gitlab-org--gitlab-foss/doc/topics/autodevops/index.md

24 KiB
Raw Blame History

stage group info
none unassigned To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#designated-technical-writers

Auto DevOps

  • Introduced in GitLab 10.0.
  • Generally available on GitLab 11.0.

Auto DevOps are default CI/CD templates that auto-discover the source code you have. They enable GitLab to automatically detect, build, test, deploy, and monitor your applications. Leveraging CI/CD best practices and tools, Auto DevOps aims to simplify the setup and execution of a mature and modern software development lifecycle.

Overview

You can spend a lot of effort to set up the workflow and processes required to build, deploy, and monitor your project. It gets worse when your company has hundreds, if not thousands, of projects to maintain. With new projects constantly starting up, the entire software development process becomes impossibly complex to manage.

Auto DevOps provides you a seamless software development process by automatically detecting all dependencies and language technologies required to test, build, package, deploy, and monitor every project with minimal configuration. Automation enables consistency across your projects, seamless management of processes, and faster creation of new projects: push your code, and GitLab does the rest, improving your productivity and efficiency.

For an introduction to Auto DevOps, watch AutoDevOps in GitLab 11.0.

For requirements, see Requirements for Auto DevOps for more information.

Enabled by default

Introduced in GitLab 11.3.

Auto DevOps is enabled by default for all projects and attempts to run on all pipelines in each project. An instance administrator can enable or disable this default in the Auto DevOps settings. Auto DevOps automatically disables in individual projects on their first pipeline failure, if it has not been explicitly enabled for the project.

Since GitLab 12.7, Auto DevOps runs on pipelines automatically only if a Dockerfile or matching buildpack exists.

If a CI/CD configuration file is present in the project, it will continue to be used, whether or not Auto DevOps is enabled.

Quick start

If you're using GitLab.com, see the quick start guide for setting up Auto DevOps with GitLab.com and a Kubernetes cluster on Google Kubernetes Engine (GKE).

If you use a self-managed instance of GitLab, you must configure the Google OAuth2 OmniAuth Provider before configuring a cluster on GKE. After configuring the provider, you can follow the steps in the quick start guide to get started.

In GitLab 13.0 and later, it is possible to leverage Auto DevOps to deploy to AWS ECS.

Comparison to application platforms and PaaS

Auto DevOps provides features often included in an application platform or a Platform as a Service (PaaS). It takes inspiration from the innovative work done by Heroku and goes beyond it in multiple ways:

  • Auto DevOps works with any Kubernetes cluster; you're not limited to running on GitLab's infrastructure. (Note that many features also work without Kubernetes).
  • There is no additional cost (no markup on the infrastructure costs), and you can use a Kubernetes cluster you host or Containers as a Service on any public cloud (for example, Google Kubernetes Engine).
  • Auto DevOps has more features including security testing, performance testing, and code quality testing.
  • Auto DevOps offers an incremental graduation path. If you need advanced customizations, you can start modifying the templates without starting over on a completely different platform. Review the customizing documentation for more information.

Features

Comprised of a set of stages, Auto DevOps brings these best practices to your project in a simple and automatic way:

  1. Auto Build
  2. Auto Test
  3. Auto Code Quality
  4. Auto SAST (Static Application Security Testing)
  5. Auto Secret Detection
  6. Auto Dependency Scanning (ULTIMATE)
  7. Auto License Compliance (ULTIMATE)
  8. Auto Container Scanning (ULTIMATE)
  9. Auto Review Apps
  10. Auto DAST (Dynamic Application Security Testing) (ULTIMATE)
  11. Auto Deploy
  12. Auto Browser Performance Testing (PREMIUM)
  13. Auto Monitoring
  14. Auto Code Intelligence

As Auto DevOps relies on many different components, you should have a basic knowledge of the following:

Auto DevOps provides great defaults for all the stages and makes use of CI templates. You can, however, customize almost everything to your needs, and manage Auto DevOps with GitLab APIs.

For an overview on the creation of Auto DevOps, read more in this blog post.

NOTE: Note: Kubernetes clusters can be used without Auto DevOps.

Kubernetes requirements

See Auto DevOps requirements for Kubernetes.

