10 Cloud Storage Solutions for Kubernetes Applications
10 cloud storage options for Kubernetes: EBS, EFS, FSx, Mountpoint for S3, Persistent Disk, Filestore, Azure Disk, Azure Files, Rook-Ceph, Portworx.
The main cloud storage solutions for Kubernetes are the provider CSI drivers for block disks (Amazon EBS, Google Persistent Disk and Hyperdisk, Azure Disk), shared file storage (Amazon EFS, Google Filestore, Azure Files), high-performance file systems (Amazon FSx), object storage mounts (Mountpoint for Amazon S3), and software-defined storage that runs across clouds, such as Rook-Ceph and Portworx.
This guide compares those ten options by access mode, performance profile and best use, with a StorageClass example for each cloud. Driver names, provisioner strings and add-on names were checked against AWS, Google Cloud, Microsoft and project documentation in November 2026. For the underlying concepts (volumes, PVs, PVCs and StorageClasses), read Kubernetes storage options explained first.
Key takeaways
- Pick storage by access pattern first. A single-writer database wants block storage (ReadWriteOnce); many pods sharing files want a network file system (ReadWriteMany); large read-mostly datasets can sit in object storage.
- All modern cloud storage for Kubernetes is delivered through the Container Storage Interface (CSI). The old in-tree cloud volume plugins have been migrated to CSI drivers.
- Block volumes are zonal. Use
volumeBindingMode: WaitForFirstConsumerso the disk is created in the same zone as the pod. - Managed file services (EFS, Filestore, Azure Files) are the simplest way to get ReadWriteMany, at the cost of higher latency than local block disks.
- Rook-Ceph and Portworx give one storage layer across clouds and on-premises, but you take on more operational work or licence cost.
- Storage is not backup. Plan snapshots and off-cluster backups separately.
How cloud storage connects to Kubernetes
A CSI driver has a controller component that talks to the cloud API (create disk, attach, snapshot) and a node component that mounts the volume into the pod. You expose a driver to developers through a StorageClass, and developers request storage with a PersistentVolumeClaim. When the claim is bound, the driver creates the disk or file share automatically. That is dynamic provisioning.
Three questions decide which solution fits a workload:
- Access mode: ReadWriteOnce (one node), ReadWriteOncePod (one pod), ReadOnlyMany or ReadWriteMany (many nodes).
- Performance: latency, IOPS and throughput needs, and whether you can tune them separately.
- Scope: does the data need to survive a zone failure, or move between clouds?
Our beginner guide to Kubernetes storage classes explains the StorageClass fields used below.
The ten solutions at a glance
| # | Solution | Type | Access modes | CSI provisioner | Best for |
|---|---|---|---|---|---|
| 1 | Amazon EBS | Block | RWO, RWOP | ebs.csi.aws.com | Databases on EKS |
| 2 | Amazon EFS | NFS file | RWX | efs.csi.aws.com | Shared content, CMS uploads |
| 3 | Amazon FSx | Managed file systems | RWX | FSx for Lustre CSI, NetApp Trident | HPC, ML training, NetApp features |
| 4 | Mountpoint for Amazon S3 | Object (mounted) | RWX for reads, limited writes | s3.csi.aws.com | Large read-heavy datasets |
| 5 | Google Persistent Disk and Hyperdisk | Block | RWO, RWOP, ROX | pd.csi.storage.gke.io | Databases on GKE |
| 6 | Google Filestore | NFS file | RWX | filestore.csi.storage.gke.io | Shared files on GKE |
| 7 | Azure Disk | Block | RWO, RWOP | disk.csi.azure.com | Databases on AKS |
| 8 | Azure Files | SMB or NFS file | RWX | file.csi.azure.com | Shared files, Windows pods |
| 9 | Rook-Ceph | Software-defined (block, file, object) | RWO, RWX | Ceph CSI | Self-managed, multi-cloud, on-prem |
| 10 | Portworx | Commercial software-defined | RWO, RWX | Portworx CSI | Enterprise data services across clouds |
Amazon Web Services
1. Amazon EBS CSI driver
Amazon Elastic Block Store volumes are network block disks attached to one EC2 instance at a time, in one Availability Zone. On EKS, install the driver as the aws-ebs-csi-driver managed add-on with an IAM role. The gp3 volume type lets you set IOPS and throughput separately from size; io2 is for the highest IOPS and durability needs.
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: ebs-gp3
provisioner: ebs.csi.aws.com
parameters:
type: gp3
iops: "6000"
throughput: "250"
encrypted: "true"
volumeBindingMode: WaitForFirstConsumer
allowVolumeExpansion: true
reclaimPolicy: Retain
Use it for: PostgreSQL, MySQL, MongoDB, Kafka and other workloads with one writer per volume. Watch for: a pod cannot move to a node in another zone and keep its EBS volume; spread replicas across zones at the application level instead. For a deeper EBS vs EFS comparison, see EBS vs EFS storage services.
