Manually creating IAM for GCP

In environments where the cloud identity and access management (IAM) APIs are not reachable, or the administrator prefers not to store an administrator-level credential secret in the cluster kube-system namespace, you can put the Cloud Credential Operator (CCO) into manual mode before you install the cluster.

Alternatives to storing administrator-level secrets in the kube-system project

The Cloud Credential Operator (CCO) manages cloud provider credentials as Kubernetes custom resource definitions (CRDs). You can configure the CCO to suit the security requirements of your organization by setting different values for the credentialsMode parameter in the install-config.yaml file.

If you prefer not to store an administrator-level credential secret in the cluster kube-system project, you can set the credentialsMode parameter for the CCO to Manual when installing OKD and manage your cloud credentials manually.

Using manual mode allows each cluster component to have only the permissions it requires, without storing an administrator-level credential in the cluster. You can also use this mode if your environment does not have connectivity to the cloud provider public IAM endpoint. However, you must manually reconcile permissions with new release images for every upgrade. You must also manually supply credentials for every component that requests them.

Additional resources

To learn how to rotate or remove the administrator-level credential secret after installing OKD, see Rotating or removing cloud provider credentials.

For a detailed description of all available CCO credential modes and their supported platforms, see the Cloud Credential Operator reference.

Manually create IAM

The Cloud Credential Operator (CCO) can be put into manual mode prior to installation in environments where the cloud identity and access management (IAM) APIs are not reachable, or the administrator prefers not to store an administrator-level credential secret in the cluster kube-system namespace.

Procedure

  1. To generate the manifests, run the following command from the directory that contains the installation program:

    1. $ openshift-install create manifests --dir=<installation_directory> (1)
    1For <installation_directory>, specify the directory name to store the files that the installation program creates.
  2. Insert a config map into the manifests directory so that the Cloud Credential Operator is placed in manual mode:

    1. $ cat <<EOF > mycluster/manifests/cco-configmap.yaml
    2. apiVersion: v1
    3. kind: ConfigMap
    4. metadata:
    5. name: cloud-credential-operator-config
    6. namespace: openshift-cloud-credential-operator
    7. annotations:
    8. release.openshift.io/create-only: "true"
    9. data:
    10. disabled: "true"
    11. EOF
  3. Remove the admin credential secret created using your local cloud credentials. This removal prevents your admin credential from being stored in the cluster:

    1. $ rm mycluster/openshift/99_cloud-creds-secret.yaml
  4. From the directory that contains the installation program, obtain details of the OKD release image that your openshift-install binary is built to use:

    1. $ openshift-install version

    Example output

    1. release image quay.io/openshift-release-dev/ocp-release:4.y.z-x86_64
  5. Locate all CredentialsRequest objects in this release image that target the cloud you are deploying on:

    1. $ oc adm release extract quay.io/openshift-release-dev/ocp-release:4.y.z-x86_64 --credentials-requests --cloud=gcp

    This displays the details for each request.

    Sample CredentialsRequest object

    1. apiVersion: cloudcredential.openshift.io/v1
    2. kind: CredentialsRequest
    3. metadata:
    4. labels:
    5. controller-tools.k8s.io: "1.0"
    6. name: openshift-image-registry-gcs
    7. namespace: openshift-cloud-credential-operator
    8. spec:
    9. secretRef:
    10. name: installer-cloud-credentials
    11. namespace: openshift-image-registry
    12. providerSpec:
    13. apiVersion: cloudcredential.openshift.io/v1
    14. kind: GCPProviderSpec
    15. predefinedRoles:
    16. - roles/storage.admin
    17. - roles/iam.serviceAccountUser
    18. skipServiceCheck: true
  6. Create YAML files for secrets in the openshift-install manifests directory that you generated previously. The secrets must be stored using the namespace and secret name defined in the spec.secretRef for each credentialsRequest. The format for the secret data varies for each cloud provider.

  7. From the directory that contains the installation program, proceed with your cluster creation:

    1. $ openshift-install create cluster --dir=<installation_directory>

    Before upgrading a cluster that uses manually maintained credentials, you must ensure that the CCO is in an upgradeable state. For details, see the Upgrading clusters with manually maintained credentials section of the installation content for your cloud provider.

Admin credentials root secret format

Each cloud provider uses a credentials root secret in the kube-system namespace by convention, which is then used to satisfy all credentials requests and create their respective secrets. This is done either by minting new credentials, with mint mode, or by copying the credentials root secret, with passthrough mode.

The format for the secret varies by cloud, and is also used for each CredentialsRequest secret.

Google Cloud Platform (GCP) secret format

  1. apiVersion: v1
  2. kind: Secret
  3. metadata:
  4. namespace: kube-system
  5. name: gcp-credentials
  6. stringData:
  7. service_account.json: <ServiceAccount>

Upgrading clusters with manually maintained credentials

If credentials are added in a future release, the Cloud Credential Operator (CCO) upgradable status for a cluster with manually maintained credentials changes to false. For minor release, for example, from 4.5 to 4.6, this status prevents you from upgrading until you have addressed any updated permissions. For z-stream releases, for example, from 4.5.10 to 4.5.11, the upgrade is not blocked, but the credentials must still be updated for the new release.

Use the Administrator perspective of the web console to determine if the CCO is upgradeable.

  1. Navigate to AdministrationCluster Settings.

  2. To view the CCO status details, click cloud-credential in the Cluster Operators list.

  3. If the Upgradeable status in the Conditions section is False, examine the credentialsRequests for the new release and update the manually maintained credentials on your cluster to match before upgrading.

In addition to creating new credentials for the release image that you are upgrading to, you must review the required permissions for existing credentials and accommodate any new permissions requirements for existing components in the new release. The CCO cannot detect these mismatches and will not set upgradable to false in this case.

The Manually creating IAM section of the installation content for your cloud provider explains how to obtain and use the credentials required for your cloud.

Mint mode

Mint mode is the default and recommended Cloud Credential Operator (CCO) credentials mode for OKD. In this mode, the CCO uses the provided administrator-level cloud credential to run the cluster. Mint mode is supported for AWS, GCP, and Azure.

In mint mode, the admin credential is stored in the kube-system namespace and then used by the CCO to process the CredentialsRequest objects in the cluster and create users for each with specific permissions.

The benefits of mint mode include:

  • Each cluster component has only the permissions it requires

  • Automatic, on-going reconciliation for cloud credentials, including additional credentials or permissions that might be required for upgrades

One drawback is that mint mode requires admin credential storage in a cluster kube-system secret.

Mint Mode with removal or rotation of the admin credential

Currently, this mode is only supported on AWS.

In this mode, a user installs OKD with an admin credential just like the normal mint mode. However, this mode removes the admin credential secret from the cluster post-installation.

The administrator can have the Cloud Credential Operator make its own request for a read-only credential that allows it to verify if all CredentialsRequest objects have their required permissions, thus the admin credential is not required unless something needs to be changed. After the associated credential is removed, it can be destroyed on the underlying cloud, if desired.

Prior to upgrade, the admin credential should be restored. In the future, upgrade might be blocked if the credential is not present.

The admin credential is not stored in the cluster permanently.

This mode still requires the admin credential in the cluster for brief periods of time. It also requires manually re-instating the secret with admin credentials for each upgrade.

Next steps