Rook - General information


needs updates (differentiate between on OpenStack and on Bare Metal)

  • We configure the cluster to use Cinder CSI volumes as backing storage. The volumes are already replicated on our OpenStack’s Ceph level, so we set the pool size to 1 (= one replica only) in the Rook cluster.

  • Cinder volumes are numbered in rook (e.g. cinder-2-ceph-data). The numbering does not always correspond to the OSD number using that volume! This is important when debugging OSD issues and when removing OSDs and their storage especially.

  • Adding volumes can cause the worker instance to crash with a kernel panic due to a known kernel bug with a race condition when detaching the Cinder volume from the instance.

  • To access ceph tools, run:

    $ kubectl -n rook-ceph exec -it $(kubectl -n rook-ceph get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0]}') bash
  • Rook stores information about the provisioning state of volumes in ConfigMap objects while the provisioning hasn’t completed yet. When removing a provisioning job half way through, it is important to also clean up the corresponding ConfigMap object, otherwise the operator will hang.

  • The ceph mons use the local-storage StorageClass which is a fancy version of hostPath and has the advantage of “binding” a pod to a node. local-storage works through a controller that presents disks (or bind-mounts of directories, as in our case) as PVs to K8s. The controller also attaches PVCs to these PVs.

  • Note that it’s usually safe to delete a ceph mon and its PVC if at least one healthy mon remains. To be on the safe side, make sure the quorum is > floor(mons / 2).