Auto DevOps base domain

The Auto DevOps base domain is required to use Auto Review Apps, Auto Deploy, and Auto Monitoring. You can define the base domain in any of the following places:

  • either under the cluster's settings, whether for an instance, projects or groups
  • or at the project level as a variable: KUBE_INGRESS_BASE_DOMAIN
  • or at the group level as a variable: KUBE_INGRESS_BASE_DOMAIN
  • or as an instance-wide fallback in Admin Area > Settings under the Continuous Integration and Delivery section

The base domain variable KUBE_INGRESS_BASE_DOMAIN follows the same order of precedence as other environment variables. If the CI/CD variable is not set and the cluster setting is left blank, the instance-wide Auto DevOps domain setting will be used if set.

TIP: Tip: If you use the GitLab managed app for Ingress, the URL endpoint should be automatically configured for you. All you must do is use its value for the KUBE_INGRESS_BASE_DOMAIN variable.

NOTE: Note: AUTO_DEVOPS_DOMAIN was deprecated in GitLab 11.8 and replaced with KUBE_INGRESS_BASE_DOMAIN, and removed in GitLab 12.0.

Auto DevOps requires a wildcard DNS A record matching the base domain(s). For a base domain of example.com, you'd need a DNS entry like:

*.example.com   3600     A     1.2.3.4

In this case, the deployed applications are served from example.com, and 1.2.3.4 is the IP address of your load balancer; generally NGINX (see requirements). Setting up the DNS record is beyond the scope of this document; check with your DNS provider for information.

Alternatively, you can use free public services like nip.io which provide automatic wildcard DNS without any configuration. For nip.io, set the Auto DevOps base domain to 1.2.3.4.nip.io.

After completing setup, all requests hit the load balancer, which routes requests to the Kubernetes pods running your application.

AWS ECS

See Auto DevOps requirements for Amazon ECS.

Enabling/Disabling Auto DevOps

When first using Auto DevOps, review the requirements to ensure all the necessary components to make full use of Auto DevOps are available. First-time users should follow the quick start guide.

GitLab.com users can enable or disable Auto DevOps only at the project level. Self-managed users can enable or disable Auto DevOps at the project, group, or instance level.

At the project level

If enabling, check that your project does not have a .gitlab-ci.yml, or if one exists, remove it.

  1. Go to your project's Settings > CI/CD > Auto DevOps.
  2. Select the Default to Auto DevOps pipeline checkbox to enable it.
  3. (Optional, but recommended) When enabling, you can add in the base domain Auto DevOps uses to deploy your application, and choose the deployment strategy.
  4. Click Save changes for the changes to take effect.

After enabling the feature, an Auto DevOps pipeline is triggered on the master branch.

At the group level

Introduced in GitLab 11.10.

Only administrators and group owners can enable or disable Auto DevOps at the group level.

When enabling or disabling Auto DevOps at group level, group configuration is implicitly used for the subgroups and projects inside that group, unless Auto DevOps is specifically enabled or disabled on the subgroup or project.

To enable or disable Auto DevOps at the group level:

  1. Go to your group's Settings > CI/CD > Auto DevOps page.
  2. Select the Default to Auto DevOps pipeline checkbox to enable it.
  3. Click Save changes for the changes to take effect.

At the instance level (Administrators only)

Even when disabled at the instance level, group owners and project maintainers can still enable Auto DevOps at the group and project level, respectively.

  1. Go to Admin Area > Settings > Continuous Integration and Deployment.
  2. Select Default to Auto DevOps pipeline for all projects to enable it.
  3. (Optional) You can set up the Auto DevOps base domain, for Auto Deploy and Auto Review Apps to use.
  4. Click Save changes for the changes to take effect.

Deployment strategy

Introduced in GitLab 11.0.

You can change the deployment strategy used by Auto DevOps by going to your project's Settings > CI/CD > Auto DevOps. The following options are available:

  • Continuous deployment to production: Enables Auto Deploy with master branch directly deployed to production.

  • Continuous deployment to production using timed incremental rollout: Sets the INCREMENTAL_ROLLOUT_MODE variable to timed. Production deployments execute with a 5 minute delay between each increment in rollout.

  • Automatic deployment to staging, manual deployment to production: Sets the STAGING_ENABLED and INCREMENTAL_ROLLOUT_MODE variables to 1 and manual. This means:

    • master branch is directly deployed to staging.
    • Manual actions are provided for incremental rollout to production.

TIP: Tip: Use the blue-green deployment technique to minimize downtime and risk.

Using multiple Kubernetes clusters

When using Auto DevOps, you can deploy different environments to different Kubernetes clusters, due to the 1:1 connection existing between them.

