Skip to main content

Disk Layout

While most of the systems are immutable and bootstrapped on demand after power cycling (see relevant section of the docs) they do have state in place. This state is persisted multiple XFS filesystems on top of partitioned disks.

Each layer as discussed in the architecture section possibly utilizes state and is thus briefly presented here. It further includes a rough estimation on storage requirement from small to large.

Kubernetes​

Kubernetes needs to persist three distinct items.

Common​

  • /etc/rancher: Static Kubernetes metadata: small (at least 300MiB)
  • /var/lib/kubelet: Scratch space for Kubernetes: medium (~50GB)
  • /var/lib/rancher: Local kubernetes state: large (typically the remainder of the disk > 100GB)

OpenStack​

Depending on your individual setup, you might require local storage on your nodes. If you fully rely on an external storage provider like ceph this is not necessary. Instead those would be provided by that provider.

Computenode​

  • /var/lib/nova: Local OpenStack virtual machine state: large

Storage​

The disks used for storage are highly dependent on your storage engine.

LVM​

  • /var/lib/glance: Storage for glance disk images: large
  • /var/lib/cinder-conversion: Scratch space for cinder: large
  • cinder partition if using LVM: as large as possible
warning

This feature is currently in private beta

Ceph​

  • /var/lib/rook: rook state: medium (a few GB)
  • devices or partition of the same size with the correct OSB label: as large as possible (refer to ceph for more information)