csi-digitalocean

A Container Storage Interface (CSI) Driver for DigitalOcean Block Storage. The CSI plugin allows you to use DigitalOcean Block Storage with your preferred Container Orchestrator.

The DigitalOcean CSI plugin is mostly tested on Kubernetes. Theoretically it should also work on other Container Orchestrators, such as Mesos or Cloud Foundry. Feel free to test it on other CO's and give us a feedback.

Releases

The DigitalOcean CSI plugin follows semantic versioning. The version will be bumped following the rules below:

• Bug fixes will be released as a PATCH update.
• New features (such as CSI spec bumps with no breaking changes) will be released as a MINOR update.
• Significant breaking changes makes a MAJOR update.

Features

Below is a list of functionality implemented by the plugin. In general, CSI features implementing an aspect of the specification are available on any DigitalOcean Kubernetes version for which beta support for the feature is provided.

See also the project examples for use cases.

Volume Expansion

Volumes can be expanded by updating the storage request value of the corresponding PVC:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: csi-pvc
  namespace: default
spec:
  [...]
  resources:
    requests:
      # The field below can be increased.
      storage: 10Gi
      [...]

After successful expansion, the status section of the PVC object will reflect the actual volume capacity.

Important notes:

• Volumes can only be increased in size, not decreased; attempts to do so will lead to an error.
• Expanding a volume that is larger than the target size will have no effect. The PVC object status section will continue to represent the actual volume capacity.
• Resizing volumes other than through the PVC object (e.g., the DigitalOcean cloud control panel) is not recommended as this can potentially cause conflicts. Additionally, size updates will not be reflected in the PVC object status section immediately, and the section will eventually show the actual volume capacity.

Raw Block Volume

Volumes can be used in raw block device mode by setting the volumeMode on the corresponding PVC:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: csi-pvc
  namespace: default
spec:
  [...]
  volumeMode: Block

Important notes:

• If using volume expansion functionality, only expansion of the underlying persistent volume is guaranteed. We do not guarantee to automatically expand the filesystem if you have formatted the device.

Volume Snapshots

Snapshots can be created and restored through VolumeSnapshot objects.

Note:

Version 1 of the CSI driver supports v1alpha1 Volume Snapshots only.

Version 2 and 3 of the CSI driver supports v1beta1 Volume Snapshots only.

Version 4 and later of the CSI driver support v1 Volume Snapshots only, which is backwards compatible to v1beta1. However, version 3 renders snapshots unusable that had previously been marked as invalid. See the csi-snapshotter documentation on the validating webhook and v1beta1 to v1 upgrade notes.

See also the example.

Volume Statistics

Volume statistics are exposed through the CSI-conformant endpoints. Monitoring systems such as Prometheus can scrape metrics and provide insights into volume usage.

Volume Transfer

Volumes can be transferred across clusters. The exact steps are outlined in our example.

Installing to Kubernetes

Kubernetes Compatibility

The following table describes the required DigitalOcean CSI driver version per supported Kubernetes release.

Note:

The DigitalOcean Kubernetes product comes with the CSI driver pre-installed and no further steps are required.

Driver modes

By default, the driver supports both the controller and node mode. It can manage DigitalOcean Volumes via the cloud API and mount them on the required node. The actually used mode is determined by how the driver is deployed and configured. The suggested release manifests provide separate deployments for controller and node modes, respectively.

When running outside of DigitalOcean droplets, the driver can only function in controller mode. This requires to set the --region flag to a valid DigitalOcean region slug in addition to the other flags.

The --region flag must not be set when running the driver on DigitalOcean droplets.

Alternatively driver can be run in node only mode on DigitalOcean droplets. Driver would only handle node related requests like mount volume. Driver runs in node only mode when --token flag is not provided.

Skip secret creation (section 1. in following deployment instructions) when using node only mode as API token is not required.

Requirements

• --allow-privileged flag must be set to true for the API server
• --allow-privileged flag must be set to true for the kubelet in Kubernetes 1.14 and below (flag does not exist in later releases)
• --feature-gates=KubeletPluginsWatcher=true,CSINodeInfo=true,CSIDriverRegistry=true feature gate flags must be set to true for both the API server and the kubelet
• Mount Propagation needs to be enabled. If you use Docker, the Docker daemon of the cluster nodes must allow shared mounts.

