CSI provisioner

The external-provisioner is a sidecar container that dynamically provisions volumes by calling CreateVolume and DeleteVolume functions of CSI drivers. It is necessary because internal persistent volume controller running in Kubernetes controller-manager does not have any direct interfaces to CSI drivers.

Overview

The external-provisioner is an external controller that monitors PersistentVolumeClaim objects created by user and creates/deletes volumes for them. The Kubernetes Container Storage Interface (CSI) Documentation explains how to develop, deploy, and test a Container Storage Interface (CSI) driver on Kubernetes.

Compatibility

This information reflects the head of this branch.

Feature status

Various external-provisioner releases come with different alpha / beta features. Check --help output for alpha/beta features in each release.

Following table reflects the head of this branch.

All other external-provisioner features and the external-provisioner itself is considered GA and fully supported.

Usage

It is necessary to create a new service account and give it enough privileges to run the external-provisioner, see deploy/kubernetes/rbac.yaml. The provisioner is then deployed as single Deployment as illustrated below:

kubectl create deploy/kubernetes/deployment.yaml

The external-provisioner may run in the same pod with other external CSI controllers such as the external-attacher, external-snapshotter and/or external-resizer.

Note that the external-provisioner does not scale with more replicas. Only one external-provisioner is elected as leader and running. The others are waiting for the leader to die. They re-elect a new active leader in ~15 seconds after death of the old leader.

Command line options

Recommended optional arguments

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

• --leader-election: Enables leader election. This is mandatory when there are multiple replicas of the same external-provisioner 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 where leader election object will be created. It is recommended that this parameter is populated from Kubernetes DownwardAPI with the namespace where the external-provisioner runs in.

• --leader-election-lease-duration <duration>: Duration, in seconds, that non-leader candidates will wait to force acquire leadership. Defaults to 15 seconds.

• --leader-election-renew-deadline <duration>: Duration, in seconds, that the acting leader will retry refreshing leadership before giving up. Defaults to 10 seconds.

• --leader-election-retry-period <duration>: Duration, in seconds, the LeaderElector clients should wait between tries of actions. Defaults to 5 seconds.

• --timeout <duration>: Timeout of all calls to CSI driver. It should be set to value that accommodates majority of ControllerCreateVolume and ControllerDeleteVolume calls. See CSI error and timeout handling for details. 15 seconds is used by default.

• --retry-interval-start <duration>: Initial retry interval of failed provisioning or deletion. It doubles with each failure, up to --retry-interval-max and then it stops increasing. Default value is 1 second. See CSI error and timeout handling for details.

• --retry-interval-max <duration>: Maximum retry interval of failed provisioning or deletion. Default value is 5 minutes. See CSI error and timeout handling for details.

• --worker-threads <num>: Number of simultaneously running ControllerCreateVolume and ControllerDeleteVolume operations. Default value is 100.

• --kube-api-qps <num>: The number of requests per second sent by a Kubernetes client to the Kubernetes API server. Defaults to 5.0.

• --kube-api-burst <num>: The number of requests to the Kubernetes API server, exceeding the QPS, that can be sent at any given time. Defaults to 10.

• --cloning-protection-threads <num>: Number of simultaneously running threads, handling cloning finalizer removal. Defaults to 1.

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

• --extra-create-metadata: Enables the injection of extra PVC and PV metadata as parameters when calling CreateVolume on the driver (keys: "csi.storage.k8s.io/pvc/name", "csi.storage.k8s.io/pvc/namespace", "csi.storage.k8s.io/pv/name")

• controller-publish-readonly: This option enables PV to be marked as readonly at controller publish volume call if PVC accessmode has been set to ROX. Defaults to false.

• --enable-pprof: Enable pprof profiling on the TCP network address specified by --http-endpoint. The HTTP path is /debug/pprof/.

Storage capacity arguments

See the storage capacity section below for details.

• --enable-capacity: This enables producing CSIStorageCapacity objects with capacity information from the driver's GetCapacity call. The default is to not produce CSIStorageCapacity objects.

