Install Ceph Storage with Rook

Refactor In Progress To Use The Lab CLI - The Original Instructions are Below

  1. Mirror Ceph Images

    labcli --ceph -m
    
  2. Install Rook Operator with Ceph cluster:

    labcli --ceph -i
    
  3. Configure the internal image registry for a Ceph PVC

    labcli --ceph -r
    

Original Instructions

Ideally, you need a full cluster for this deployment; 3 master and 3 worker nodes. Additionally, you need to give the worker nodes a second disk that will be used by Ceph. If you followed the guide for adding working nodes here; Add Worker Nodes, then you are good to proceed. If not, you will need to recreate your cluster and add a second drive to the three dual role, master/worker nodes.

If you need to recreate your cluster to add disks for Ceph, follow these instructions: Destroy and Recreate your Cluster.

Follow these steps to deploy a Ceph cluster:

Initial setup for Ceph install

The first thing that we are going to do is mirror the Rook Operator and Ceph Storage images into Nexus. Remember, our cluster is isolated from the internet. If this were a “real” data center, we would run security scans on the images before releasing them to be ingested by applications in our cluster. In a future post, we will actually do that. We’ll set up a two stage process to validate that images are safe to run in our data center. But, more on that later…

  1. Make sure that you have podman installed on your workstation.

    Mac OS:

    brew install podman
    podman machine init
    podman machine start
    
  2. Pull the Rook and Ceph images so that we can mirror them to Nexus:

    podman pull  quay.io/cephcsi/cephcsi:v3.5.1
    podman pull  k8s.gcr.io/sig-storage/csi-node-driver-registrar:v2.5.0
    podman pull  k8s.gcr.io/sig-storage/csi-resizer:v1.4.0
    podman pull  k8s.gcr.io/sig-storage/csi-provisioner:v3.1.0
    podman pull  k8s.gcr.io/sig-storage/csi-snapshotter:v5.0.1
    podman pull  k8s.gcr.io/sig-storage/csi-attacher:v3.4.0
    podman pull  docker.io/rook/ceph:v1.8.7
    podman pull  quay.io/ceph/ceph:v16.2.7
    
  3. Select the Lab subdomain that you want to work with:

    labctx
    
  4. Tag the images for Nexus:

    LOCAL_REGISTRY=$(yq e ".local-registry" ${OKD_LAB_PATH}/lab-config/${SUB_DOMAIN}-cluster.yaml)
    
    podman tag quay.io/cephcsi/cephcsi:v3.5.1 ${LOCAL_REGISTRY}/cephcsi/cephcsi:v3.5.1
    podman tag k8s.gcr.io/sig-storage/csi-node-driver-registrar:v2.5.0 ${LOCAL_REGISTRY}/sig-storage/csi-node-driver-registrar:v2.5.0
    podman tag k8s.gcr.io/sig-storage/csi-resizer:v1.4.0 ${LOCAL_REGISTRY}/sig-storage/csi-resizer:v1.4.0
    podman tag k8s.gcr.io/sig-storage/csi-provisioner:v3.1.0 ${LOCAL_REGISTRY}/sig-storage/csi-provisioner:v3.1.0
    podman tag k8s.gcr.io/sig-storage/csi-snapshotter:v5.0.1 ${LOCAL_REGISTRY}/sig-storage/csi-snapshotter:v5.0.1
    podman tag k8s.gcr.io/sig-storage/csi-attacher:v3.4.0 ${LOCAL_REGISTRY}/sig-storage/csi-attacher:v3.4.0
    podman tag docker.io/rook/ceph:v1.8.7 ${LOCAL_REGISTRY}/rook/ceph:v1.8.7
    podman tag quay.io/ceph/ceph:v16.2.7 ${LOCAL_REGISTRY}/ceph/ceph:v16.2.7
    
  5. Log into Nexus:

    podman login -u openshift-mirror ${LOCAL_REGISTRY}
    
  6. Push the images: (The --tls-verify is for Mac users. It’s not ideal, but it works.)

    podman push ${LOCAL_REGISTRY}/cephcsi/cephcsi:v3.5.1 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/sig-storage/csi-node-driver-registrar:v2.5.0 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/sig-storage/csi-resizer:v1.4.0 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/sig-storage/csi-provisioner:v3.1.0 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/sig-storage/csi-snapshotter:v5.0.1 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/sig-storage/csi-attacher:v3.4.0 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/rook/ceph:v1.8.7 --tls-verify=false
    podman push ${LOCAL_REGISTRY}/ceph/ceph:v16.2.7 --tls-verify=false
    
