Whether your OpenShift cluster(s) are hosted on-premise or in the cloud downtime happens. This could be a temporary outage or it can be an extended outage with no resolution in sight. This article will explain how GitOps can be used for the rapid redeployment of your Kubernetes objects. One important thing to note is that GitOps can only restore Kubernetes objects so that means any persistent data required for an application to correctly function must be restored for stateful applications, such as databases, to be back in service.
Continuing with our usage of Argo CD we will discuss two different ways to start the process in restoring these objects. For both of these procedures we will assume that a new cluster has been deployed and that Argo CD has also been deployed. We also assume that the same OpenShift routes and DNS zones will be used because the OpenShift routes should be stored within git as well.
Using the Argo CD binary you will manually need to define the repositories and Argo CD applications. This process will require you to have a list of repositories and command to define them within Argo CD. For example, we could run the following to restore our simple-app project.
argocd repo add https://github.com/cooktheryan/blogpost
argocd app create --project default \
--name simple-app --repo https://github.com/cooktheryan/blogpost.git \
--path . --dest-server https://kubernetes.default.svc \
-dest-namespace simple-app --revision master
This process works as long as you have a list of all repositories, git branches, and namespaces documented. Once these items are all defined and loaded into Argo CD, the objects will begin to deploy within the cluster and sync with Argo CD.
With some planning we can make this process better though by using git to manage our GitOps resources. Storing a copy of the configmap and the various ArgoCD applications within Git or even something simple as a file or object share that exists outside of the data center hosting the OpenShift cluster will allow us to rapidly redefine the objects managed by Argo CD.
First, let's take a look at the configmap in YAML format. We will see the repositories currently defined within Argo CD.
oc get configmap -n argocd argocd-cm -o yaml
- url: https://github.com/cooktheryan/blogpost
- url: http://github.com/openshift/federation-dev.git
We will next save the configmap in YAML format.
oc get configmap -n argocd argocd-cm -o yaml --export > argocd-cm.yaml
But what if my repository requires a ssh key? If that is the case then we will need to export
the secret as well. If your repositories do not require a ssh key or authentication then ignore this step.
The configmap identifies the name of the secret that is used by the repository.
oc get secrets -n argocd repo-federation-dev-3296805493 -o yaml > repo-federation-dev-secret.yaml
We will now need to backup any Argo CD applications. This can be done per individual application or by just exporting
all of the applications to a YAML file. For this example since we only have one application within Argo CD.
It would be suggested to store the applications individually and within the same git repository that the Kubernetes
objects are defined so that they will be under revision control and available in the event of a disaster.
oc get applications -o yaml --export > simple-app-backup.yaml
Now that we have all of required Argo CD objects we will now import them into our new server that has been deployed when
the new environment was brought online.
First, we will update the configmap to include our previously defined repositories.
oc apply -f argocd-repos.yaml -n argocd-cm.yaml
OPTIONAL: If credentials were used for any of the repositories then the credentials in the secret must be imported before running a repo list.
oc apply -f repo-federation-dev-secret.yaml -n argocd
Next, we will restore our Argo CD applications which will cause the our Kubernetes objects like namespaces, services, deployments, and routes to deploy
onto the cluster.
oc create -f backup.yaml -n argocd
At this point all of the objects should begin to deploy and the applications within Argo CD should post a healthy state. As with
all backup solutions it makes sense to test this DR procedure frequently. This could be done per application basis on another cluster or with
Code Ready Containers.
In the coming weeks we will publish another disaster recovery post containing information on what to do if your cluster fails and how
a Global Load Balancer can keep the lights on.