• --capacity-ownerref-level <levels>: The level indicates the number of objects that need to be traversed starting from the pod identified by the POD_NAME and NAMESPACE environment variables to reach the owning object for CSIStorageCapacity objects: 0 for the pod itself, 1 for a StatefulSet and DaemonSet, 2 for a Deployment, etc. Defaults to 1 (= StatefulSet). Ownership is optional and can be disabled with -1.

• --capacity-threads <num>: Number of simultaneously running threads, handling CSIStorageCapacity objects. Defaults to 1.

• --capacity-poll-interval <interval>: How long the external-provisioner waits before checking for storage capacity changes. Defaults to 1m.

• --capacity-for-immediate-binding <bool>: Enables producing capacity information for storage classes with immediate binding. Not needed for the Kubernetes scheduler, maybe useful for other consumers or for debugging. Defaults to false.

Distributed provisioning

• --node-deployment: Enables deploying the external-provisioner together with a CSI driver on nodes to manage node-local volumes. Off by default.

• --node-deployment-immediate-binding: Determines whether immediate binding is supported when deployed on each node. Enabled by default, use --node-deployment-immediate-binding=false if not desired. Disabling it may be useful for example when a custom controller will select nodes for PVCs.

• --node-deployment-base-delay: Determines how long the external-provisioner sleeps initially before trying to own a PVC with immediate binding. Defaults to 20 seconds.

• --node-deployment-max-delay: Determines how long the external-provisioner sleeps at most before trying to own a PVC with immediate binding. Defaults to 60 seconds.

Other recognized arguments

• --feature-gates <gates>: A set of comma separated <feature-name>=<true|false> pairs that describe feature gates for alpha/experimental features. See list of features or --help output for list of recognized features. Example: --feature-gates Topology=true to enable Topology feature that's disabled by default.

• --strict-topology: This controls what topology information is passed to CreateVolumeRequest.AccessibilityRequirements in case of delayed binding. See the table below for an explanation how this option changes the result. This option has no effect if either Topology feature is disabled or Immediate volume binding mode is used.

• --immediate-topology: This controls what topology information is passed to CreateVolumeRequest.AccessibilityRequirements in case of immediate binding. See the table below for an explanation how this option changes the result. This option has no effect if either Topology feature is disabled or WaitForFirstConsumer (= delayed) volume binding mode is used. The default is true, so use --immediate-topology=false to disable it. It should not be disabled if the CSI driver might create volumes in a topology segment that is not accessible in the cluster. Such a driver should use the topology information to create new volumes where they can be accessed.

• --kubeconfig <path>: Path to Kubernetes client configuration that the external-provisioner 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-provisioner does not run as a Kubernetes pod, e.g. for debugging. Either this or --master needs to be set if the external-provisioner is being run out of cluster.

• --master <url>: Master URL to build a client config from. When omitted, default token provided by Kubernetes will be used. This option is useful only when the external-provisioner does not run as a Kubernetes pod, e.g. for debugging. Either this or --kubeconfig needs to be set if the external-provisioner is being run out of cluster.

• --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.

• --volume-name-prefix <prefix>: Prefix of PersistentVolume names created by the external-provisioner. Default value is "pvc", i.e. created PersistentVolume objects will have name pvc-<uuid>.

• --volume-name-uuid-length: Length of UUID to be added to --volume-name-prefix. Default behavior is to NOT truncate the UUID.

• --version: Prints current external-provisioner version and quits.

• --prevent-volume-mode-conversion: Prevents an unauthorized user from modifying the volume mode when creating a PVC from an existing VolumeSnapshot. Defaults to true.

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

Design

External-provisioner interacts with Kubernetes by watching PVCs and PVs and implementing the external provisioner protocol. The design document explains this in more detail.

Topology support

When Topology feature is enabled* and the driver specifies VOLUME_ACCESSIBILITY_CONSTRAINTS in its plugin capabilities, external-provisioner prepares CreateVolumeRequest.AccessibilityRequirements while calling Controller.CreateVolume. The driver has to consider these topology constraints while creating the volume. Below table shows how these AccessibilityRequirements are prepared:

*) Topology feature gate is enabled by default since v5.0.

