Upgrading deployments for newer Auto Deploy dependencies

DETAILS: Tier: Free, Premium, Ultimate Offering: GitLab.com, Self-managed, GitLab Dedicated

Auto Deploy is a feature that deploys your application to a Kubernetes cluster. It consists of several dependencies:

Auto Deploy template is a set of pipeline jobs and scripts that makes use of auto-deploy-image.
auto-deploy-image is the executable image that communicates with the Kubernetes cluster.
auto-deploy-app chart is the Helm chart for deploying your application.

The auto-deploy-image and auto-deploy-app charts use Semantic Versioning. By default, your Auto DevOps project keeps using the stable and non-breaking version. However, these dependencies could be upgraded in a major version release of GitLab with breaking changes requiring you to upgrade your deployments.

This guide explains how to upgrade your deployments with newer or different major versions of Auto Deploy dependencies.

Verify dependency versions

The process to check the current versions differs depending on which template you are using. First verify which template is in use:

For self-managed instances, the stable Auto Deploy template bundled with the GitLab package is being used.
The GitLab.com stable Auto Deploy template is being used if one of the following is true:
- Your Auto DevOps project doesn't have a .gitlab-ci.yml file.
- Your Auto DevOps project has a .gitlab-ci.yml and includes the Auto-DevOps.gitlab-ci.yml template.
The latest Auto Deploy template is being used if both of the following is true:
- Your Auto DevOps project has a .gitlab-ci.yml file and includes the Auto-DevOps.gitlab-ci.yml template.
- It also includes the latest Auto Deploy template

If you know what template is being used:

The auto-deploy-image version is in the template (for example auto-deploy-image:v1.0.3).
The auto-deploy-app chart version is in the auto-deploy-image repository (for example version: 1.0.3).

Compatibility

The following table explains the version compatibility between GitLab and Auto Deploy dependencies:

GitLab version	`auto-deploy-image` version	Notes
v10.0 to v14.0	v0.1.0 to v2.0.0	v0 and v1 auto-deploy-image are backwards compatible.
v13.4 and later	v2.0.0 and later	v2 auto-deploy-image contains breaking changes, as explained in the upgrade guide.

You can find the current stable version of auto-deploy-image in the Auto Deploy stable template.

Upgrade Guide

Projects using Auto DevOps must use the unmodified chart managed by GitLab. Customized charts are unsupported.

Upgrade deployments to the v1 `auto-deploy-image`

The v1 chart is backward compatible with the v0 chart, so no configuration changes are needed.

Upgrade deployments to the v2 `auto-deploy-image`

The v2 auto-deploy-image contains multiple dependency and architectural changes. If your Auto DevOps project has an active environment deployed with the v1 auto-deploy-image, proceed with the following upgrade guide. Otherwise, you can skip this process.

Kubernetes 1.16+

The v2 auto-deploy-image drops support for Kubernetes 1.15 and earlier. If you need to upgrade your Kubernetes cluster, follow your cloud provider's instructions. Here's an example on GKE.

Helm v3

The auto-deploy-image uses the Helm binary to manipulate the releases. Previously, auto-deploy-image used Helm v2, which used Tiller in a cluster. In the v2 auto-deploy-image, it uses Helm v3 that doesn't require Tiller anymore.

If your Auto DevOps project has an active environment that was deployed with the v1 auto-deploy-image, use the following steps to upgrade to v2, which uses Helm v3:

Include the Helm 2to3 migration CI/CD template:
- If you are on GitLab.com, or GitLab 14.0.1 or later, this template is already included in Auto DevOps.
- On other versions of GitLab, you can modify your .gitlab-ci.yml to include the templates:
```
include:
  - template: Auto-DevOps.gitlab-ci.yml
  - remote: https://gitlab.com/gitlab-org/gitlab/-/raw/master/lib/gitlab/ci/templates/Jobs/Helm-2to3.gitlab-ci.yml
```
Set the following CI/CD variables:
- MIGRATE_HELM_2TO3 to true. If this variable is not present, migration jobs do not run.
- AUTO_DEVOPS_FORCE_DEPLOY_V2 to 1.
- Optional: BACKUP_HELM2_RELEASES to 1. If you set this variable, the migration job saves a backup for 1 week in a job artifact called helm-2-release-backups. If you accidentally delete the Helm v2 releases before you are ready, you can restore this backup from a Kubernetes manifest file by using kubectl apply -f $backup.
  
  WARNING: Do not use this if you have public pipelines. This artifact can contain secrets and is visible to any user who can see your job.
Run a pipeline and trigger the <environment-name>:helm-2to3:migrate job.
Deploy your environment as usual. This deployment uses Helm v3.
If the deployment succeeds, you can safely run <environment-name>:helm-2to3:cleanup. This deletes all Helm v2 release data from the namespace.
Remove the MIGRATE_HELM_2TO3 CI/CD variable or set it to false. You can do this one environment at a time using environment scopes.

In-Cluster PostgreSQL Channel 2

The v2 auto-deploy-image drops support for legacy in-cluster PostgreSQL. If your Kubernetes cluster still depends on it, upgrade and migrate your data with the v1 auto-deploy-image.

Traffic routing change for canary deployments and incremental rollouts

Auto Deploy supports advanced deployment strategies such as canary deployments and incremental rollouts.

Previously, auto-deploy-image created one service to balance the traffic between unstable and stable tracks by changing the replica ratio. In the v2 auto-deploy-image, it controls the traffic with Canary Ingress.

For more details, see the v2 auto-deploy-app chart resource architecture.