1. Create a secret with your DigitalOcean API Access Token

Replace the placeholder string starting with a05... with your own secret and save it as secret.yml:

apiVersion: v1
kind: Secret
metadata:
  name: digitalocean
  namespace: kube-system
stringData:
  access-token: "a05dd2f26b9b9ac2asdas__REPLACE_ME____123cb5d1ec17513e06da"

and create the secret using kubectl:

$ kubectl create -f ./secret.yml
secret "digitalocean" created

You should now see the digitalocean secret in the kube-system namespace along with other secrets

$ kubectl -n kube-system get secrets
NAME                  TYPE                                  DATA      AGE
default-token-jskxx   kubernetes.io/service-account-token   3         18h
digitalocean          Opaque                                1         18h

2. Provide authentication data for the snapshot validation webhook

Snapshots are validated through a ValidatingWebhookConfiguration which requires proper CA, certificate, and key data. The manifests in snapshot-validation-webhook.yaml should provide sufficient scaffolding to inject the data accordingly. However, the details on how to create and manage them is up to the user and dependent on the exact environment the webhook runs in. See the XXX-marked comments in the manifests file for user-required injection points.

The official snapshot webhook example offers a non-production-ready solution suitable for testing. For full production readiness, something like cert-manager can be leveraged.

3. Deploy the CSI plugin and sidecars

Always use the latest release compatible with your Kubernetes release (see the compatibility information).

The releases directory holds manifests for all plugin releases. You can deploy a specific version by executing the command

# Do *not* add a blank space after -f
kubectl apply -fhttps://raw.githubusercontent.com/digitalocean/csi-digitalocean/master/deploy/kubernetes/releases/csi-digitalocean-vX.Y.Z/{crds.yaml,driver.yaml,snapshot-controller.yaml}

where vX.Y.Z is the plugin target version. (Note that for releases older than v2.0.0, the driver was contained in a single YAML file. If you'd like to deploy an older release you need to use kubectl apply -fhttps://raw.githubusercontent.com/digitalocean/csi-digitalocean/master/deploy/kubernetes/releases/csi-digitalocean-vX.Y.Z.yaml)

If you see any issues during the installation, this could be because the newly created CRDs haven't been established yet. If you call kubectl apply -f again on the same file, the missing resources will be applied again.

The above does not include the snapshot validating webhook which needs extra configuration as outlined above. You may append ,snapshot-validation-webhook.yaml to the {...} list if you want to install a (presumably configured) webhook as well.

4. Test and verify

Create a PersistentVolumeClaim. This makes sure a volume is created and provisioned on your behalf:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: csi-pvc
spec:
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 5Gi
  storageClassName: do-block-storage

Check that a new PersistentVolume is created based on your claim:

$ kubectl get pv
NAME                                       CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS    CLAIM             STORAGECLASS       REASON    AGE
pvc-0879b207-9558-11e8-b6b4-5218f75c62b9   5Gi        RWO            Delete           Bound     default/csi-pvc   do-block-storage             3m

The above output means that the CSI plugin successfully created (provisioned) a new Volume on behalf of you. You should be able to see this newly created volume under the Volumes tab in the DigitalOcean UI

The volume is not attached to any node yet. It'll only attached to a node if a workload (i.e: pod) is scheduled to a specific node. Now let us create a Pod that refers to the above volume. When the Pod is created, the volume will be attached, formatted and mounted to the specified Container:

kind: Pod
apiVersion: v1
metadata:
  name: my-csi-app
spec:
  containers:
    - name: my-frontend
      image: busybox
      volumeMounts:
      - mountPath: "/data"
        name: my-do-volume
      command: [ "sleep", "1000000" ]
  volumes:
    - name: my-do-volume
      persistentVolumeClaim:
        claimName: csi-pvc

Check if the pod is running successfully:

kubectl describe pods/my-csi-app

Write inside the app container:

$ kubectl exec -ti my-csi-app /bin/sh
/ # touch /data/hello-world
/ # exit
$ kubectl exec -ti my-csi-app /bin/sh
/ # ls /data
hello-world

Upgrading

When upgrading to a new Kubernetes minor version, you should upgrade the CSI driver to match. See the table above for which driver version is used with each Kubernetes version.

Special consideration is necessary when upgrading from Kubernetes 1.11 or earlier, which uses CSI driver version 0.2 or earlier. In these early releases, the driver name was com.digitalocean.csi.dobs, while in all subsequent releases it is dobs.csi.digitalocean.com. When upgrading, use the commandline flag --driver-name to force the new driver to use the old name. Failing to do so will cause any existing PVs to be unusable since the new driver will not manage them and the old driver is no longer running.

Configuration

Default volumes paging size

Some CSI driver operations require paging through the volumes returned from the DO Volumes API. By default, the page size is not defined and causes the DO API to choose a value as specified in the API reference. In the vast majority of cases, this should work fine. However, for accounts with a very large number of volumes, the API server-chosen default page size may be too small to return all volumes within the configured (sidecar-provided) timeout.

For that reason, the default page size can be customized by passing the --default-volumes-page-size flag a positive number.