When enabling topology support in a CSI driver that had it disabled, please make sure the topology is first enabled in the driver's node DaemonSet and topology labels are populated on all nodes. The topology can be then updated in the driver's Deployment and its external-provisioner sidecar.

Capacity support

The external-provisioner can be used to create CSIStorageCapacity objects that hold information about the storage capacity available through the driver. The Kubernetes scheduler then uses that information when selecting nodes for pods with unbound volumes that wait for the first consumer.

All CSIStorageCapacity objects created by an instance of the external-provisioner have certain labels:

• csi.storage.k8s.io/drivername: the CSI driver name
• csi.storage.k8s.io/managed-by: external-provisioner for central provisioning, external-provisioner- for distributed provisioning

They get created in the namespace identified with the NAMESPACE environment variable.

Each external-provisioner instance manages exactly those objects with the labels that correspond to the instance.

Optionally, all CSIStorageCapacity objects created by an instance of the external-provisioner can have an owner. This ensures that the objects get removed automatically when uninstalling the CSI driver. The owner is determined with the POD_NAME/NAMESPACE environment variables and the --capacity-ownerref-level parameter. Setting an owner reference is highly recommended whenever possible (i.e. in the most common case that drivers are run inside containers).

If ownership is disabled the storage admin is responsible for removing orphaned CSIStorageCapacity objects, and the following command can be used to clean up orphaned objects of a driver:

kubectl delete csistoragecapacities -l csi.storage.k8s.io/drivername=my-csi.example.com

When switching from a deployment without ownership to one with ownership, managed objects get updated such that they have the configured owner. When switching in the other direction, the owner reference is not removed because the new deployment doesn't know what the old owner was.

To enable this feature in a driver deployment with a central controller (see also the deploy/kubernetes/storage-capacity.yaml example):

• Set the POD_NAME and NAMESPACE environment variables like this:

  env:
  - name: NAMESPACE
   valueFrom:
     fieldRef:
       fieldPath: metadata.namespace
  - name: POD_NAME
   valueFrom:
     fieldRef:
       fieldPath: metadata.name

• Add --enable-capacity to the command line flags.
• Add StorageCapacity: true to the CSIDriver information object. Without it, external-provisioner will publish information, but the Kubernetes scheduler will ignore it. This can be used to first deploy the driver without that flag, then when sufficient information has been published, enabled the scheduler usage of it.
• If external-provisioner is not deployed with a StatefulSet, then configure with --capacity-ownerref-level which object is meant to own CSIStorageCapacity objects.
• Optional: configure how often external-provisioner polls the driver to detect changed capacity with --capacity-poll-interval.
• Optional: configure how many worker threads are used in parallel with --capacity-threads.
• Optional: enable producing information also for storage classes that use immediate volume binding with --capacity-for-immediate-binding. This is usually not needed because such volumes are created by the driver without involving the Kubernetes scheduler and thus the published information would just be ignored.

To determine how many different topology segments exist, external-provisioner uses the topology keys and labels that the CSI driver instance on each node reports to kubelet in the NodeGetInfoResponse.accessible_topology field. The keys are stored by kubelet in the CSINode objects and the actual values in Node annotations.

CSI drivers must report topology information that matches the storage pool(s) that it has access to, with granularity that matches the most restrictive pool.

For example, if the driver runs in a node with region/rack topology and has access to per-region storage as well as per-rack storage, then the driver should report topology with region/rack as its keys. If it only has access to per-region storage, then it should just use region as key. If it uses region/rack, then redundant CSIStorageCapacity objects will be published, but the information is still correct. See the KEP for details.

For each segment and each storage class, CSI GetCapacity is called once with the topology of the segment and the parameters of the class. If there is no error and the capacity is non-zero, a CSIStorageCapacity object is created or updated (if it already exists from a prior call) with that information. Obsolete objects are removed.

To ensure that CSIStorageCapacity objects get removed when the external-provisioner gets removed from the cluster, they all have an owner and therefore get garbage-collected when that owner disappears. The owner is not the external-provisioner pod itself but rather one of its parents as specified by --capacity-ownerref-level. This way, it is possible to switch between external-provisioner instances without losing the already gathered information.

