Rook Ceph#
Rook is used to operate a Ceph cluster.
Ceph cluster configuration#
Current setup#
Currently the setup is based on the test recommendations since there is no need for high availability yet and data restore is part of the cluster bootstrap process already.
When expanding to 3+ nodes this should be re-considered.
Monitor#
Quorum: 1, 3, …
HA: 3
Manager#
Redundancy: At least two
HA: At least two
OSDs (Object Storage Daemon)#
Redundancy: At least 3
HA: At least 3
CSI snapshotter version coupling#
The Rook operator deploys CSI provisioner pods with a csi-snapshotter
sidecar. The sidecar version is hardcoded per Rook release and must be
compatible with the VolumeGroupSnapshot CRDs installed by the
snapshot-controller Helm chart (Piraeus).
Both components originate from the same upstream project (kubernetes-csi/external-snapshotter) but are deployed independently. When upgrading either Rook or the snapshot-controller chart, verify that the snapshotter sidecar version and the CRD API versions match.
A mismatch causes the snapshotter informer to fail silently, blocking all VolumeSnapshot operations (backups and restores).
Check the current sidecar version:
kubectl get pods -n rook-ceph -l app=csi-rbdplugin-provisioner \
-o jsonpath='{.items[0].spec.containers[?(@.name=="csi-snapshotter")].image}'
Override via operator Helm values if needed (csi.snapshotter.tag).