# Ceph CSI Drivers (/docs/runtime/components/ceph-csi-drivers)



Component Category [#component-category]

Data and storage / storage driver

Component Description [#component-description]

Ceph CSI Drivers is the Container Storage Interface (CSI) driver set that lets Kubernetes dynamically provision, attach, and mount volumes backed by the `Rook-Ceph` cluster — both RBD block volumes and CephFS shared filesystem volumes. It is deployed as its own Helm release, separate from the Ceph cluster it serves.

This is distinct from `CSI S3`, which is a different CSI driver (Mountpoint for Amazon S3) used specifically to mount the Rook-Ceph RGW's S3-compatible object storage as a filesystem for model-cache workloads.

Why It Is Used [#why-it-is-used]

In BullSequana AI Runtime, Ceph CSI Drivers is what makes `StorageClass` resources backed by Rook-Ceph usable by ordinary Kubernetes workloads. Without it, pods could not claim `PersistentVolumeClaim`s against CephFS or RBD-backed storage classes. It is deployed and versioned separately from the Ceph cluster itself so the driver's lifecycle (upgrades, compatibility with the cluster kubelet) can be managed independently of cluster operations like OSD management.

Learn More [#learn-more]

* [Ceph CSI documentation](https://github.com/ceph/ceph-csi/blob/devel/docs/deploy-rbd.md)
* [ceph/ceph-csi on GitHub](https://github.com/ceph/ceph-csi)
* [Rook documentation](https://rook.io/docs/rook/latest/)

Deployment notes [#deployment-notes]

Ceph CSI Drivers deploys into the `rook-ceph` namespace alongside the Rook operator and Ceph cluster, and is versioned and synced as its own component in the `common` tier. Splitting the CSI driver out from the cluster component keeps driver upgrades (which affect every pod mounting Ceph-backed storage cluster-wide) as an explicit, independently reviewable change rather than a side effect of a cluster chart bump.

Interacts With [#interacts-with]

* `Rook-Ceph`, whose CephFS and RBD pools this driver exposes as Kubernetes `StorageClass` targets.
* Any platform workload with a `PersistentVolumeClaim` against a Ceph-backed storage class, since every mount and unmount operation is brokered through this driver.