CSIStorageCapacity objects are namespaced and get created in the namespace of the external-provisioner. Only CSIStorageCapacity objects with the right owner are modified by external-provisioner and their name is generated, so it is possible to deploy different drivers in the same namespace. However, Kubernetes does not check who is creating CSIStorageCapacity objects, so in theory a malfunctioning or malicious driver deployment could also publish incorrect information about some other driver.

The deployment with distributed provisioning is almost the same as above, with some minor change:

• Use --capacity-ownerref-level=1 and the NAMESPACE/POD_NAME variables to make the DaemonSet that contains the external-provisioner the owner of CSIStorageCapacity objects for the node.

Deployments of external-provisioner outside the Kubernetes cluster are also possible, albeit only without an owner for the objects. NAMESPACE still needs to be set to some existing namespace also in this case.

CSI error and timeout handling

The external-provisioner invokes all gRPC calls to CSI driver with timeout provided by --timeout command line argument (15 seconds by default).

Correct timeout value and number of worker threads depends on the storage backend and how quickly it is able to process ControllerCreateVolume and ControllerDeleteVolume calls. The value should be set to accommodate majority of them. It is fine if some calls time out - such calls will be retried after exponential backoff (starting with 1s by default), however, this backoff will introduce delay when the call times out several times for a single volume.

Frequency of ControllerCreateVolume and ControllerDeleteVolume retries can be configured by --retry-interval-start and --retry-interval-max parameters. The external-provisioner starts retries with retry-interval-start interval (1s by default) and doubles it with each failure until it reaches retry-interval-max (5 minutes by default). The external provisioner stops increasing the retry interval when it reaches retry-interval-max, however, it still retries provisioning/deletion of a volume until it's provisioned. The external-provisioner keeps its own number of provisioning/deletion failures for each volume.

The external-provisioner can invoke up to --worker-threads (100 by default) ControllerCreateVolume and up to --worker-threads (100 by default) ControllerDeleteVolume calls in parallel, i.e. these two calls are counted separately. The external-provisioner assumes that the storage backend can cope with such high number of parallel requests and that the requests are handled in relatively short time (ideally sub-second). Lower value should be used for storage backends that expect slower processing related to newly created / deleted volumes or can handle lower amount of parallel calls.

Details of error handling of individual CSI calls:

• ControllerCreateVolume: The call might have timed out just before the driver provisioned a volume and was sending a response. From that reason, timeouts from ControllerCreateVolume is considered as "volume may be provisioned" or "volume is being provisioned in the background." The external-provisioner will retry calling ControllerCreateVolume after exponential backoff until it gets either successful response or final (non-timeout) error that the volume cannot be created.
• ControllerDeleteVolume: This is similar to ControllerCreateVolume, The external-provisioner will retry calling ControllerDeleteVolume with exponential backoff after timeout until it gets either successful response or a final error that the volume cannot be deleted.
• Probe: The external-provisioner retries calling Probe until the driver reports it's ready. It retries also when it receives timeout from Probe call. The external-provisioner has no limit of retries. It is expected that ReadinessProbe on the driver container will catch case when the driver takes too long time to get ready.
• GetPluginInfo, GetPluginCapabilitiesRequest, ControllerGetCapabilities: The external-provisioner expects that these calls are quick and does not retry them on any error, including timeout. Instead, it assumes that the driver is faulty and exits. Note that Kubernetes will likely start a new provisioner container and it will start with Probe call.

HTTP endpoint

The external-provisioner 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 https://github.com/kubernetes-csi/csi-lib-utils/issues/66 for details.

Deployment on each node

Normally, external-provisioner is deployed once in a cluster and communicates with a control instance of the CSI driver which then provisions volumes via some kind of storage backend API. CSI drivers which manage local storage on a node don't have such an API that a central controller could use.

