Does ceph-csi support volume level QoS, not just StorageClass level QoS?

[microyahoo]

the question as the title described, thanks!

[Ganesh-403]

@microyahoo From what I could find, ceph-csi has historically exposed QoS through the StorageClass, so every volume provisioned from that class inherits the same settings.

However, newer versions have started adding support for per-volume QoS using Kubernetes' VolumeAttributesClass API rather than putting all QoS parameters in the StorageClass.

Specifically:

• RBD support was added in ceph-csi v3.17, allowing QoS to be modified per volume via ControllerModifyVolume.
• This currently requires the nbd mounter. :contentReference[oaicite:0]{index=0}

The workflow looks roughly like this:

apiVersion: storage.k8s.io/v1
kind: VolumeAttributesClass
metadata:
  name: high-iops
driverName: rbd.csi.ceph.com
parameters:
  rwIosPerSecond: "10000"
  rwMbytesPerSecond: "500"

Then reference it from the PVC:

spec:
  storageClassName: ceph-rbd
  volumeAttributesClassName: high-iops

This allows different PVCs using the same StorageClass to have different QoS policies. IBM's Ceph CSI documentation describes this model for CSI drivers using VolumeAttributesClass, and ceph-csi's v3.17 release notes specifically mention RBD VolumeAttributesClass QoS support. :contentReference[oaicite:1]{index=1}

A couple of clarifying questions:

• Which ceph-csi version are you running?
• Are you asking about RBD, CephFS, or NVMe-oF volumes?

The answer depends on the driver:

• RBD: Per-volume QoS is supported in newer releases via VolumeAttributesClass (with the current limitations noted above). :contentReference[oaicite:2]{index=2}
• Older ceph-csi releases: QoS is effectively at the StorageClass level.
• CephFS/NVMe-oF: Support differs by driver and implementation, so the available QoS mechanisms are not identical. :contentReference[oaicite:3]{index=3}

If this solves your problem, feel free to mark it as the accepted answer so others can find it easily.