Migrate from GitLab Managed Apps to Cluster Management Projects (DEPRECATED) (FREE)

The GitLab Managed Apps were deprecated in GitLab 14.0 in favor of user-controlled Cluster Management projects. Managing your cluster applications through a project enables you a lot more flexibility to manage your cluster than through the late GitLab Managed Apps. To migrate to the cluster management project you need GitLab Runners available and be familiar with Helm.

Migrate to a Cluster Management Project

To migrate from GitLab Managed Apps to a Cluster Management Project, follow the steps below. See also video walk-throughs with examples.

  1. Create a new project based on the Cluster Management Project template.

  2. Install an agent for this project in your cluster.

  3. Set the KUBE_CONTEXT CI/CD variable to the newly installed agent's context, as instructed in the .gitlab-ci.yml from the Project Template.

  4. Detect apps deployed through Helm v2 releases by using the pre-configured .gitlab-ci.yml file:

    • In case you had overwritten the default GitLab Managed Apps namespace, edit .gitlab-ci.yml, and make sure the script is receiving the correct namespace as an argument:

      script:
        - gl-fail-if-helm2-releases-exist <your_custom_namespace>
    • If you kept the default name (gitlab-managed-apps), then the script is already set up.

    Either way, run a pipeline manually and read the logs of the detect-helm2-releases job to know if you have any Helm v2 releases and which are they.

  5. If you have no Helm v2 releases, skip this step. Otherwise, follow the official Helm docs on how to migrate from Helm v2 to Helm v3, and clean up the Helm v2 releases after you are confident that they have been successfully migrated.

  6. In this step you should already have only Helm v3 releases. Uncomment from the main ./helmfile.yaml the paths for the applications that you would like to manage with this project. Although you could uncomment all the ones you want to managed at once, we recommend you repeat the following steps separately for each app, so you do not get lost during the process.

  7. Edit the associated applications/{app}/helmfiles.yaml to match the chart version currently deployed for your app. Take a GitLab Runner Helm v3 release as an example:

    The following command lists the releases and their versions:

    helm ls -n gitlab-managed-apps
    
    NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
    runner gitlab-managed-apps 1 2021-06-09 19:36:55.739141644 +0000 UTC deployed gitlab-runner-0.28.0 13.11.0

    Take the version from the CHART column which is in the format {release}-v{chart_version}, then edit the version: attribute in the ./applications/gitlab-runner/helmfile.yaml, so that it matches the version you have currently deployed. This is a safe step to avoid upgrading versions during this migration. Make sure you replace gitlab-managed-apps from the above command if you have your apps deployed to a different namespace.

  8. Edit the applications/{app}/values.yaml associated with your app to match the currently deployed values. For example, for GitLab Runner:

    1. Copy the output of the following command (it might be big):

      helm get values runner -n gitlab-managed-apps -a --output yaml
    2. Overwrite applications/gitlab-runner/values.yaml with the output of the previous command.

    This safe step will guarantee that no unexpected default values overwrite your currently deployed values. For instance, your GitLab Runner could have its gitlabUrl or runnerRegistrationToken overwritten by mistake.

  9. Some apps require special attention:

    • Ingress: Due to an existing chart issue, you might see spec.clusterIP: Invalid value when trying to run the ./gl-helmfile command. To work around this, after overwriting the release values in applications/ingress/values.yaml, you might need to overwrite all the occurrences of omitClusterIP: false, setting it to omitClusterIP: true. Another approach,could be to collect these IPs by running kubectl get services -n gitlab-managed-apps and then overwriting each ClusterIP that it complains about with the value you got from that command.

    • Vault: This application introduces a breaking change from the chart we used in Helm v2 to the chart used in Helm v3. So, the only way to integrate it with this Cluster Management Project is to actually uninstall this app and accept the chart version proposed in applications/vault/values.yaml.

    • Cert-manager:

      • For users on Kubernetes version 1.20 or above, the deprecated cert-manager v0.10 is no longer valid and the upgrade includes a breaking change. So we suggest that you backup and uninstall cert-manager v0.10, and install the latest cert-manager instead. To install this version, uncomment applications/cert-manager/helmfile.yaml from ./helmfile.yaml. This triggers a pipeline to install the new version.

      • For users on Kubernetes versions lower than 1.20, you can stick to v0.10 by uncommenting applications/cert-manager-legacy/helmfile.yaml in your project's main Helmfile (./helmfile.yaml).

        WARNING: Cert-manager v0.10 breaks when Kubernetes is upgraded to version 1.20 or later.

  10. After following all the previous steps, run a pipeline manually and watch the apply job logs to see if any of your applications were successfully detected, installed, and whether they got any unexpected updates.

    Some annotation checksums are expected to be updated, as well as this attribute:

    --- heritage: Tiller
    +++ heritage: Tiller

After getting a successful pipeline, repeat these steps for any other deployed apps you want to manage with the Cluster Management Project.

Backup and uninstall cert-manager v0.10

  1. Follow the official docs on how to backup your cert-manager v0.10 data.
  2. Uninstall cert-manager by editing the setting all the occurrences of installed: true to installed: false in the applications/cert-manager/helmfile.yaml file.
  3. Search for any left-over resources by executing the following command kubectl get Issuers,ClusterIssuers,Certificates,CertificateRequests,Orders,Challenges,Secrets,ConfigMaps -n gitlab-managed-apps | grep certmanager.
  4. For each of the resources found in the previous step, delete them with kubectl delete -n gitlab-managed-apps {ResourceType} {ResourceName}. For example, if you found a resource of type ConfigMap named cert-manager-controller, delete it by executing: kubectl delete configmap -n gitlab-managed-apps cert-manager-controller.

Video walk-throughs

You can watch these videos with examples on how to migrate from GMA to a Cluster Management project: