Updating GitOps ZTP - Clusters at the network far edge | Scalability and performance

Overview of the GitOps ZTP update process
Preparing for the upgrade
Labeling the existing clusters
Stopping the existing GitOps ZTP applications
Required changes to the Git repository
Installing the new GitOps ZTP applications
Rolling out the GitOps ZTP configuration changes

You can update the Gitops zero touch provisioning (ZTP) infrastructure independently from the hub cluster, Red Hat Advanced Cluster Management (RHACM), and the managed OKD clusters.

You can update the Red Hat OpenShift GitOps Operator when new versions become available. When updating the GitOps ZTP plugin, review the updated files in the reference configuration and ensure that the changes meet your requirements.

Overview of the GitOps ZTP update process

You can update GitOps zero touch provisioning (ZTP) for a fully operational hub cluster running an earlier version of the GitOps ZTP infrastructure. The update process avoids impact on managed clusters.

Any changes to policy settings, including adding recommended content, results in updated polices that must be rolled out to the managed clusters and reconciled.

At a high level, the strategy for updating the GitOps ZTP infrastructure is as follows:

Label all existing clusters with the ztp-done label.
Stop the ArgoCD applications.
Install the new GitOps ZTP tools.
Update required content and optional changes in the Git repository.
Update and restart the application configuration.

Preparing for the upgrade

Use the following procedure to prepare your site for the GitOps zero touch provisioning (ZTP) upgrade.

Procedure

Get the latest version of the GitOps ZTP container that has the custom resources (CRs) used to configure Red Hat OpenShift GitOps for use with GitOps ZTP.
Extract the argocd/deployment directory by using the following commands:
```
$ mkdir -p ./update
```
```
$ podman run --log-driver=none --rm registry.redhat.io/openshift4/ztp-site-generate-rhel8:v4.11 extract /home/ztp --tar | tar x -C ./update
```
The /update directory contains the following subdirectories:
- update/extra-manifest: contains the source CR files that the SiteConfig CR uses to generate the extra manifest configmap.
- update/source-crs: contains the source CR files that the PolicyGenTemplate CR uses to generate the Red Hat Advanced Cluster Management (RHACM) policies.
- update/argocd/deployment: contains patches and YAML files to apply on the hub cluster for use in the next step of this procedure.
- update/argocd/example: contains example SiteConfig and PolicyGenTemplate files that represent the recommended configuration.
Update the clusters-app.yaml and policies-app.yaml files to reflect the name of your applications and the URL, branch, and path for your Git repository.

If the upgrade includes changes that results in obsolete policies, the obsolete policies should be removed prior to performing the upgrade.

Diff the changes between the configuration and deployment source CRs in the /update folder and Git repo where you manage your fleet site CRs. Apply and push the required changes to your site repository.

When you update GitOps ZTP to the latest version, you must apply the changes from the update/argocd/deployment directory to your site repository. Do not use older versions of the argocd/deployment/ files.

Labeling the existing clusters

To ensure that existing clusters remain untouched by the tool updates, label all existing managed clusters with the ztp-done label.

This procedure only applies when updating clusters that were not provisioned with Topology Aware Lifecycle Manager (TALM). Clusters that you provision with TALM are automatically labeled with ztp-done.

Procedure

Find a label selector that lists the managed clusters that were deployed with zero touch provisioning (ZTP), such as local-cluster!=true:
```
$ oc get managedcluster -l 'local-cluster!=true'
```
Ensure that the resulting list contains all the managed clusters that were deployed with ZTP, and then use that selector to add the ztp-done label:
```
$ oc label managedcluster -l 'local-cluster!=true' ztp-done=
```

Stopping the existing GitOps ZTP applications

Removing the existing applications ensures that any changes to existing content in the Git repository are not rolled out until the new version of the tools is available.

Use the application files from the deployment directory. If you used custom names for the applications, update the names in these files first.

Procedure

Perform a non-cascaded delete on the clusters application to leave all generated resources in place:
```
$ oc delete -f update/argocd/deployment/clusters-app.yaml
```

Perform a cascaded delete on the policies application to remove all previous policies:

$ oc patch -f policies-app.yaml -p '{"metadata": {"finalizers": ["resources-finalizer.argocd.argoproj.io"]}}' --type merge

$ oc delete -f update/argocd/deployment/policies-app.yaml

Required changes to the Git repository

When upgrading the ztp-site-generate container from an earlier release of GitOps ZTP to v4.10 or later, there are additional requirements for the contents of the Git repository. Existing content in the repository must be updated to reflect these changes.

