CSI Snapshotter

The CSI snapshotter is part of Kubernetes implementation of Container Storage Interface (CSI).

The volume snapshot feature supports CSI v1.0 and higher. It was introduced as an Alpha feature in Kubernetes v1.12 and has been promoted to an Beta feature in Kubernetes 1.17. In Kubernetes 1.20, the volume snapshot feature moves to GA.

⚠️ WARNING: There is a new validating webhook server which provides tightened validation on snapshot objects. This SHOULD be installed by all users of this feature. More details below.

Overview

With the promotion of Volume Snapshot to GA, the feature is enabled by default on standard Kubernetes deployments and cannot be turned off.

Blog post for the GA feature can be found here

Compatibility

This information reflects the head of this branch.

Compatible with CSI Version	Container Image	Min K8s Version
CSI Spec v1.2.0	k8s.gcr.io/sig-storage/csi-snapshotter	1.17
CSI Spec v1.2.0	k8s.gcr.io/sig-storage/snapshot-controller	1.17
CSI Spec v1.2.0	k8s.gcr.io/sig-storage/snapshot-validation-webhook	1.17

Note: snapshot-controller, snapshot-validation-webhook, csi-snapshotter v4.0 requires v1 snapshot CRDs to be installed, but it supports both v1 and v1beta1 snapshot objects.

Feature Status

The VolumeSnapshotDataSource feature gate was introduced in Kubernetes 1.12 and it is enabled by default in Kubernetes 1.17 when the volume snapshot feature is promoted to beta. In Kubernetes 1.20, the feature gate is enabled by default on standard Kubernetes deployments and cannot be turned off.

Design

Both the snapshot controller and CSI external-snapshotter sidecar follow controller pattern and uses informers to watch for events. The snapshot controller watches for VolumeSnapshot and VolumeSnapshotContent create/update/delete events.

The CSI external-snapshotter sidecar only watches for VolumeSnapshotContent create/update/delete events. It filters out these objects with Driver==<CSI driver name> specified in the associated VolumeSnapshotClass object and then processes these events in workqueues with exponential backoff.

The CSI external-snapshotter sidecar talks to CSI over socket (/run/csi/socket by default, configurable by -csi-address).

Snapshot v1 APIs

Other than introducing tightening validation, there is no difference between the v1beta1 and v1 Kubernetes volume snapshot API. In the current release, both v1 and v1beta1 are served while the stored API version is still v1beta1. Future releases will switch the stored version to v1 and gradually remove v1beta1 support.

Usage

Volume Snapshot feature contains the following components:

• Kubernetes Volume Snapshot CRDs
• Volume snapshot controller
• Snapshot validation webhook
• CSI Driver along with CSI Snapshotter sidecar

The Volume Snapshot feature depends on a volume snapshot controller and the volume snapshot CRDs. Both the volume snapshot controller and the CRDs are independent of any CSI driver. Regardless of the number CSI drivers deployed on the cluster, there must be only one instance of the volume snapshot controller running and one set of volume snapshot CRDs installed per cluster.

Therefore, it is strongly recommended that Kubernetes distributors bundle and deploy the controller and CRDs as part of their Kubernetes cluster management process (independent of any CSI Driver).

If your Kubernetes distribution does not bundle the snapshot controller, you may manually install these components by executing the following steps. Note that the snapshot controller YAML files in the git repository deploy into the default namespace for system testing purposes. For general use, update the snapshot controller YAMLs with an appropriate namespace prior to installing. For example, on a Vanilla Kubernetes cluster update the namespace from 'default' to 'kube-system' prior to issuing the kubectl create command.

There is a new validating webhook server which provides tightened validation on snapshot objects. The cluster admin or Kubernetes distribution admin should install the webhook alongside the snapshot controllers and CRDs. More details below.

Install Snapshot Beta CRDs:

• kubectl create -f client/config/crd
• https://github.com/kubernetes-csi/external-snapshotter/tree/master/client/config/crd
• Do this once per cluster

Install Common Snapshot Controller:

• Update the namespace to an appropriate value for your environment (e.g. kube-system)
• kubectl create -f deploy/kubernetes/snapshot-controller
• https://github.com/kubernetes-csi/external-snapshotter/tree/master/deploy/kubernetes/snapshot-controller
• Do this once per cluster

Install CSI Driver:

• Follow instructions provided by your CSI Driver vendor.
• Here is an example to install the sample hostpath CSI driver
  • kubectl create -f deploy/kubernetes/csi-snapshotter
  • https://github.com/kubernetes-csi/external-snapshotter/tree/master/deploy/kubernetes/csi-snapshotter

Validating Webhook

The snapshot validating webhook is an HTTP callback which responds to admission requests. It is part of a larger plan to tighten validation for volume snapshot objects. This webhook introduces the ratcheting validation mechanism targeting the tighter validation. The cluster admin or Kubernetes distribution admin should install the webhook alongside the snapshot controllers and CRDs.