2. Amazon EFS CSI driver
Amazon Elastic File System is a managed NFS file system that many pods across zones can mount at once. The driver can provision a new access point per PVC on an existing file system, which gives each claim its own directory and POSIX identity.
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: efs-shared
provisioner: efs.csi.aws.com
parameters:
provisioningMode: efs-ap
fileSystemId: fs-0123456789abcdef0
directoryPerms: "700"
Use it for: shared uploads, CMS media, shared configuration and ML artefacts read by many pods. Watch for: higher per-operation latency than EBS, so avoid it for database data files. Storage size is elastic, and the PVC size value is not enforced.
3. Amazon FSx (Lustre and NetApp ONTAP)
Amazon FSx offers managed file systems for specialised needs. FSx for Lustre has its own CSI driver and delivers high throughput for HPC and ML training, with optional linking to an S3 bucket. FSx for NetApp ONTAP is used through NetApp's Trident CSI driver, which brings NetApp features such as snapshots, clones and replication, and supports both NFS and iSCSI.
Use it for: training jobs reading large datasets quickly (Lustre) or teams that already rely on NetApp data management (ONTAP). Watch for: cost and sizing; these are higher-end services, so match them to workloads that need them.
4. Mountpoint for Amazon S3 CSI driver
The Mountpoint for Amazon S3 CSI driver mounts an S3 bucket as a file system inside pods, using AWS's Mountpoint client. It is available as the aws-mountpoint-s3-csi-driver EKS add-on. It is optimised for high-throughput reads of large objects, and supports sequential writes of new files but not in-place edits.
Use it for: ML datasets, media archives and analytics inputs that many pods read. Watch for: it uses static provisioning (you point a PersistentVolume at an existing bucket), it does not behave like a POSIX disk, and AWS documents limits such as no support on Fargate. Applications that need random writes should use EBS or EFS. Our article on Amazon S3 covers the object store itself.
Google Cloud
5. Compute Engine Persistent Disk and Hyperdisk CSI driver
On GKE, the Compute Engine persistent disk CSI driver provisions both Persistent Disk types (pd-balanced, pd-ssd and others) and the newer Hyperdisk family, which lets you set performance separately from capacity on supported machine types. GKE enables the driver by default on new clusters. Regional persistent disks replicate data synchronously across two zones in a region.
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: pd-balanced-regional
provisioner: pd.csi.storage.gke.io
parameters:
type: pd-balanced
replication-type: regional-pd
volumeBindingMode: WaitForFirstConsumer
allowVolumeExpansion: true
Use it for: stateful sets and databases on GKE; regional disks when a single-replica workload must survive a zone outage. Watch for: Hyperdisk types are supported only on certain machine series, so check compatibility before choosing.
6. Filestore CSI driver
Filestore is Google Cloud's managed NFS service. The Filestore CSI driver, enabled as a GKE add-on, provisions Filestore instances or shares for ReadWriteMany volumes. Tiers range from basic HDD to zonal and regional high-performance options.
Use it for: shared files across pods and zones, legacy apps that expect an NFS mount, and content repositories. Watch for: minimum instance sizes on some tiers make small volumes expensive; check whether multishare instances fit your use.
Microsoft Azure
7. Azure Disk CSI driver
Azure managed disks attach to one node and are the standard choice for single-writer workloads on AKS. AKS installs the Azure Disk CSI driver by default and ships built-in StorageClasses such as managed-csi and managed-csi-premium. Premium SSD v2 and Ultra Disk let you tune IOPS and throughput independently.
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: azure-premium-zrs
provisioner: disk.csi.azure.com
parameters:
skuName: Premium_ZRS
volumeBindingMode: WaitForFirstConsumer
allowVolumeExpansion: true
reclaimPolicy: Retain
Use it for: databases and stateful services on AKS. Zone-redundant storage (ZRS) SKUs let a disk attach to nodes in different zones of a region. Watch for: each VM size has a maximum number of data disks, which caps how many volumes a node can host.
8. Azure Files CSI driver
Azure Files provides managed file shares over SMB or NFS. The Azure Files CSI driver, also installed by default on AKS, supports ReadWriteMany and works with both Linux and Windows nodes. Built-in classes include azurefile-csi and azurefile-csi-premium.
Use it for: shared application data, Windows containers that need SMB, and lift-and-shift apps expecting a network share. Watch for: choose NFS shares on the premium tier for Linux workloads that need POSIX permissions, and test performance for small-file workloads.
Software-defined storage across clouds
9. Rook-Ceph
Rook is a CNCF graduated project that runs Ceph inside Kubernetes as an operator. Ceph then provides block (RBD), file (CephFS) and S3-compatible object storage from the disks attached to your nodes, whether those are cloud volumes, local NVMe drives or bare-metal disks.
Use it for: one storage platform across clouds and data centres, on-prem clusters with no cloud storage, or when you need RWX block and object storage under your own control. Watch for: Ceph is powerful but complex. Plan for at least three storage nodes, monitoring and people who understand Ceph recovery.
