Reference Documentation

Design docs, concept definitions, and references for APIs and CLIs.

Edit This Page

Persistent Volumes

This document describes the current state of PersistentVolumes in Kubernetes. Familiarity with volumes is suggested.


Managing storage is a distinct problem from managing compute. The PersistentVolume subsystem provides an API for users and administrators that abstracts details of how storage is provided from how it is consumed. To do this we introduce two new API resources: PersistentVolume and PersistentVolumeClaim.

A PersistentVolume (PV) is a piece of networked storage in the cluster that has been provisioned by an administrator. It is a resource in the cluster just like a node is a cluster resource. PVs are volume plugins like Volumes, but have a lifecycle independent of any individual pod that uses the PV. This API object captures the details of the implementation of the storage, be that NFS, iSCSI, or a cloud-provider-specific storage system.

A PersistentVolumeClaim (PVC) is a request for storage by a user. It is similar to a pod. Pods consume node resources and PVCs consume PV resources. Pods can request specific levels of resources (CPU and Memory). Claims can request specific size and access modes (e.g, can be mounted once read/write or many times read-only).

While PersistentVolumeClaims allow a user to consume abstract storage resources, it is common that users need PersistentVolumes with varying properties, such as performance, for different problems. Cluster administrators need to be able to offer a variety of PersistentVolumes that differ in more ways than just size and access modes, without exposing users to the details of how those volumes are implemented. For these needs there is the StorageClass resource.

A StorageClass provides a way for administrators to describe the “classes” of storage they offer. Different classes might map to quality-of-service levels, or to backup policies, or to arbitrary policies determined by the cluster administrators. Kubernetes itself is unopinionated about what classes represent. This concept is sometimes called “profiles” in other storage systems.

Please see the detailed walkthrough with working examples.

Lifecycle of a volume and claim

PVs are resources in the cluster. PVCs are requests for those resources and also act as claim checks to the resource. The interaction between PVs and PVCs follows this lifecycle:


There are two ways PVs may be provisioned: statically or dynamically.


A cluster administrator creates a number of PVs. They carry the details of the real storage which is available for use by cluster users. They exist in the Kubernetes API and are available for consumption.


When none of the static PVs the administrator created matches a user’s PersistentVolumeClaim, the cluster may try to dynamically provision a volume specially for the PVC. This provisioning is based on StorageClasses: the PVC must request a class and the administrator must have created and configured that class in order for dynamic provisioning to occur. Claims that request the class "" effectively disable dynamic provisioning for themselves.


A user creates, or has already created in the case of dynamic provisioning, a PersistentVolumeClaim with a specific amount of storage requested and with certain access modes. A control loop in the master watches for new PVCs, finds a matching PV (if possible), and binds them together. If a PV was dynamically provisioned for a new PVC, the loop will always bind that PV to the PVC. Otherwise, the user will always get at least what they asked for, but the volume may be in excess of what was requested. Once bound, PersistentVolumeClaim binds are exclusive, regardless of the mode used to bind them.

Claims will remain unbound indefinitely if a matching volume does not exist. Claims will be bound as matching volumes become available. For example, a cluster provisioned with many 50Gi PVs would not match a PVC requesting 100Gi. The PVC can be bound when a 100Gi PV is added to the cluster.


Pods use claims as volumes. The cluster inspects the claim to find the bound volume and mounts that volume for a pod. For volumes which support multiple access modes, the user specifies which mode desired when using their claim as a volume in a pod.

Once a user has a claim and that claim is bound, the bound PV belongs to the user for as long as they need it. Users schedule Pods and access their claimed PVs by including a persistentVolumeClaim in their Pod’s volumes block. See below for syntax details.


When a user is done with their volume, they can delete the PVC objects from the API which allows reclamation of the resource. The volume is considered “released” when the claim is deleted, but it is not yet available for another claim. The previous claimant’s data remains on the volume which must be handled according to policy.


The reclaim policy for a PersistentVolume tells the cluster what to do with the volume after it has been released of its claim. Currently, volumes can either be Retained, Recycled or Deleted. Retention allows for manual reclamation of the resource. For those volume plugins that support it, deletion removes both the PersistentVolume object from Kubernetes as well as deletes associated storage asset in external infrastructure such as AWS EBS, GCE PD or Cinder volume. Volumes that were dynamically provisioned are always deleted. If supported by appropriate volume plugin, recycling performs a basic scrub (rm -rf /thevolume/*) on the volume and makes it available again for a new claim.

Types of Persistent Volumes

PersistentVolume types are implemented as plugins. Kubernetes currently supports the following plugins:

Persistent Volumes

Each PV contains a spec and status, which is the specification and status of the volume.

  apiVersion: v1
  kind: PersistentVolume
    name: pv0003
    annotations: "slow"
      storage: 5Gi
      - ReadWriteOnce
    persistentVolumeReclaimPolicy: Recycle
      path: /tmp


Generally, a PV will have a specific storage capacity. This is set using the PV’s capacity attribute. See the Kubernetes Resource Model to understand the units expected by capacity.

Currently, storage size is the only resource that can be set or requested. Future attributes may include IOPS, throughput, etc.

