auto DevOps builds a Docker image based on the Dockerfile, rather than using buildpacks. This can be much faster and result in smaller images, especially if your Dockerfile is based on Alpine.

If you set the DOCKERFILE_PATH CI/CD variable, Auto Build looks for a Dockerfile there instead.

Passing arguments to docker build

Arguments can be passed to the docker build command using the AUTO_DEVOPS_BUILD_IMAGE_EXTRA_ARGS project CI/CD variable. For example, to build a Docker image based on based on the ruby:alpine instead of the default ruby:latest:

1. Set AUTO_DEVOPS_BUILD_IMAGE_EXTRA_ARGS to --build-arg=RUBY_VERSION=alpine.

2. Add the following to a custom Dockerfile:

   ARG RUBY_VERSION=latest
   FROM ruby:$RUBY_VERSION
   
   # ... put your stuff here
   

Use Base64 encoding if you need to pass complex values, such as newlines and spaces. Left unencoded, complex values like these can cause escaping issues due to how Auto DevOps uses the arguments.

Extend Auto DevOps with the API

You can extend and manage your Auto DevOps configuration with GitLab APIs:

• Settings that can be accessed with API calls, which include auto_devops_enabled, to enable Auto DevOps on projects by default.
• Creating a new project.
• Editing groups.
• Editing projects.

Forward CI/CD variables to the build environment

CI/CD variables can be forwarded into the build environment using the AUTO_DEVOPS_BUILD_IMAGE_FORWARDED_CI_VARIABLES CI/CD variable. The forwarded variables should be specified by name in a comma-separated list. For example, to forward the variables CI_COMMIT_SHA and CI_ENVIRONMENT_NAME, set AUTO_DEVOPS_BUILD_IMAGE_FORWARDED_CI_VARIABLES to CI_COMMIT_SHA,CI_ENVIRONMENT_NAME.

• When using Buildpacks, the forwarded variables are available automatically as environment variables.

• When using a Dockerfile, the following additional steps are required:

  1. Activate the experimental Dockerfile syntax by adding the following code to the top of the file:

     # syntax = docker/dockerfile:experimental
     

  2. To make secrets available in any RUN $COMMAND in the Dockerfile, mount the secret file and source it prior to running $COMMAND:

     RUN --mount=type=secret,id=auto-devops-build-secrets . /run/secrets/auto-devops-build-secrets && $COMMAND
     

When AUTO_DEVOPS_BUILD_IMAGE_FORWARDED_CI_VARIABLES is set, Auto DevOps enables the experimental Docker BuildKit feature to use the --secret flag.

Custom Helm Chart

Auto DevOps uses Helm to deploy your application to Kubernetes. You can override the Helm chart used by bundling up a chart into your project repository or by specifying a project CI/CD variable:

• Bundled chart - If your project has a ./chart directory with a Chart.yaml file in it, Auto DevOps detects the chart and uses it instead of the default chart, enabling you to control exactly how your application is deployed.
• Project variable - Create a project CI/CD variable AUTO_DEVOPS_CHART with the URL of a custom chart to use, or create two project variables: AUTO_DEVOPS_CHART_REPOSITORY with the URL of a custom chart repository, and AUTO_DEVOPS_CHART with the path to the chart.

Customize values for Helm Chart

You can override the default values in the values.yaml file in the default Helm chart by either:

• Adding a file named .gitlab/auto-deploy-values.yaml to your repository, which is automatically used, if found.
• Adding a file with a different name or path to the repository, and setting the HELM_UPGRADE_VALUES_FILE CI/CD variable with the path and name.

Customize the helm upgrade command

You can customize the helm upgrade command used in the auto-deploy-image by passing options to the command with the HELM_UPGRADE_EXTRA_ARGS CI/CD variable. For example, set the value of HELM_UPGRADE_EXTRA_ARGS to --no-hooks to disable pre-upgrade and post-upgrade hooks when the command is executed.

See the official documentation for the full list of options.

Custom Helm chart per environment

You can specify the use of a custom Helm chart per environment by scoping the CI/CD variable to the desired environment. See Limit environment scope of CI/CD variables.

Customizing .gitlab-ci.yml

Auto DevOps is completely customizable because the Auto DevOps template is just an implementation of a .gitlab-ci.yml file, and uses only features available to any implementation of .gitlab-ci.yml.

To modify the CI/CD pipeline used by Auto DevOps, include the template, and customize it as needed by adding a .gitlab-ci.yml file to the root of your repository containing the following:

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

Add your changes, and your additions are merged with the Auto DevOps template using the behavior described for include.

If you need to specifically remove a part of the file, you can also copy and paste the contents of the Auto DevOps template into your project and edit it as needed.