The Deploy Job template used by Auto DevOps currently defines 3 environment names:

  • review/ (every environment starting with review/)
  • staging
  • production

Those environments are tied to jobs using Auto Deploy, so except for the environment scope, they must have a different deployment domain. You must define a separate KUBE_INGRESS_BASE_DOMAIN variable for each of the above based on the environment.

The following table is an example of how to configure the three different clusters:

Cluster name Cluster environment scope KUBE_INGRESS_BASE_DOMAIN variable value Variable environment scope Notes
review review/* review.example.com review/* The review cluster which runs all Review Apps. * is a wildcard, used by every environment name starting with review/.
staging staging staging.example.com staging (Optional) The staging cluster which runs the deployments of the staging environments. You must enable it first.
production production example.com production The production cluster which runs the production environment deployments. You can use incremental rollouts.

To add a different cluster for each environment:

  1. Navigate to your project's Operations > Kubernetes.
  2. Create the Kubernetes clusters with their respective environment scope, as described from the table above.
  3. After creating the clusters, navigate to each cluster and install Ingress. Wait for the Ingress IP address to be assigned.
  4. Make sure you've configured your DNS with the specified Auto DevOps domains.
  5. Navigate to each cluster's page, through Operations > Kubernetes, and add the domain based on its Ingress IP address.

After completing configuration, you can test your setup by creating a merge request and verifying your application is deployed as a Review App in the Kubernetes cluster with the review/* environment scope. Similarly, you can check the other environments.

Limitations

The following restrictions apply.

Private registry support

No documented way of using private container registry with Auto DevOps exists. We strongly advise using GitLab Container Registry with Auto DevOps to simplify configuration and prevent any unforeseen issues.

Install applications behind a proxy

GitLab's Helm integration does not support installing applications when behind a proxy. Users who want to do so must inject their proxy settings into the installation pods at runtime, such as by using a PodPreset:

apiVersion: settings.k8s.io/v1alpha1
kind: PodPreset
metadata:
  name: gitlab-managed-apps-default-proxy
  namespace: gitlab-managed-apps
spec:
  env:
    - name: http_proxy
      value: "PUT_YOUR_HTTP_PROXY_HERE"
    - name: https_proxy
      value: "PUT_YOUR_HTTPS_PROXY_HERE"

Troubleshooting

Unable to select a buildpack

Auto Build and Auto Test may fail to detect your language or framework with the following error:

Step 5/11 : RUN /bin/herokuish buildpack build
 ---> Running in eb468cd46085
    -----> Unable to select a buildpack
The command '/bin/sh -c /bin/herokuish buildpack build' returned a non-zero code: 1

The following are possible reasons:

  • Your application may be missing the key files the buildpack is looking for. Ruby applications require a Gemfile to be properly detected, even though it's possible to write a Ruby app without a Gemfile.
  • No buildpack may exist for your application. Try specifying a custom buildpack.

Pipeline that extends Auto DevOps with only / except fails

If your pipeline fails with the following message:

Found errors in your .gitlab-ci.yml:

  jobs:test config key may not be used with `rules`: only

This error appears when the included jobs rules configuration has been overridden with the only or except syntax. To fix this issue, you must either:

Failure to create a Kubernetes namespace

Auto Deploy will fail if GitLab can't create a Kubernetes namespace and service account for your project. For help debugging this issue, see Troubleshooting failed deployment jobs.

Detected an existing PostgreSQL database

After upgrading to GitLab 13.0, you may encounter this message when deploying with Auto DevOps:

Detected an existing PostgreSQL database installed on the
deprecated channel 1, but the current channel is set to 2. The default
channel changed to 2 in of GitLab 13.0.
[...]

Auto DevOps, by default, installs an in-cluster PostgreSQL database alongside your application. The default installation method changed in GitLab 13.0, and upgrading existing databases requires user involvement. The two installation methods are:

  • channel 1 (deprecated): Pulls in the database as a dependency of the associated Helm chart. Only supports Kubernetes versions up to version 1.15.
  • channel 2 (current): Installs the database as an independent Helm chart. Required for using the in-cluster database feature with Kubernetes versions 1.16 and greater.

If you receive this error, you can do one of the following actions:

  • You can safely ignore the warning and continue using the channel 1 PostgreSQL database by setting AUTO_DEVOPS_POSTGRES_CHANNEL to 1 and redeploying.