If your Auto DevOps project has active canary or rollout track releases in the production environment deployed with the v1 auto-deploy-image, use the following steps to upgrade to v2:

Verify your project is using the v1 auto-deploy-image. If not, specify the version.
If you're in the process of deploying canary or rollout deployments, promote them to production first to delete the unstable tracks.
Verify your project is using the v2 auto-deploy-image. If not, specify the version.
Add an AUTO_DEVOPS_FORCE_DEPLOY_V2 CI/CD variable with a value of true in the GitLab CI/CD settings.
Create a new pipeline and run the production job to renew the resource architecture with the v2 auto-deploy-app chart.
Remove the AUTO_DEVOPS_FORCE_DEPLOY_V2 variable.

Use a specific version of Auto Deploy dependencies

To use a specific version of Auto Deploy dependencies, specify the previous Auto Deploy stable template that contains the desired version of auto-deploy-image and auto-deploy-app.

For example, if the template is bundled in GitLab 16.10, change your .gitlab-ci.yml to:

include:
  - template: Auto-DevOps.gitlab-ci.yml
  - remote: https://gitlab.com/gitlab-org/gitlab/-/raw/v16.10.0-ee/lib/gitlab/ci/templates/Jobs/Deploy.gitlab-ci.yml

Ignore warnings and continue deploying

If you are certain that the new chart version is safe to be deployed, you can add the AUTO_DEVOPS_FORCE_DEPLOY_V<major-version-number> CI/CD variable to force the deployment to continue.

For example, if you want to deploy the v2.0.0 chart on a deployment that previously used the v0.17.0 chart, add AUTO_DEVOPS_FORCE_DEPLOY_V2.

Early adopters

If you want to use the latest beta or unstable version of auto-deploy-image, include the latest Auto Deploy template into your .gitlab-ci.yml:

include:
  - template: Auto-DevOps.gitlab-ci.yml
  - template: Jobs/Deploy.latest.gitlab-ci.yml

WARNING: Using a beta or unstable auto-deploy-image could cause unrecoverable damage to your environments. Do not test it with important projects or environments.

Resource Architectures of the `auto-deploy-app` chart

v0 and v1 chart resource architecture

%%{init: { "fontFamily": "GitLab Sans" }}%%
graph TD;
accTitle: v0 and v1 chart resource architecture
accDescr: Shows the relationships between the components of the v0 and v1 charts.

subgraph gl-managed-app
  Z[Nginx Ingress]
  end
  Z[Nginx Ingress] --> A(Ingress);
  Z[Nginx Ingress] --> B(Ingress);
  subgraph stg namespace
  B[Ingress] --> H(...);
end

subgraph prd namespace
  A[Ingress] --> D(Service);
  D[Service] --> E(Deployment:Pods:app:stable);
  D[Service] --> F(Deployment:Pods:app:canary);
  D[Service] --> I(Deployment:Pods:app:rollout);
  E(Deployment:Pods:app:stable)---id1[(Pods:Postgres)]
  F(Deployment:Pods:app:canary)---id1[(Pods:Postgres)]
  I(Deployment:Pods:app:rollout)---id1[(Pods:Postgres)]
end

v2 chart resource architecture

%%{init: { "fontFamily": "GitLab Sans" }}%%
graph TD;
accTitle: v2 chart resource architecture
accDescr: Shows the relationships between the components of the v2 chart.

subgraph gl-managed-app
  Z[Nginx Ingress]
  end
  Z[Nginx Ingress] --> A(Ingress);
  Z[Nginx Ingress] --> B(Ingress);
  Z[Nginx Ingress] --> |If canary is present or incremental rollout/|J(Canary Ingress);
  subgraph stg namespace
  B[Ingress] --> H(...);
end

subgraph prd namespace

  subgraph stable track
    A[Ingress] --> D[Service];
    D[Service] --> E(Deployment:Pods:app:stable);
  end

  subgraph canary track
    J(Canary Ingress) --> K[Service]
    K[Service] --> F(Deployment:Pods:app:canary);
  end

E(Deployment:Pods:app:stable)---id1[(Pods:Postgres)]
F(Deployment:Pods:app:canary)---id1[(Pods:Postgres)]
end

Troubleshooting

Major version mismatch warning

If deploying a chart that has a major version that is different from the previous one, the new chart might not be correctly deployed. This could be due to an architectural change. If that happens, the deployment job fails with a message similar to:

*************************************************************************************
                                   [WARNING]
Detected a major version difference between the chart that is currently deploying (auto-deploy-app-v0.7.0), and the previously deployed chart (auto-deploy-app-v1.0.0).
A new major version might not be backward compatible with the current release (production). The deployment could fail or be stuck in an unrecoverable status.
...

To clear this error message and resume deployments, you must do one of the following:

Manually upgrade the chart version.
Use a specific chart version.

Error: `missing key "app.kubernetes.io/managed-by": must be set to "Helm"`

If your cluster has a deployment that was deployed with the v1 auto-deploy-image, you might encounter the following error:

Error: rendered manifests contain a resource that already exists. Unable to continue with install: Secret "production-postgresql" in namespace "<project-name>-production" exists and cannot be imported into the current release: invalid ownership metadata; label validation error: missing key "app.kubernetes.io/managed-by": must be set to "Helm"; annotation validation error: missing key "meta.helm.sh/release-name": must be set to "production-postgresql"; annotation validation error: missing key "meta.helm.sh/release-namespace": must be set to "<project-name>-production"

This is because the previous deployment was deployed with Helm2, which is not compatible with Helm3. To resolve the problem, follow the upgrade guide.