Kubernetes Persistent Volume Lifecycle: Create, Bind, Bill

Introduction

Pods come and go. Databases, AI pipelines, and business apps can't. They need storage that survives Pod replacement and rescheduling, and that's where PersistentVolumes come in.

The lifecycle has four beats. Kubernetes provisions storage, binds it to a claim, mounts it into a workload, and then the cloud provider starts billing for it. That last step is the one teams forget.

Lucidity's internal analysis points to over 50% storage underutilization across cloud block storage, with an estimated $21.5 billion in wasted spend by 2028.

This guide covers static versus dynamic provisioning, binding problems, reclaim behavior, and how to keep unused volumes off your bill.

Key Takeaways

  • A PV is the storage resource; a PVC is the workload's request for it.
  • StorageClasses and CSI drivers handle dynamic provisioning in most production clusters.
  • A Bound PVC still generates capacity, performance, snapshot, or replication charges.
  • Deleting a PVC doesn't always delete the cloud volume; the reclaim policy decides.

Kubernetes Storage Objects: Volume, PV, PVC, and StorageClass

Ephemeral volumes follow a Pod's lifecycle. A PersistentVolume exists independently of any single Pod.

  • PersistentVolume (PV): cluster-scoped storage with capacity, access modes, storage class, reclaim policy, and backend details.
  • PersistentVolumeClaim (PVC): a namespaced request for capacity, access mode, and optionally a StorageClass. Pods reference the PVC, never the PV directly.
  • StorageClass: defines the provisioner (usually a CSI driver) and parameters for dynamic provisioning, plus reclaim policy, expansion, and binding mode.

The lifecycle at a glance: provision or discover storage, bind PV to PVC, mount the claim into a Pod, monitor usage and cost, then reclaim or delete.

Five-stage Kubernetes PersistentVolume lifecycle from provisioning to reclamation

A PV moves through Available, Bound, Released, and Failed phases. Volume mode is Filesystem by default; Block requests a raw device.

Create: Provision Persistent Volumes Statically or Dynamically

Static vs. dynamic

With static provisioning, an administrator creates PVs by hand. With dynamic provisioning, a PVC plus a StorageClass makes Kubernetes request a new volume from the storage backend when no suitable static PV exists. Production clusters mostly use dynamic.

Fields that matter on the PVC

  • Requested capacity
  • Access mode
  • storageClassName
  • Volume mode
  • Any performance needs your CSI driver supports

If you omit storageClassName, the default StorageClass applies. With multiple defaults, Kubernetes picks the most recently created.

StorageClass controls

A production StorageClass sets:

  • provisioner
  • parameters
  • reclaimPolicy (defaults to Delete if omitted)
  • allowVolumeExpansion (grow only, never shrink)
  • volumeBindingMode

Defaults vary by platform. AKS ships managed-csi (Standard SSD LRS) and managed-csi-premium. GKE installs CSI-backed standard-rwo and premium-rwo. EKS Auto Mode doesn't create a StorageClass at all, so don't assume gp3 is a universal EKS default.

Topology and WaitForFirstConsumer

Immediate binding provisions the volume at PVC creation, possibly in a zone where your Pod can't run. WaitForFirstConsumer delays provisioning until a Pod uses the claim, so the scheduler can account for zone and node constraints. A Pending claim under this mode can be expected. Setting nodeName bypasses the scheduler and can leave it Pending forever.

Immediate versus WaitForFirstConsumer volume binding behavior comparison

Example

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: orders-db-data
  namespace: prod
spec:
  accessModes: ["ReadWriteOnce"]
  storageClassName: managed-csi
  resources:
    requests:
      storage: 100Gi

Avoid hostPath outside local learning setups. It exposes a node path, doesn't limit disk use between Pods, and Kubernetes suggests local volumes instead.

Bind and Use: How a PVC Reaches a Pod

How binding works

Kubernetes matches capacity, access modes, StorageClass, selectors, volume mode, and topology. Binding is exclusive and one-to-one. Key state transitions:

  • PV: Available to Bound, then Released after its claim is deleted
  • PVC: Pending to Bound

A claim stays Pending until a compatible volume is found or provisioned. A PVC naming a PV already claimed by another PVC will also sit Pending.

Mounting into a workload

The Pod (or Deployment, StatefulSet) lists the PVC under volumes and mounts it with volumeMounts. The Pod and PVC must share a namespace.

Access modes

  • ReadWriteOnce: read-write from one node. Multiple Pods on that node can share it.
  • ReadOnlyMany: read-only from many nodes.
  • ReadWriteMany: read-write from many nodes; needs a backend that supports it.
  • ReadWriteOncePod: one Pod only. CSI-only, stable since Kubernetes v1.29.

Cloud block storage has its own limits. EBS Multi-Attach is restricted to io1/io2 volumes in one Availability Zone, and AWS warns that ordinary XFS or EXT4 isn't built for concurrent multi-server writes. A read-write Google Persistent Disk can't attach to multiple nodes at once.

Once the claim is Bound and mounted, walk this sequence to confirm the volume path.

