rook-ceph-csi-drivers
Purpose
Deploys the Ceph-CSI drivers (RBD + CephFS) via the upstream ceph-csi-drivers Helm chart. As of Rook v1.20 the Rook operator no longer deploys the CSI drivers itself — it only installs the ceph-csi-operator (controller + CRDs) as a subchart of the rook-ceph-operator chart. This HelmRelease creates the Driver custom resources that the ceph-csi-operator reconciles into the actual node-plugin DaemonSets and controller-plugin Deployments.
Without this release, CSI provisioning and volume mounts fail.
Configuration
Values mirror Rook’s recommended defaults plus this cluster’s customizations:
operatorConfig.namespace: rook-ceph— must match the operator namespace.operatorConfig.driverSpecDefaults.imageSet.name: rook-csi-operator-image-set-configmap— reuses the image set ConfigMap published by the Rook operator chart so CSI images stay in lockstep with the Rook version.- Driver names keep the
rook-ceph.prefix (rook-ceph.rbd.csi.ceph.com,rook-ceph.cephfs.csi.ceph.com). This MUST match existing PVspec.csi.drivervalues and the StorageClass provisioner names (csiDriverNamePrefixin therook-ceph-clusterchart). Changing it orphans every existing volume. nodePlugin.tolerations— node plugin DaemonSet runs on all nodes (migrated from the old operatorcsi.pluginTolerations).controllerPlugin.tolerations— provisioner pinned to control-plane (migrated from the old operatorcsi.provisionerTolerations).drivers.cephfs.kernelMountOptions.ms_mode: crc— migrated from the old operatorcsi.cephFSKernelMountOptions: ms_mode=crc; required because the cluster enforces msgr2 (requireMsgr2: true).drivers.rbd.snapshotPolicy: volumeSnapshot— the chart defaults the RBD driver’ssnapshotPolicytonone, which omits thecsi-snapshottersidecar from the RBDcontrollerPlugin(5/5 containers instead of 6/6). The old Rook-operator-managed CSI enabled the snapshotter by default, so the v1.20 split silently broke RBDVolumeSnapshotcreation — every RBD-backed VolSyncReplicationSourcehung atreadyToUse: false. Setting this restores parity with thecephfsdriver (which the chart already defaults tovolumeSnapshot). Gotcha: when adding new RBD-backed apps that need snapshots, this must stay set.
Dropped during the v1.20 migration
csi.enableLiveness— no equivalent inceph-csi-driversv1.0.1.cephClusterSpec.csi.readAffinity— the CephCluster.spec.csifield was removed and the v1.0.1 drivers chart exposes noreadAffinity. Read-locality is lost (negligible impact on a small homelab cluster).
Dependencies
rook-ceph-operator(installs the ceph-csi-operator CRDs) — enforced via FluxdependsOn.
Source
- Chart:
ceph-csi-drivers1.0.1fromhttps://ceph.github.io/ceph-csi-operator(published only as an HTTP Helm repo; no OCI registry). Version tracks theceph-csi-operatorsubchart bundled by Rook v1.20.0.