Along with the validation webhook, the volume snapshot controller will start labeling invalid snapshot objects which already existed. This is to enable quick identification of invalid snapshot objects in the system by running:

kubectl get volumesnapshots --selector=snapshot.storage.kubernetes.io/invalid-snapshot-resource: ""
kubectl get volumesnapshotcontents --selector=snapshot.storage.kubernetes.io/invalid-snapshot-content-resource: ""

Users should run this to identify, remove any invalid objects, and correct their workflows before upgrading to v1. Once the API has been switched to the v1 type, those invalid objects will not be deletable from the system.

If there are no existing invalid v1beta1 objects, after upgrading to v1, the webhook and schema validation will prevent the user from creating new invalid v1 and v1beta1 objects.

If there are existing invalid v1beta1 objects, the user should make sure that the snapshot controller is upgraded to v3.0.0 or higher (v3.0.3 is the latest recommended v3.0.x release) and install the corresponding validation webhook before upgrading to v1 so that those invalid objects will be labeled and can be identified easily and removed before upgrading to v1.

If there are existing invalid v1beta1 objects and the user didn't upgrade to the snapshot controller 3.0.0 or higher and install the corresponding validation webhook before upgrading to v1, those existing invalid v1beta1 objects will not be labeled by the snapshot controller.

So the recommendation is that before upgrading to v1 CRDs and upgrading snapshot controller and validation webhook to v4.0, the user should upgrade to the snapshot controller 3.0.0 and higher (v3.0.3 is the latest recommended version for 3.0.x) and install the corresponding validation webhook so that all existing invalid objects will be labeled and can be easily identified and deleted.

⚠️ WARNING: Cluster admins choosing not to install the webhook server and participate in the phased release process can cause future problems when upgrading from v1beta1 to v1 volumesnapshot API, if there are currently persisted objects which fail the new stricter validation. Potential impacts include being unable to delete invalid snapshot objects.

Read more about how to install the example webhook here.

Validating Webhook Command Line Options

• --tls-cert-file: File containing the x509 Certificate for HTTPS. (CA cert, if any, concatenated after server cert). Required.

• --tls-private-key-file: File containing the x509 private key matching --tls-cert-file. Required.

• --port: Secure port that the webhook listens on (default 443)

Snapshot controller command line options

Important optional arguments that are highly recommended to be used

• --leader-election: Enables leader election. This is useful when there are multiple replicas of the same snapshot controller running for the same Kubernetes deployment. Only one of them may be active (=leader). A new leader will be re-elected when current leader dies or becomes unresponsive for ~15 seconds.

• --leader-election-namespace <namespace>: The namespace where the leader election resource exists. Defaults to the pod namespace if not set.

• --http-endpoint: The TCP network address where the HTTP server for diagnostics, including metrics and leader election health check, will listen (example: :8080 which corresponds to port 8080 on local host). The default is empty string, which means the server is disabled.

• --metrics-address: (deprecated) The TCP network address where the prometheus metrics endpoint will run (example: :8080 which corresponds to port 8080 on local host). The default is empty string, which means metrics endpoint is disabled.

• --metrics-path: The HTTP path where prometheus metrics will be exposed. Default is /metrics.

• --worker-threads: Number of worker threads. Default value is 10.

Other recognized arguments

• --kubeconfig <path>: Path to Kubernetes client configuration that the snapshot controller uses to connect to Kubernetes API server. When omitted, default token provided by Kubernetes will be used. This option is useful only when the snapshot controller does not run as a Kubernetes pod, e.g. for debugging.

• --resync-period <duration>: Internal resync interval when the snapshot controller re-evaluates all existing VolumeSnapshot instances and tries to fulfill them, i.e. create / delete corresponding snapshots. It does not affect re-tries of failed calls! It should be used only when there is a bug in Kubernetes watch logic. Default is 15 minutes.

• --version: Prints current snapshot controller version and quits.

• All glog / klog arguments are supported, such as -v <log level> or -alsologtostderr.

CSI external snapshotter sidecar command line options

Important optional arguments that are highly recommended to be used

• --csi-address <path to CSI socket>: This is the path to the CSI driver socket inside the pod that the external-snapshotter container will use to issue CSI operations (/run/csi/socket is used by default).

• --leader-election: Enables leader election. This is useful when there are multiple replicas of the same external-snapshotter running for one CSI driver. Only one of them may be active (=leader). A new leader will be re-elected when current leader dies or becomes unresponsive for ~15 seconds.

• --leader-election-namespace <namespace>: The namespace where the leader election resource exists. Defaults to the pod namespace if not set.