Access Modes

A PersistentVolume can be mounted on a host in any way supported by the resource provider. As shown in the table below, providers will have different capabilities and each PV’s access modes are set to the specific modes supported by that particular volume. For example, NFS can support multiple read/write clients, but a specific NFS PV might be exported on the server as read-only. Each PV gets its own set of access modes describing that specific PV’s capabilities.

The access modes are:

In the CLI, the access modes are abbreviated to:

Important! A volume can only be mounted using one access mode at a time, even if it supports many. For example, a GCEPersistentDisk can be mounted as ReadWriteOnce by a single node or ReadOnlyMany by many nodes, but not at the same time.

Volume Plugin ReadWriteOnce ReadOnlyMany ReadWriteMany
AWSElasticBlockStore x - -
AzureFile x x x
CephFS x x x
Cinder x - -
FC x x -
FlexVolume x x -
GCEPersistentDisk x x -
Glusterfs x x x
HostPath x - -
iSCSI x x -
NFS x x x
RBD x x -
VsphereVolume x - -


A PV can have a class, which is specified by setting the annotation to the name of a StorageClass. A PV of a particular class can only be bound to PVCs requesting that class. A PV with no annotation or its class annotation set to "" has no class and can only be bound to PVCs that request no particular class.

In the future after beta, the annotation will become an attribute.

Recycling Policy

Current recycling policies are:

Currently, only NFS and HostPath support recycling. AWS EBS, GCE PD and Cinder volumes support deletion.


A volume will be in one of the following phases:

The CLI will show the name of the PVC bound to the PV.


Each PVC contains a spec and status, which is the specification and status of the claim.

kind: PersistentVolumeClaim
apiVersion: v1
  name: myclaim
  annotations: "slow"
    - ReadWriteOnce
      storage: 8Gi
      release: "stable"
      - {key: environment, operator: In, values: [dev]}

Access Modes

Claims use the same conventions as volumes when requesting storage with specific access modes.


Claims, like pods, can request specific quantities of a resource. In this case, the request is for storage. The same resource model applies to both volumes and claims.


Claims can specify a label selector to further filter the set of volumes. Only the volumes whose labels match the selector can be bound to the claim. The selector can consist of two fields:

All of the requirements, from both matchLabels and matchExpressions are ANDed together – they must all be satisfied in order to match.


A claim can request a particular class by specifying the name of a StorageClassusing the annotation Only PVs of the requested class, ones with the same annotation as the PVC, can be bound to the PVC.

PVCs don’t necessarily have to request a class. A PVC with its annotation set equal to "" is always interpreted to be requesting a PV with no class, so it can only be bound to PVs with no class (no annotation or one set equal to ""). A PVC with no annotation is not quite the same and is treated differently by the cluster depending on whether the DefaultStorageClass admission plugin is turned on.

When a PVC specifies a selector in addition to requesting a StorageClass, the requirements are ANDed together: only a PV of the requested class and with the requested labels may be bound to the PVC. Note that currently, a PVC with a non-empty selector can’t have a PV dynamically provisioned for it.

In the future after beta, the annotation will become an attribute.

Claims As Volumes

Pods access storage by using the claim as a volume. Claims must exist in the same namespace as the pod using the claim. The cluster finds the claim in the pod’s namespace and uses it to get the PersistentVolume backing the claim. The volume is then mounted to the host and into the pod.

kind: Pod
apiVersion: v1
  name: mypod
    - name: myfrontend
      image: dockerfile/nginx
      - mountPath: "/var/www/html"
        name: mypd
    - name: mypd
        claimName: myclaim

A Note on Namespaces

PersistentVolumes binds are exclusive, and since PersistentVolumeClaims are namespaced objects, mounting claims with “Many” modes (ROX, RWX) is only possible within one namespace.


Each StorageClass contains the fields provisioner and parameters, which are used when a PersistentVolume belonging to the class needs to be dynamically provisioned.

The name of a StorageClass object is significant, and is how users can request a particular class. Administrators set the name and other parameters of a class when first creating StorageClass objects, and the objects cannot be updated once they are created.

Administrators can specify a default StorageClass just for PVCs that don’t request any particular class to bind to: see the PersistentVolumeClaim section for details.

kind: StorageClass
  name: standard
  type: gp2


Storage classes have a provisioner that determines what volume plugin is used for provisioning PVs. This field must be specified. During beta, the available provisioner types are and


Storage classes have parameters that describe volumes belonging to the storage class. Different parameters may be accepted depending on the provisioner. For example, the value io1, for the parameter type, and the parameter iopsPerGB are specific to EBS. When a parameter is omitted, some default is used.


kind: StorageClass
  name: slow
  type: io1
  zone: us-east-1d
  iopsPerGB: "10"


kind: StorageClass
  name: slow
  type: pd-standard
  zone: us-central1-a


kind: StorageClass
  name: slow
  endpoint: "glusterfs-cluster"
  resturl: ""
  restauthenabled: "true"
  restuser: "admin"
  restuserkey: "password"

OpenStack Cinder

kind: StorageClass
  name: gold
  type: fast
  availability: nova

Writing Portable Configuration

If you’re writing configuration templates or examples that run on a wide range of clusters and need persistent storage, we recommend that you use the following pattern:


Create Issue Edit This Page