  • You can delete the channel 1 PostgreSQL database and install a fresh channel 2 database by setting AUTO_DEVOPS_POSTGRES_DELETE_V1 to a non-empty value and redeploying.

    DANGER: Warning: Deleting the channel 1 PostgreSQL database permanently deletes the existing channel 1 database and all its data. See Upgrading PostgreSQL for more information on backing up and upgrading your database.

  • If you are not using the in-cluster database, you can set POSTGRES_ENABLED to false and re-deploy. This option is especially relevant to users of custom charts without the in-chart PostgreSQL dependency. Database auto-detection is based on the postgresql.enabled Helm value for your release. This value is set based on the POSTGRES_ENABLED CI variable and persisted by Helm, regardless of whether or not your chart uses the variable.

DANGER: Warning: Setting POSTGRES_ENABLED to false permanently deletes any existing channel 1 database for your environment.

Error: unable to recognize "": no matches for kind "Deployment" in version "extensions/v1beta1"

After upgrading your Kubernetes cluster to v1.16+, you may encounter this message when deploying with Auto DevOps:

UPGRADE FAILED
Error: failed decoding reader into objects: unable to recognize "": no matches for kind "Deployment" in version "extensions/v1beta1"

This can occur if your current deployments on the environment namespace were deployed with a deprecated/removed API that doesn't exist in Kubernetes v1.16+. For example, if your in-cluster PostgreSQL was installed in a legacy way, the resource was created via the extensions/v1beta1 API. However, the deployment resource was moved to the app/v1 API in v1.16.

To recover such outdated resources, you must convert the current deployments by mapping legacy APIs to newer APIs. There is a helper tool called mapkubeapis that works for this problem. Follow these steps to use the tool in Auto DevOps:

  1. Modify your .gitlab-ci.yml with:

    include:
      - template: Auto-DevOps.gitlab-ci.yml
      - remote: https://gitlab.com/shinya.maeda/ci-templates/-/raw/master/map-deprecated-api.gitlab-ci.yml
    
    variables:
      HELM_VERSION_FOR_MAPKUBEAPIS: "v2" # If you're using auto-depoy-image v2 or above, please specify "v3".
    
  2. Run the job <environment-name>:map-deprecated-api. Ensure that this job succeeds before moving to the next step. You should see something like the following output:

    2020/10/06 07:20:49 Found deprecated or removed Kubernetes API:
    "apiVersion: extensions/v1beta1
    kind: Deployment"
    Supported API equivalent:
    "apiVersion: apps/v1
    kind: Deployment"
    
  3. Revert your .gitlab-ci.yml to the previous version. You no longer need to include the supplemental template map-deprecated-api.

  4. Continue the deployments as usual.

Error: error initializing: Looks like "https://kubernetes-charts.storage.googleapis.com" is not a valid chart repository or cannot be reached

As announced in the official CNCF blogpost, the stable Helm chart repository will be deprecated and removed on November 13th, 2020. You may encounter this error after that date.

Some GitLab features had dependencies on the stable chart. To mitigate the impact, we changed them to use new official repositories or the Helm Stable Archive repository maintained by GitLab. Auto Deploy contains an example fix.

In Auto Deploy, v1.0.6+ of auto-deploy-image no longer adds the deprecated stable repository to the helm command. If you use a custom chart and it relies on the deprecated stable repository, specify an older auto-deploy-image like this example:

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

.auto-deploy:
  image: "registry.gitlab.com/gitlab-org/cluster-integration/auto-deploy-image:v1.0.5"

Keep in mind that this approach will eventually stop working when the stable repository is removed, so you must eventually fix your custom chart.

To fix your custom chart:

  1. In your chart directory, update the repository value in your requirements.yaml file from :

    repository: "https://kubernetes-charts.storage.googleapis.com/"
    

    to:

    repository: "https://charts.helm.sh/stable"
    
  2. In your chart directory, run helm dep update . using the same Helm major version as Auto DevOps.

  3. Commit the changes for the requirements.yaml file.

  4. If you previously had a requirements.lock file, commit the changes to the file. If you did not previously have a requirements.lock file in your chart, you do not need to commit the new one. This file is optional, but when present, it's used to verify the integrity of the downloaded dependencies.

You can find more information in issue #263778, "Migrate PostgreSQL from stable Helm repo".

Development guides

Development guide for Auto DevOps