Skip to content

Latest commit

 

History

History
69 lines (52 loc) · 3.44 KB

migration.md

File metadata and controls

69 lines (52 loc) · 3.44 KB
Raw

Migrating ClusterInstallation to Mattermost

As of the new release, the Custom Resource managed by the Mattermost Operator changes from ClusterInstallation to Mattermost. Besides the name change, some new functionality is introduced while other functionality is changed or removed.

BlueGreen and Canary deployments were not widely used and introduced a lot of complexity, so those features were dropped. In most cases, the multi-replica Mattermost cluster proved to be enough for save updates between versions. Behavior similar to BlueGreen could be mimicked by using multiple Mattermost resources and with some extra ingresses and services.

Automatic migration

It is possible for the Operator to migrate ClusterInstallation to Mattermost as long as none of the unsupported features are enabled (like BlueGreen or Canary). During the migration, old Pods are deleted only after the new Mattermost resource reaches the stable state, therefore the Mattermost instance should not experience any downtime.

For migration to be possible, the Mattermost Operator needs to be version v1.12.x.

NOTE: Make sure that Mattermost Operator is in version v1.12.x before starting the migration. To do that, inspect the Operator image tag. To get full image name and tag, run:

kubectl -n [OPERATOR_NAMESPACE] get deployment mattermost-operator -o jsonpath='{.spec.template.spec.containers[0].image}'

If the Operator's minor version is earlier than 12, update it to v1.12.0 by updating manifests to the released version.

To run the migration, follow the steps:

  1. Record the namespace and name of the ClusterInstallation and export those as environment variables on the device from which you are triggering the migration:

    export CI_NAMESPACE=[YOUR_NAMESPACE]
export CI_NAME=[NAME_OF_YOUR_INSTALLATION]

  2. Trigger the migration by setting spec.migrate to true on the ClusterInstsallation instance:

    kubectl -n ${CI_NAMESPACE} patch clusterinstallation ${CI_NAME} --type merge --patch "spec:
  migrate: true"

  3. Wait for migration to finish. The migration is done when the ClusterInstallation is removed and the Mattermost CR with the same name is created and it's status.state is equal to stable.

  4. Ensure that Mattermost CR spec is correct and matches your needs:

    kubectl -n ${CI_NAMESPACE} get mm ${CI_NAME} -oyaml

Troubleshooting the migration

To see the status and potential errors that occurred during the migration, query status.migration from the ClusterInstallation:

kubectl -n ${CI_NAMESPACE} get clusterinstallation ${CI_NAME} -o jsonpath='{.status.migration}'

If the migration failed, it can be reverted with several steps, which may vary depending on when the failure occurred.

CAUTION: If the migration finished successfully it cannot be reverted.

  1. Set spec.migrate to false or remove the field from the resource.

    kubectl -n ${CI_NAMESPACE} patch clusterinstallation ${CI_NAME} --type merge --patch "spec:
  migrate: false"

  2. Remove the new Mattermost resource if it was created.

    kubectl -n ${CI_NAMESPACE} delete mm ${CI_NAME}

  3. Remove new Deployment if it was created.

    kubectl -n ${CI_NAMESPACE} delete deployment ${CI_NAME}