Storage

Storage classes

Class	Access mode	Description
hcloud-volumes	ReadWriteOnce	Block storage, the cluster default, supports volume expansion
nfs-rwx	ReadWriteMany	NFS, for when several pods need to write at the same time

hcloud-volumes is the default class: a PVC without storageClassName lands there.

The RWX class is named nfs-rwx

Not nfs. A PVC with storageClassName: nfs stays Pending.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: data
spec:
  accessModes:
    - ReadWriteOnce
  storageClassName: hcloud-volumes
  resources:
    requests:
      storage: 10Gi

Billing

On hcloud-volumes there is a billing minimum of 10 GiB per volume. Smaller PVCs still provision normally but are billed at 10 GiB, so five 2 GiB volumes cost the same as five 10 GiB ones. The floor comes from the block storage itself, not from us.

nfs-rwx has no minimum. It is billed at the size you request, so a 1 GiB shared volume is billed as 1 GiB.

For uploads, media and anything else that does not need a filesystem, object storage is usually cheaper than a volume and is not tied to a single namespace.

Autoresize

Volumes can grow automatically, controlled by three annotations on the PVC:

Annotation	Meaning
resize.topolvm.io/threshold	How little free space triggers an expansion
resize.topolvm.io/increase	How much to add per step
resize.topolvm.io/storage_limit	Upper bound for expansion

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: data
  annotations:
    resize.topolvm.io/threshold: "20%"
    resize.topolvm.io/increase: "10Gi"
    resize.topolvm.io/storage_limit: "100Gi"
spec:
  accessModes:
    - ReadWriteOnce
  storageClassName: hcloud-volumes
  resources:
    requests:
      storage: 10Gi

threshold is free space, not used space

"20%" means: expand once less than 20% is free, that is at 80% full. This is the opposite of what most people assume.

Set it to "80%" and the volume expands at 20% full. It then keeps growing step by step up to storage_limit without ever having been close to full, and you pay for that space the whole time.

Where it works

• On hcloud-volumes. Autoresize does not work on nfs-rwx.
• On CloudNativePG and MariaDB volumes as well.

Volumes only grow

A volume can be expanded but never shrunk. That applies to autoresize and to a manually raised resources.requests.storage alike. To get smaller, create a new PVC, copy the data, and delete the old one.

Platform backups

Your namespace is backed up daily, including the contents of persistent volumes. Daily backups are kept for 30 days. On a paid plan a weekly backup is also kept for 12 weeks and a monthly one for 6 months; the free tier gets the daily backups only. Backups are stored at a second location in the EU.

To restore from one, open a request through the portal and support will carry it out, for the whole namespace or for individual volumes. There is no self-service restore button: restores are handled for you rather than run by you.

This does not replace your own backups

A daily namespace backup is a floor, not a strategy. It will not give you a database consistent to the second, and it will not help with data your application corrupted rather than lost. Keep application-level backups of anything you cannot rebuild, see Databases.

What's next

• Databases, which use these classes
• Billing and quotas for how storage is charged