10. Portworx
Portworx, from Pure Storage, is a commercial software-defined storage platform for Kubernetes. It pools storage across nodes and clouds and adds data services such as replication, snapshots, encryption, disaster recovery and backup (via PX-Backup) under one control plane.
Use it for: enterprises running stateful apps on several clouds or on-prem that want vendor support and one set of data features everywhere. Watch for: licence cost, and lock-in to its data management model. Pricing is not public in a simple form, so request a quote.
Lighter open-source options
Two open-source projects are popular for smaller clusters, edge sites and labs where Rook-Ceph would be too heavy. Longhorn is a CNCF project, originally from Rancher, that replicates block volumes across nodes and includes snapshots and backups to S3 or NFS. OpenEBS, a CNCF Sandbox project, offers local persistent volumes and replicated storage engines. Both are easier to start with than Ceph but deserve the same testing for failure and recovery.
How to choose
| Workload | AWS | Google Cloud | Azure | Multi-cloud / on-prem |
|---|---|---|---|---|
| Relational or NoSQL database | EBS gp3 or io2 | PD balanced/SSD or Hyperdisk | Azure Disk Premium SSD v2 | Rook-Ceph RBD, Portworx |
| Shared files across pods | EFS | Filestore | Azure Files | CephFS, Portworx shared volumes |
| ML training data | FSx for Lustre, Mountpoint for S3 | Filestore high tiers, Cloud Storage FUSE | Azure Files premium, Blob CSI | CephFS or Ceph object |
| Survive a zone outage with one replica | EFS (multi-AZ) | Regional PD | ZRS disks | Replicated SDS |
| Windows containers | FSx for Windows File Server (SMB) | Filestore (NFS) or PD | Azure Files SMB | Depends on vendor |
Google Cloud also has a Cloud Storage FUSE CSI driver and Azure has a Blob storage CSI driver, which play a similar role to Mountpoint for S3. They are worth a look for read-heavy datasets.
Practices that apply to every option
- Use
WaitForFirstConsumerfor zonal block storage so disks are created where pods are scheduled. - Set
reclaimPolicy: Retainfor production data so deleting a PVC by mistake does not delete the disk. - Enable
allowVolumeExpansionso volumes can grow without migration. - Encrypt at rest with provider-managed or customer-managed keys.
- Snapshot with CSI VolumeSnapshots and copy backups off the cluster. See Kubernetes backup tools and Kubernetes backup strategies.
- Monitor capacity and latency: kubelet volume stats show used and available bytes; alert before volumes fill.
- Test failure: drain a node, delete a pod and simulate a zone loss to confirm the data comes back as expected.
apiVersion: snapshot.storage.k8s.io/v1
kind: VolumeSnapshot
metadata:
name: orders-db-snap
namespace: shop
spec:
volumeSnapshotClassName: csi-snapclass
source:
persistentVolumeClaimName: data-orders-db-0
Learn Kubernetes storage hands-on
Storage questions appear in CKA labs and real production incidents alike. Our CKA training covers PVs, PVCs, StorageClasses and troubleshooting, and the AWS Solutions Architect Associate training covers EBS, EFS, FSx and S3 design. Both run at the Kharadi, Pune classroom and live online.
Frequently asked questions
What is the best storage for databases on Kubernetes?
Use the cloud provider's block storage through its CSI driver: Amazon EBS on EKS, Persistent Disk or Hyperdisk on GKE, and Azure Disk on AKS. Block storage gives the low latency and single-writer semantics databases expect.
Which Kubernetes storage supports ReadWriteMany?
Managed file services do: Amazon EFS, Google Filestore and Azure Files. CephFS through Rook and Portworx shared volumes also support ReadWriteMany across clouds and on-premises.
What is a CSI driver?
A Container Storage Interface driver is a plugin that lets Kubernetes create, attach, mount, resize and snapshot volumes on a storage system. Each cloud provides CSI drivers for its disk and file services.
Can I mount an S3 bucket in a Kubernetes pod?
Yes, with the Mountpoint for Amazon S3 CSI driver. It suits high-throughput reads and sequential writes of new files, but it does not support in-place edits like a normal disk, and it uses static provisioning.
Why should I use WaitForFirstConsumer?
Block disks live in one zone. WaitForFirstConsumer delays disk creation until a pod is scheduled, so the disk is created in the pod's zone and the pod is not stuck waiting for a volume it cannot attach.
Is Rook-Ceph better than cloud storage?
Not by default. Cloud CSI drivers are simpler and fully managed. Rook-Ceph helps when you need one storage layer across clouds or on-premises, and you have people to run Ceph.
Do cloud volumes replace backups?
No. Replication protects against hardware failure, not deletion, corruption or ransomware. Take CSI snapshots and copy backups to separate storage or another region with a tool such as Velero.
References
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Angry
0
Sad
0
Wow
0