To support this case, external-provisioner can be deployed alongside each CSI driver on different nodes. The CSI driver deployment must:

• support topology, usually with one topology key ("csi.example.org/node") and the Kubernetes node name as value
• use a service account that has the same RBAC rules as for a normal deployment
• invoke external-provisioner with --node-deployment
• tweak --node-deployment-base-delay and --node-deployment-max-delay to match the expected cluster size and desired response times (only relevant when there are storage classes with immediate binding, see below for details)
• set the NODE_NAME environment variable to the name of the Kubernetes node
• implement GetCapacity

Usage of --strict-topology and --immediate-topology=false is recommended because it makes the CreateVolume invocations simpler. Topology information is always derived exclusively from the information returned by the CSI driver that runs on the same node, without combining that with information stored for other nodes. This works as long as each node is in its own topology segment, i.e. usually with a single topology key and one unique value for each node.

Volume provisioning with late binding works as before, except that each external-provisioner instance checks the "selected node" annotation and only creates volumes if that node is the one it runs on. It also only deletes volumes on its own node.

Immediate binding is also supported, but not recommended. It is implemented by letting the external-provisioner instances assign a PVC to one of them: when they see a new PVC with immediate binding, they all attempt to set the "selected node" annotation with their own node name as value. Only one update request can succeed, all others get a "conflict" error and then know that some other instance was faster. To avoid the thundering herd problem, each instance waits for a random period before issuing an update request.

When CreateVolume call fails with ResourcesExhausted, the normal recovery mechanism is used, i.e. the external-provisioner instance removes the "selected node" annotation and the process repeats. But this triggers events for the PVC and delays volume creation, in particular when storage is exhausted on most nodes. Therefore external-provisioner checks with GetCapacity before attempting to own a PVC whether the currently available capacity is sufficient for the volume. When it isn't, the PVC is ignored and some other instance can own it.

The --node-deployment-base-delay parameter determines the initial wait period. It also sets the jitter, so in practice the initial wait period will be in the range from zero to the base delay. If the value is high, volumes with immediate binding get created more slowly. If it is low, then the risk of conflicts while setting the "selected node" annotation increases and the apiserver load will be higher.

There is an exponential backoff per PVC which is used for unexpected problems. Normally, an owner for a PVC is chosen during the first attempt, so most PVCs will use the base delays. A maximum can be set with --node-deployment-max-delay anyway, to avoid very long delays when something went wrong repeatedly.

During scale testing with 100 external-provisioner instances, a base delay of 20 seconds worked well. When provisioning 3000 volumes, there were only 500 conflicts which the apiserver handled without getting overwhelmed. The average provisioning rate of around 40 volumes/second was the same as with a delay of 10 seconds. The worst-case latency per volume was probably higher, but that wasn't measured.

Note that the QPS settings of kube-controller-manager and external-provisioner have to be increased at the moment (Kubernetes 1.19) to provision volumes faster than around 4 volumes/second. Those settings will eventually get replaced with flow control in the API server itself.

Beware that if no node has sufficient storage available, then also no CreateVolume call is attempted and thus no events are generated for the PVC, i.e. some other means of tracking remaining storage capacity must be used to detect when the cluster runs out of storage.

Because PVCs with immediate binding get distributed randomly among nodes, they get spread evenly. If that is not desirable, then it is possible to disable support for immediate binding in distributed provisioning with --node-deployment-immediate-binding=false and instead implement a custom policy in a separate controller which sets the "selected node" annotation to trigger local provisioning on the desired node.

Deleting local volumes after a node failure or removal

When a node with local volumes gets removed from a cluster before deleting those volumes, the PV and PVC objects may still exist. It may be possible to remove the PVC normally if the volume was not in use by any pod on the node, but normal deletion of the volume and thus deletion of the PV is not possible anymore because the CSI driver instance on the node is not available or reachable anymore and therefore Kubernetes cannot be sure that it is okay to remove the PV.

When an administrator is sure that the node is never going to come back, then the local volumes can be removed manually:

• force-delete objects: kubectl delete pv <pv> --wait=false --grace-period=0 --force
• remove all finalizers: kubectl patch pv <pv> -p '{"metadata":{"finalizers":null}}'

It may also be necessary to scrub disks before reusing them because the CSI driver had no chance to do that.

If there still was a PVC which was bound to that PV, it then will be moved to phase "Lost". It has to be deleted and re-created if still needed because no new volume will be created for it. Editing the PVC to revert it to phase "Unbound" is not allowed by the Kubernetes API server.

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.