Verification sequence

  1. kubectl get pvc -n <ns> to check status
  2. kubectl describe pvc <name> to read events
  3. kubectl get pv to confirm the volume and phase
  4. kubectl describe pod <pod> for mount errors
  5. kubectl get events --sort-by=.lastTimestamp for provisioner messages

Five-step Kubernetes PVC and Pod storage verification sequence

Common causes of failure:

  • StorageClass mismatch
  • Unsupported access mode
  • Requested size too large
  • Wrong namespace
  • Zone conflicts
  • Mount permission problems

Bill: How Persistent Volumes Become Cloud Costs

Kubernetes doesn't charge you. The cloud provider bills for the underlying disk. Requested capacity is what's billed on most disk types, not what your application actually uses.

Cost drivers by provider

  • AWS EBS: provisioned capacity (including when unattached), plus gp3 IOPS and throughput above baseline (3,000 IOPS, 125 MiB/s), plus snapshots. See EBS pricing.
  • Azure Managed Disks: billed by provisioned tier, so a 200 GiB request maps to the 256 GiB E15 tier. Unattached disks keep billing. Premium SSD bursting can add fees.
  • Google Persistent Disk/Hyperdisk: PD is billed on provisioned space with I/O included; Hyperdisk adds provisioned IOPS and throughput. Standard snapshots in Iowa run $0.05/GB-month, per Google's pricing.

Rates vary by region and date, so confirm current pricing for your region.

Lifecycle state vs. money

State Billing impact
Pending PVC Often no disk yet (expected under WaitForFirstConsumer)
Bound PVC Active provisioned capacity
Released / unattached Charges continue until the cloud disk is deleted
Zero-I/O volume Still billed for full provisioned size

Kubernetes PersistentVolume lifecycle states and cloud billing impact

Cost control checklist

  • Request realistic capacity
  • Match performance tier to the workload
  • Enable expansion only where safe
  • Monitor utilization and I/O
  • Audit unattached volumes regularly
  • Use ownership labels and namespace quotas
  • Reconcile Kubernetes inventory against billing exports

Kubernetes tools show claims, not cloud invoices. Lucidity fills that gap across AWS, Azure, and Google Cloud with real-time visibility into underlying block storage and autonomous optimization that can cut costs by up to 70%. It complements PV, PVC, StorageClass, and CSI controls rather than replacing them.

Reclaim and Control Storage Waste

Retain vs. Delete

  • Retain: After the PVC is deleted, the PV goes to Released. Data stays for manual recovery, but the PV isn't reusable until an admin cleans it up.
  • Delete: Removes the PV and, if the provisioner supports it, the backing storage and data.

Confirm actual behavior for your StorageClass before relying on either. Recycle is deprecated.

Safe cleanup workflow

  1. Identify the owning application so the right team can approve deletion
  2. Confirm backups or snapshots exist before any destructive step
  3. Check that no Pod still references the PVC
  4. Review the reclaim policy on the PV and StorageClass
  5. Delete only after explicit approval
  6. Verify both the Kubernetes objects and the cloud disk are gone

Six-step safe Kubernetes PersistentVolume cleanup workflow

Preventing silent waste

  • Use ResourceQuota to cap requests.storage and PVC counts per namespace, including per StorageClass; violations are rejected
  • Use RBAC to block PVC deletion by omitting the delete verb where needed
  • Alert on Released and unattached volumes.
  • Label volumes with lifecycle and owner, and document retention periods.
  • Use separate policies for production, staging, and temporary environments.

Keep these limits in mind:

  • Finalizers delay deletion while a Pod uses a PVC, but they don't stop billing
  • Quotas govern requests, not the provider's billed capacity
  • Snapshots are point-in-time copies that depend on CSI driver support; they aren't a complete backup or disaster-recovery plan on their own

Frequently Asked Questions

What is the difference between a Kubernetes volume and a persistent volume?

A regular volume is tied to a Pod, and its lifecycle depends on the volume type. A PersistentVolume is a separate cluster resource built to outlive Pods, normally consumed through a PVC.

What is the difference between a PersistentVolume and a PersistentVolumeClaim?

The PV is the supply of storage. The PVC is a namespaced request for capacity, access mode, and possibly a StorageClass.

How does a PVC get bound to a PV?

Kubernetes matches capacity, access mode, StorageClass, selectors, volume mode, and topology. A claim with no match stays Pending.

What happens to a PV when its PVC is deleted?

With Retain, the PV becomes Released and keeps its data. With Delete, the PV and often the underlying cloud disk are removed, data included.

How do Kubernetes Persistent Volumes affect cloud billing?

The underlying cloud disk is billed, not the Kubernetes object. Charges can include provisioned capacity, performance, snapshots, and replication.

Why is my PVC stuck in Pending?

Common causes are no matching PV, a wrong or missing StorageClass, an unsupported access mode, topology limits, quota, or a provisioner error. Run kubectl describe pvc and read the Events section. Under WaitForFirstConsumer, it may just be waiting for a Pod.