Notes:

1. The user is responsible for selecting a value below the maximum limit mandated by the DO API. Please see the API reference link above to see the current limit.
2. The configured sidecar timeout values may need to be aligned with the chosen page size. In particular, csi-attacher invokes ListVolumes to periodically synchronize the API and cluster-local volume states; as such, its timeout must be large enough to account for the expected number of volumes in the given account and region.
3. The default page size does not become effective if an explicit page size (more precisely, max entries in CSI spec speak) is passed to a given gRPC method.

API rate limiting

DO API usage is subject to certain rate limits. In order to protect against running out of quota for extremely heavy regular usage or pathological cases (e.g., bugs or API thrashing due to an interfering third-party controller), a custom rate limit can be configured via the --do-api-rate-limit flag. It accepts a float value, e.g., --do-api-rate-limit=3.5 to restrict API usage to 3.5 queries per second.

Flags

The --validate-attachment options adds an additional validation which checks for the /sys/class/block/<device name>/device/state file content for the running status. When enabling this flag, it prevents a racing condition where the DOBS volumes aren't fully attached which can be misinterpreted by the CSI implementation causing a force format of the volume which results in data loss.

Development

Requirements:

• Go at the version specified in .github/workflows/test.yaml
• Docker (for building via the Makefile, post-unit testing, and publishing)

Dependencies are managed via Go modules.

PRs from the code-hosting repository are automatically unit- and end-to-end-tested in our CI (implemented by Github Actions). See the .github/workflows directory for details.

For every green build of the master branch, the container image digitalocean/do-csi-plugin:master is updated and pushed at the end of the CI run. This allows to test the latest commit easily.

Steps to run the tests manually are outlined below.

Unit Tests

To execute the unit tests locally, run:

make test

End-to-End Tests

To manually run the end-to-end tests, you need to build a container image for your change first and publish it to a registry. Repository owners can publish under digitalocean/do-csi-plugin:dev:

VERSION=dev make publish

If you do not have write permissions to digitalocean/do-csi-plugin on Docker Hub or are worried about conflicting usage of that tag, you can also publish under a different (presumably personal) organization:

DOCKER_REPO=johndoe VERSION=latest-feature make publish

This would yield the published container image johndoe/do-csi-plugin:latest-feature.

Assuming you have your DO API token assigned to the DIGITALOCEAN_ACCESS_TOKEN environment variable, you can then spin up a DOKS cluster on-the-fly and execute the upstream end-to-end tests for a given set of Kubernetes versions like this:

make test-e2e E2E_ARGS="-driver-image johndoe/do-csi-plugin:latest-feature 1.16 1.15 1.14"

See our documentation for an overview on how the end-to-end tests work as well as usage instructions.

Integration Tests

There is a set of custom integration tests which are mostly useful for Kubernetes pre-1.14 installations as these are not covered by the upstream end-to-end tests.

To run the integration tests on a DOKS cluster, follow the instructions.

Prepare CSI driver for a new Kubernetes minor version

1. Review recently merged PRs and any in-progress / planned work to ensure any bugs scheduled for the release have been fixed and merged.
2. Bump kubernetes dependency versions
3. Support running e2e on new $MAJOR.$MINOR
   1. Since we only support three minor versions at a time. E2e tests for the oldest supported version can be removed.
4. Verify e2e tests pass - see here about running tests locally
5. Prepare for release
6. Perform release

See e2e test README on how to run conformance tests locally.

Updating the Kubernetes dependencies

Run

make NEW_KUBERNETES_VERSION=X.Y.Z update-k8s

to update the Kubernetes dependencies to version X.Y.Z.

Note: Make sure to also add support to the e2e tests for the new kubernetes version, following these instructions.

Releasing

Releases may happen either for the latest minor version of the CSI driver maintained in the master branch, or an older minor version still maintained in one of the release-* branches. In this section, we will call that branch the release branch.

To release a new version vX.Y.Z, first check out the release branch and bump the version:

make NEW_VERSION=vX.Y.Z bump-version

This will create the set of files specific to a new release. Make sure everything looks good; in particular, ensure that the change log is up-to-date and is not missing any important, user-facing changes.

Create a new branch with all changes:

git checkout -b prepare-release-vX.Y.Z
git add .
git push origin

After it is merged to the release branch, wait for the release branch build to go green. (This will entail another run of the entire test suite.)

Finally, check out the release branch again, tag the release, and push it:

git checkout <release branch>
git pull
git tag vX.Y.Z
git push origin vX.Y.Z

(This works for non-master release branches as well since the checkout Github Action we use defaults to checking out the ref/SHA that triggered the workflow.)

The CI will publish the container image digitalocean/do-csi-plugin:vX.Y.Z and create a Github Release under the name vX.Y.Z automatically. Nothing else needs to be done.

Contributing

At DigitalOcean we value and love our community! If you have any issues or would like to contribute, feel free to open an issue or PR.