CSI Provisioning and Supported Operations

Nutanix

Audience
Public
Product
FlashArray
FlashArray > Purity//FA
Technology Integrations
Source Type
Documentation

Nutanix Volume Groups and CSI Behavior

When a Kubernetes PersistentVolumeClaim (PVC) is created, the Nutanix CSI driver (csi.nutanix.com) provisions a Nutanix Volume Group (VG) in the storageclass specified. If that storageclass is backed by FlashArray, the resulting Persistent Volume is ultimately backed by corresponding FlashArray Kublock volumes provisioned via AOS. The Nutanix CSI driver manages the full lifecycle of these Volume Groups, including provisioning, attachment to Kubernetes worker node VMs, expansion, snapshot creation, and restore operations.

Supported CSI and Storage Operations

The following operations are supported when using FlashArray-backed storage containers with Nutanix CSI:

CSI/Storage Operation Supported with FlashArray? Notes
PVC Create/Delete Yes Dynamically provisions/deletes Nutanix Volume Groups
PV Binding Yes Automatic binding via Kubernetes
Attach / Detach Yes Volume Group attached to / removed from worker node VM
Mount / Unmount Yes Filesystem mounted inside pod
Pod Restart with Volume Yes Remount occurs automatically
Pod Migration Across Nodes Yes Triggers detach/attach cycle
Worker Node Scale-Up Yes New nodes attach PVs normally
Worker Node Scale-Down Yes PVs detach before VM deletion
Worker Node Reboot Yes Volumes reattach after recovery
Online Volume Expansion Yes Filesystem resized automatically
CSI Snapshot Create Yes Supported
CSI Snapshot Restore Yes Creates new PVC from snapshot (no in-place overwrite)
Velero Backup / Restore Yes Supported

Volume Access Modes

FlashArray-backed CSI volumes are block devices. The following access modes apply:

PVC Access Mode Supported Notes
ReadWriteOnce (RWO) Yes Primary access mode
ReadOnlyMany (ROX) Yes Multiple nodes may mount the volume in read-only mode (no concurrent writes)
ReadWriteMany (RWX) No Block storage limitation; generally requires file-based storage

Hypervisor-Attached vs Guest-Attached Volumes

NKP supports two CSI attachment models:

  • Guest-Attached (default)

  • Hypervisor-Attached

Cluster creation flag example:

--csi-hypervisor-attached-volumes=false

Support & Recommendation Matrix

Mode Supported Recommended Use Case
Guest-Attached Yes Yes General workloads
Hypervisor-Attached Yes Conditional Latency-sensitive DB / benchmarking

Recommendation: Guest-Attached vs Hypervisor-Attached Volumes

For most NKP deployments using FlashArray-backed storage, guest-attached volumes are the recommended default.

Guest-attached volumes provide:

  • Cleaner scaling and node mutation workflows

  • Better alignment with Kubernetes automation

  • Faster and more predictable recovery during AHV host failure

Because attachment occurs at the NKP worker node VM level, failover aligns closely with Kubernetes pod lifecycles.

Hypervisor-Attached (Validate before Using)

Hypervisor-attached volumes are supported and may be appropriate for more performance/latency-sensitive workloads.

However:

  • Performance differences are workload-dependent.

  • Customers should perform internal benchmarking with their specific applications before selecting this mode.

  • During AHV host failure or maintenance operations, recovery can take longer because the volume must be reconciled at the hypervisor layer rather than the VM layer.

NKP Persistent Volume Filesystem Support

Nutanix CSI presents Persistent Volumes as block devices to the Kubernetes worker node. The filesystem (ext4 or xfs) is created and managed at the guest OS level.

Both ext4 and xfs are supported on the FlashArray.

Nutanix and FlashArray do not require or mandate a specific filesystem for CSI-provisioned volumes.

Persistent Volume Expansion

  • Online volume expansion is supported.

  • Filesystem resize occurs automatically inside the pod when the PVC is expanded.