Make required changes to PolicyGenTemplate files:

All PolicyGenTemplate files must be created in a Namespace prefixed with ztp. This ensures that the GitOps zero touch provisioning (ZTP) application is able to manage the policy CRs generated by GitOps ZTP without conflicting with the way Red Hat Advanced Cluster Management (RHACM) manages the policies internally.

Add the kustomization.yaml file to the repository:

All SiteConfig and PolicyGenTemplate CRs must be included in a kustomization.yaml file under their respective directory trees. For example:

├── policygentemplates
│   ├── site1-ns.yaml
│   ├── site1.yaml
│   ├── site2-ns.yaml
│   ├── site2.yaml
│   ├── common-ns.yaml
│   ├── common-ranGen.yaml
│   ├── group-du-sno-ranGen-ns.yaml
│   ├── group-du-sno-ranGen.yaml
│   └── kustomization.yaml
└── siteconfig
    ├── site1.yaml
    ├── site2.yaml
    └── kustomization.yaml

The files listed in the generator sections must contain either SiteConfig or PolicyGenTemplate CRs only. If your existing YAML files contain other CRs, for example, Namespace, these other CRs must be pulled out into separate files and listed in the resources section.

The PolicyGenTemplate kustomization file must contain all PolicyGenTemplate YAML files in the generator section and Namespace CRs in the resources section. For example:

apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization

generators:
- common-ranGen.yaml
- group-du-sno-ranGen.yaml
- site1.yaml
- site2.yaml

resources:
- common-ns.yaml
- group-du-sno-ranGen-ns.yaml
- site1-ns.yaml
- site2-ns.yaml

The SiteConfig kustomization file must contain all SiteConfig YAML files in the generator section and any other CRs in the resources:

apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization

generators:
- site1.yaml
- site2.yaml

Remove the pre-sync.yaml and post-sync.yaml files.

In OKD 4.10 and later, the pre-sync.yaml and post-sync.yaml files are no longer required. The update/deployment/kustomization.yaml CR manages the policies deployment on the hub cluster.

There is a set of pre-sync.yaml and post-sync.yaml files under both the SiteConfig and PolicyGenTemplate trees.
Review and incorporate recommended changes

Each release may include additional recommended changes to the configuration applied to deployed clusters. Typically these changes result in lower CPU use by the OpenShift platform, additional features, or improved tuning of the platform.

Review the reference SiteConfig and PolicyGenTemplate CRs applicable to the types of cluster in your network. These examples can be found in the argocd/example directory extracted from the GitOps ZTP container.

Installing the new GitOps ZTP applications

Using the extracted argocd/deployment directory, and after ensuring that the applications point to your site Git repository, apply the full contents of the deployment directory. Applying the full contents of the directory ensures that all necessary resources for the applications are correctly configured.

Procedure

To patch the ArgoCD instance in the hub cluster by using the patch file that you previously extracted into the update/argocd/deployment/ directory, enter the following command:
```
$ oc patch argocd openshift-gitops \
-n openshift-gitops --type=merge \
--patch-file update/argocd/deployment/argocd-openshift-gitops-patch.json
```
To apply the contents of the argocd/deployment directory, enter the following command:
```
$ oc apply -k update/argocd/deployment
```

Rolling out the GitOps ZTP configuration changes

If any configuration changes were included in the upgrade due to implementing recommended changes, the upgrade process results in a set of policy CRs on the hub cluster in the Non-Compliant state. With the ZTP GitOps v4.10 and later ztp-site-generate container, these policies are set to inform mode and are not pushed to the managed clusters without an additional step by the user. This ensures that potentially disruptive changes to the clusters can be managed in terms of when the changes are made, for example, during a maintenance window, and how many clusters are updated concurrently.

To roll out the changes, create one or more ClusterGroupUpgrade CRs as detailed in the TALM documentation. The CR must contain the list of Non-Compliant policies that you want to push out to the managed clusters as well as a list or selector of which clusters should be included in the update.

Additional resources

For information about the Topology Aware Lifecycle Manager (TALM), see About the Topology Aware Lifecycle Manager configuration.
For information about creating ClusterGroupUpgrade CRs, see About the auto-created ClusterGroupUpgrade CR for ZTP.