• --timeout <duration>: Timeout of all calls to CSI driver. It should be set to value that accommodates majority of CreateSnapshot, DeleteSnapshot, and ListSnapshots calls. 1 minute is used by default.

• snapshot-name-prefix: Prefix to apply to the name of a created snapshot. Default is snapshot.

• snapshot-name-uuid-length: Length in characters for the generated uuid of a created snapshot. Defaults behavior is to NOT truncate.

• --worker-threads: Number of worker threads for running create snapshot and delete snapshot operations. Default value is 10.

Other recognized arguments

• --kubeconfig <path>: Path to Kubernetes client configuration that the CSI external-snapshotter uses to connect to Kubernetes API server. When omitted, default token provided by Kubernetes will be used. This option is useful only when the external-snapshotter does not run as a Kubernetes pod, e.g. for debugging.

• --resync-period <duration>: Internal resync interval when the CSI external-snapshotter re-evaluates all existing VolumeSnapshotContent instances and tries to fulfill them, i.e. update / delete corresponding snapshots. It does not affect re-tries of failed CSI calls! It should be used only when there is a bug in Kubernetes watch logic. Default is 15 minutes.

• --version: Prints current CSI external-snapshotter version and quits.

• All glog / klog arguments are supported, such as -v <log level> or -alsologtostderr.

HTTP endpoint

The external-snapshotter optionally exposes an HTTP endpoint at address:port specified by --http-endpoint argument. When set, these two paths are exposed:

• Metrics path, as set by --metrics-path argument (default is /metrics).

• Leader election health check at /healthz/leader-election. It is recommended to run a liveness probe against this endpoint when leader election is used to kill external-provisioner leader that fails to connect to the API server to renew its leadership. See kubernetes-csi/csi-lib-utils#66 for details.

Upgrade

Upgrade from v1alpha1 to v1beta1

The change from v1alpha1 to v1beta1 snapshot APIs is not backward compatible.

If you have already deployed v1alpha1 snapshot APIs and external-snapshotter sidecar controller and want to upgrade to v1beta1, you need to do the following:

• Note: The underlying snapshots on the storage system will be deleted in the upgrade process!!!

1. Delete volume snapshots created using v1alpha1 snapshot CRDs and external-snapshotter sidecar controller.
2. Uninstall v1alpha1 snapshot CRDs, external-snapshotter sidecar controller, and CSI driver.
3. Install v1beta1 snapshot CRDs, snapshot controller, CSI external-snapshotter sidecar and CSI driver.

Upgrade from v1beta1 to v1

Validation webhook should be installed before upgrading to v1. Potential impacts of not installing the validation webhook before upgrading to v1 include being unable to delete invalid snapshot objects. See the section on Validation Webhook for details.

Change from v1beta1 to v1 is backward compatible. Both v1 and v1beta1 are served while the stored API version is still v1beta1. Future releases will switch the stored version to v1 and gradually remove v1beta1 support.

Testing

Running Unit Tests:

go test -timeout 30s  github.com/kubernetes-csi/external-snapshotter/pkg/common-controller

go test -timeout 30s  github.com/kubernetes-csi/external-snapshotter/pkg/sidecar-controller

CRDs and Client Library

Volume snapshot APIs and client library are now in a separate sub-module: github.com/kubernetes-csi/external-snapshotter/client/v4.

Use the command go get -u github.com/kubernetes-csi/external-snapshotter/client/[email protected] to get the client library.

Setting Quota limits with Snapshot custom resources

ResourceQuotas are namespaced objects that can be used to set limits on objects of a particular Group.Version.Kind. Before we set resource quota, make sure that snapshot CRDs are installed in the cluster. If not please follow this guide.

kubectl get crds | grep snapshot

Now create a ResourceQuota object which sets the limits on number of volumesnapshots that can be created:

apiVersion: v1
kind: ResourceQuota
metadata:
  name: snapshot-quota
spec:
  hard:
    count/volumesnapshots.snapshot.storage.k8s.io: "10"

If you try to create more snapshots than what is allowed, you will see error like the following:

Error from server (Forbidden): error when creating "csi-snapshot.yaml": volumesnapshots.snapshot.storage.k8s.io "new-snapshot-demo" is forbidden: exceeded quota: snapshot-quota, requested: count/volumesnapshots.snapshot.storage.k8s.io=1, used: count/volumesnapshots.snapshot.storage.k8s.io=10, limited: count/volumesnapshots.snapshot.storage.k8s.io=10

Dependency Management

external-snapshotter uses go modules.

Community, discussion, contribution, and support

Learn how to engage with the Kubernetes community on the community page.

You can reach the maintainers of this project at:

• Slack channel
• Mailing list

Code of conduct

Participation in the Kubernetes community is governed by the Kubernetes Code of Conduct.