Customizing the Kubernetes namespace

For clusters not managed by GitLab, you can customize the namespace in .gitlab-ci.yml by specifying environment:kubernetes:namespace. For example, the following configuration overrides the namespace used for production deployments:

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

production:
  environment:
    kubernetes:
      namespace: production

When deploying to a custom namespace with Auto DevOps, the service account provided with the cluster needs at least the edit role within the namespace.

• If the service account can create namespaces, then the namespace can be created on-demand.
• Otherwise, the namespace must exist prior to deployment.

Using components of Auto DevOps

If you only require a subset of the features offered by Auto DevOps, you can include individual Auto DevOps jobs into your own .gitlab-ci.yml. Each component job relies on a stage that should be defined in the .gitlab-ci.yml that includes the template.

For example, to make use of Auto Build, you can add the following to your .gitlab-ci.yml:

stages:
  - build

include:
  - template: Jobs/Build.gitlab-ci.yml

See the Auto DevOps template for information on available jobs.

Use images hosted in a local Docker registry

You can configure many Auto DevOps jobs to run in an offline environment:

1. Copy the required Auto DevOps Docker images from Docker Hub and registry.gitlab.com to their local GitLab container registry.

2. After the images are hosted and available in a local registry, edit .gitlab-ci.yml to point to the locally-hosted images. For example:

   include:
     - template: Auto-DevOps.gitlab-ci.yml
   
   variables:
     REGISTRY_URL: "registry.gitlab.example"
   
   build:
     image: "$REGISTRY_URL/docker/auto-build-image:v0.6.0"
     services:
       - name: "$REGISTRY_URL/greg/docker/docker:20.10.6-dind"
         command: ['--tls=false', '--host=tcp://0.0.0.0:2375']
   

PostgreSQL database support

To support applications requiring a database, PostgreSQL is provisioned by default. The credentials to access the database are preconfigured, but can be customized by setting the associated CI/CD variables. You can use these credentials to define a DATABASE_URL:

postgres://user:password@postgres-host:postgres-port/postgres-database

Upgrading PostgresSQL

The version of the chart used to provision PostgreSQL:

• Is 8.2.1 in GitLab 13.0 and later, but can be set back to 0.7.1 if needed.
• Can be set to from 0.7.1 to 8.2.1 in GitLab 12.9 and 12.10.
• Is 0.7.1 in GitLab 12.8 and earlier.

GitLab encourages users to migrate their database to the newer PostgreSQL.

Using external PostgreSQL database providers

While Auto DevOps provides out-of-the-box support for a PostgreSQL container for production environments, for some use cases, it may not be sufficiently secure or resilient, and you may want to use an external managed provider (such as AWS Relational Database Service) for PostgreSQL.

You must define environment-scoped CI/CD variables for POSTGRES_ENABLED and DATABASE_URL in your project’s CI/CD settings:

1. Disable the built-in PostgreSQL installation for the required environments using environment-scoped CI/CD variables. For this use case, it’s likely that only production must be added to this list. The built-in PostgreSQL setup for Review Apps and staging is sufficient.

   

2. Define the DATABASE_URL variable as an environment-scoped variable that is available to your application. This should be a URL in the following format:

   postgres://user:password@postgres-host:postgres-port/postgres-database
   

You must ensure that your Kubernetes cluster has network access to wherever PostgreSQL is hosted.

CI/CD variables

The following variables can be used for setting up the Auto DevOps domain, providing a custom Helm chart, or scaling your application. PostgreSQL can also be customized, and you can use a custom buildpack.

Build and deployment

The following table lists CI/CD variables related to building and deploying applications.

Database

The following table lists CI/CD variables related to the database.

Disable jobs

The following table lists variables used to disable jobs.

Application secret variables

Some applications need to define secret variables that are accessible by the deployed application. Auto DevOps detects CI/CD variables starting with K8S_SECRET_, and makes these prefixed variables available to the deployed application as environment variables.

To configure your application variables:

1. Go to your project’s Settings > CI/CD, then expand the Variables section.

2. Create a CI/CD variable, ensuring the key is prefixed with K8S_SECRET_. For example, you can create a variable with key K8S_SECRET_RAILS_MASTER_KEY.

3. Run an Auto DevOps pipeline, either by manually creating a new pipeline or by pushing a code change to GitLab.

Auto DevOps pipelines take your application secret variables to populate a Kubernetes secret. This secret is unique per environment. When deploying your application, the secret is loaded as environment variables in the container running the application. Following the example above, you can see the secret below containing the RAILS_MASTER_KEY variable.

