This is a subordinate charm of kubernetes-control-plane, deployed as part of Charmed Kubernetes and enables a Kubernetes cluster to use Ceph as a storage backend. Currently supported storage options are ceph-xfs, ceph-ext4, and cephfs.

Note: This charm creates various Kubernetes resources, including pods. Therefore it requires kubernetes-control-plane to run in privileged mode (config option allow-privileged=true)

As this charm has no standalone functionality, we'll need Kubernetes and Ceph clusters first.

Deploy Charmed Kubernetes configured to allow privileged containers as follows:

juju deploy charmed-kubernetes
juju config kubernetes-control-plane allow-privileged=true

Deploy a Ceph cluster consisting of three monitor and three storage nodes:

 juju deploy -n 3 ceph-mon
 juju deploy -n 3 ceph-osd --storage osd-devices=32G,2 --storage osd-journals=8G,1
 juju integrate ceph-osd:mon ceph-mon:osd

The storage nodes above will have two 32GB devices for storage and 8GB for journaling. As we have asked for 3 machines, this means a total of 192GB of storage and 24GB of journal space. The storage comes from whatever the default storage class is for the cloud (e.g., on AWS this will be EBS volumes).

Once the deployment has settled, We can add ceph-csi:

juju deploy ceph-csi
juju integrate ceph-csi:kubernetes kubernetes-control-plane:juju-info
juju integrate ceph-csi:ceph-client ceph-mon:client

If desired, add support for the Ceph distributed filesystem (CephFS) as follows:

juju deploy ceph-fs
juju integrate ceph-fs:ceph-mds ceph-mon:mds
juju config ceph-csi cephfs-enable=True

Check the Charmed Kubernetes cluster to verify Ceph cluster resources are available. Running:

juju ssh kubernetes-control-plane/leader -- kubectl get sc,po --namespace default

... should return output similar to:

NAME                                             PROVISIONER           RECLAIMPOLICY   VOLUMEBINDINGMODE   ALLOWVOLUMEEXPANSION   AGE
storageclass.storage.k8s.io/ceph-ext4            rbd.csi.ceph.com      Delete          Immediate           true                   142m
storageclass.storage.k8s.io/ceph-xfs (default)   rbd.csi.ceph.com      Delete          Immediate           true                   142m
storageclass.storage.k8s.io/cephfs               cephfs.csi.ceph.com   Delete          Immediate           true                   127m

NAME                                               READY   STATUS    RESTARTS   AGE
pod/csi-cephfsplugin-4gs22                         3/3     Running   0          127m
pod/csi-cephfsplugin-ljfw9                         3/3     Running   0          127m
pod/csi-cephfsplugin-mlbdx                         3/3     Running   0          127m
pod/csi-cephfsplugin-provisioner-9f479bcc4-48v8f   5/5     Running   0          127m
pod/csi-cephfsplugin-provisioner-9f479bcc4-92bt6   5/5     Running   0          127m
pod/csi-cephfsplugin-provisioner-9f479bcc4-hbb82   5/5     Running   0          127m
pod/csi-cephfsplugin-wlp2w                         3/3     Running   0          127m
pod/csi-cephfsplugin-xwdb2                         3/3     Running   0          127m
pod/csi-rbdplugin-b8nk8                            3/3     Running   0          142m
pod/csi-rbdplugin-bvqwn                            3/3     Running   0          142m
pod/csi-rbdplugin-provisioner-85dc49c6c-9rckg      7/7     Running   0          142m
pod/csi-rbdplugin-provisioner-85dc49c6c-f6h6k      7/7     Running   0          142m
pod/csi-rbdplugin-provisioner-85dc49c6c-n47fx      7/7     Running   0          142m
pod/csi-rbdplugin-vm25h                            3/3     Running   0          142m

When the ceph-csi charm is removed, it will not clean up Ceph pools that were created when the relation with ceph-mon:client was joined. The interface of ceph-mon:client does not seem to offer "removal" functionality. If you wish to remove ceph pools, use the delete-pool action on a ceph-mon unit.

Create and activate a virtualenv with the development requirements:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements-dev.txt

The Python operator framework includes a very nice harness for testing operator behaviour without full deployment. To execute unit tests, run:

tox -e unit

This charm contains functional tests which fully deploy the ceph-csi charm along with kubernetes and ceph cluster to test its functionality in a real environment. There are a few ways to run functional tests

tox -e func                          # Deploys new juju model and runs tests
tox -e func -- --model <model_name>  # Runs tests against existing model
tox -e func -- --keep-models         # Does not tear down model after tests are done (useful for debugging failing tests )

NOTE: If the environment which runs functional tests is behind a http proxy, you must export TEST_HTTPS_PROXY environment variable. Otherwise Kubernetes might have problem fetching docker images. Example:

export TEST_HTTPS_PROXY=http://10.0.0.1:3128
tox -e func