  7. Clean up after yourself and log off the bastion host:

    Remove the images so that we don’t fill up tmpfs

    podman image rm -a
    
  8. If you are on a Mac like me, the stop the podman VM:

    podman machine stop
    

Now we are ready to install Ceph

  1. If you haven’t pulled a recent copy of my companion project to this blog. Do that now. We need the Ceph manifests:

    The companion project is located at: https://github.com/cgruver/okd-home-lab. You cloned it during the cluster setup, so it should still be located at: ${OKD_LAB_PATH}/okd-home-lab

    cd ${OKD_LAB_PATH}/okd-home-lab
    git fetch
    git pull
    

    The Ceph installation files, located in ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install, are modified from the project at https://github.com/rook/rook.

  2. Next, we need to label our nodes so that Ceph knows where to deploy. If you look at the cluster.yml file, you will see the nodeAffinity configuration.

    If you created worker nodes by following this guide, Add Worker Nodes, then set the following variable:

    export CEPH_NODE=worker
    

    If you are using a three node cluster as mentioned above, then set this:

    export CEPH_NODE=master
    

    Add a label to the nodes that are going to host Ceph:

    oc login -u admin https://api.okd4.${SUB_DOMAIN}.${LAB_DOMAIN}:6443
       
    for i in 0 1 2
    do
          oc label nodes okd4-${CEPH_NODE}-${i}.${SUB_DOMAIN}.${LAB_DOMAIN} role=storage-node
    done
    
  3. Create the Ceph Namspace, CRDs, and Roles:

    oc apply -f ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install/crds.yaml
    oc apply -f ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install/common.yaml
    oc apply -f ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install/rbac.yaml
    
  4. Deploy the Ceph Operator:

    export REGION=${SUB_DOMAIN}
    export PROXY_REGISTRY=$(yq e ".proxy-registry" ${OKD_LAB_PATH}/lab-config/${SUB_DOMAIN}-cluster.yaml)
    envsubst < ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install/operator-openshift.yaml | oc apply -f -
    

    Wait for the Operator pod to completely deploy before executing the next step.

  5. Deploy the Ceph cluster:

    envsubst < ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/install/cluster.yaml | oc apply -f -
    

    This will take a while to complete.

    It is finished when you see all of the rook-ceph-osd-prepare-okd4-... pods in a Completed state.

    oc get pods -n rook-ceph | grep rook-ceph-osd-prepare
    
  6. If you have worker nodes and designated the control plane as Infra nodes:

    If you followed this guide here: Add Worker Nodes

    Then we also need to allow pods that run on the control plane nodes to access Ceph volumes.

    Add the following patch to your Ceph cluster:

    oc patch configmap rook-ceph-operator-config -n rook-ceph --type merge --patch '"data": {"CSI_PLUGIN_TOLERATIONS": "- key: \"node-role.kubernetes.io/master\"\n  operator: \"Exists\"\n  effect: \"NoSchedule\"\n"}'
    

Now, let’s create a PVC for the Image Registry.

  1. First, we need a Storage Class:

    oc apply -f ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/configure/ceph-storage-class.yml
    
  2. Now create the PVC:

    oc apply -f ${OKD_LAB_PATH}/okd-home-lab/rook-ceph/configure/registry-pvc.yml
    
  3. Now, patch the imageregistry operator to use the PVC that you just created:

    If you previously added emptyDir as a storage type to the Registry, you need to remove it first:

    oc patch configs.imageregistry.operator.openshift.io cluster --type json -p '[{ "op": "remove", "path": "/spec/storage/emptyDir" }]'
    

    Now patch it to use the new PVC:

    oc patch configs.imageregistry.operator.openshift.io cluster --type merge --patch '{"spec":{"rolloutStrategy":"Recreate","managementState":"Managed","storage":{"pvc":{"claim":"registry-pvc"}}}}'
    
  4. Make sure that the PVC gets bound to a new PV:

    oc get pvc -n openshift-image-registry
    

    You should see output similar to:

    NAME           STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS      AGE
    registry-pvc   Bound    pvc-bcee4ccd-aa6e-4c8c-89b0-3f8da1c17df0   100Gi      RWO            rook-ceph-block   4d17h
    
  5. If you want to designate your new storage class as the default storage class, the do the following:

    oc patch storageclass rook-ceph-block -p '{"metadata": {"annotations":{"storageclass.kubernetes.io/is-default-class":"true"}}}'
    

You just created a Ceph cluster and bound your image registry to a Persistent Volume!