$ kubectl get secret production-secret -n minimal-ruby-app-54 -o yaml

apiVersion: v1
data:
  RAILS_MASTER_KEY: MTIzNC10ZXN0
kind: Secret
metadata:
  creationTimestamp: 2018-12-20T01:48:26Z
  name: production-secret
  namespace: minimal-ruby-app-54
  resourceVersion: "429422"
  selfLink: /api/v1/namespaces/minimal-ruby-app-54/secrets/production-secret
  uid: 57ac2bfd-03f9-11e9-b812-42010a9400e4
type: Opaque

Environment variables are generally considered immutable in a Kubernetes pod. If you update an application secret without changing any code, then manually create a new pipeline, any running application pods don’t receive the updated secrets. To update the secrets, either:

• Push a code update to GitLab to force the Kubernetes deployment to recreate pods.
• Manually delete running pods to cause Kubernetes to create new pods with updated secrets.

Variables with multi-line values are not currently supported due to limitations with the current Auto DevOps scripting environment.

Advanced replica variables setup

Apart from the two replica-related variables for production mentioned above, you can also use other variables for different environments.

The Kubernetes’ label named track, GitLab CI/CD environment names, and the replicas environment variable are combined into the format TRACK_ENV_REPLICAS, enabling you to define your own variables for scaling the pod’s replicas:

• TRACK: The capitalized value of the track Kubernetes label in the Helm Chart app definition. If not set, it isn’t taken into account to the variable name.
• ENV: The capitalized environment name of the deploy job, set in .gitlab-ci.yml.

In the example below, the environment’s name is qa, and it deploys the track foo, which results in an environment variable named FOO_QA_REPLICAS:

QA testing:
  stage: deploy
  environment:
    name: qa
  script:
    - deploy foo

The track foo being referenced must also be defined in the application’s Helm chart, like:

replicaCount: 1
image:
  repository: gitlab.example.com/group/project
  tag: stable
  pullPolicy: Always
  secrets:
    - name: gitlab-registry
application:
  track: foo
  tier: web
service:
  enabled: true
  name: web
  type: ClusterIP
  url: http://my.host.com/
  externalPort: 5000
  internalPort: 5000

Deploy policy for staging and production environments

The normal behavior of Auto DevOps is to use continuous deployment, pushing automatically to the production environment every time a new pipeline is run on the default branch. However, there are cases where you might want to use a staging environment, and deploy to production manually. For this scenario, the STAGING_ENABLED CI/CD variable was introduced.

If you define STAGING_ENABLED with a non-empty value, then GitLab automatically deploys the application to a staging environment, and creates a production_manual job for you when you’re ready to manually deploy to production.

Deploy policy for canary environments

You can use a canary environment before deploying any changes to production.

If you define CANARY_ENABLED with a non-empty value, then two manual jobs are created:

• canary - Deploys the application to the canary environment.
• production_manual - Manually deploys the application to production.

Incremental rollout to production

When you’re ready to deploy a new version of your app to production, you may want to use an incremental rollout to replace just a few pods with the latest code to check how the application is behaving before manually increasing the rollout up to 100%.

If INCREMENTAL_ROLLOUT_MODE is set to manual in your project, then instead of the standard production job, 4 different manual jobs are created:

1. rollout 10%
2. rollout 25%
3. rollout 50%
4. rollout 100%

The percentage is based on the REPLICAS CI/CD variable, and defines the number of pods you want to have for your deployment. If the value is 10, and you run the 10% rollout job, there is 1 new pod and 9 old ones.

To start a job, click the play icon () next to the job’s name. You’re not required to go from 10% to 100%, you can jump to whatever job you want. You can also scale down by running a lower percentage job, just before hitting 100%. Once you get to 100%, you can’t scale down, and you’d have to roll back by redeploying the old version using the rollback button in the environment page.

Below, you can see how the pipeline appears if the rollout or staging variables are defined.

Without INCREMENTAL_ROLLOUT_MODE and without STAGING_ENABLED:

Without INCREMENTAL_ROLLOUT_MODE and with STAGING_ENABLED:

With INCREMENTAL_ROLLOUT_MODE set to manual and without STAGING_ENABLED:

With INCREMENTAL_ROLLOUT_MODE set to manual and with STAGING_ENABLED

Timed incremental rollout to production

This configuration is based on incremental rollout to production.

Everything behaves the same way, except:

• To enable it, set the INCREMENTAL_ROLLOUT_MODE CI/CD variable to timed.

• Instead of the standard production job, the following jobs are created with a 5 minute delay between each:

  1. timed rollout 10%
  2. timed rollout 25%
  3. timed rollout 50%
  4. timed rollout 100%

Auto DevOps banner

The following Auto DevOps banner displays for users with Maintainer or greater permissions on new projects when Auto DevOps is not enabled:

The banner can be disabled for:

• A user, when they dismiss it themselves.
• A project, by explicitly disabling Auto DevOps.
• An entire GitLab instance:

  • By an administrator running the following in a Rails console:

    Feature.enable(:auto_devops_banner_disabled)
    

  • Through the REST API with an admin access token:

    curl --data "value=true" --header "PRIVATE-TOKEN: <personal_access_token>" "https://gitlab.example.com/api/v4/features/auto_devops_banner_disabled"