Rook cluster-wide encryption can now use the native Kubernetes authentication to interact with vault KMS instead of using the token method. Signed-off-by: Sébastien Han <seb@redhat.com>
23 lines
915 B
Markdown
23 lines
915 B
Markdown
# Major Themes
|
|
|
|
v1.8...
|
|
|
|
## K8s Version Support
|
|
|
|
## Upgrade Guides
|
|
|
|
## Breaking Changes
|
|
|
|
- Flex driver is fully deprecated. If you are still using flex volumes, before upgrading to v1.8
|
|
you will need to convert them to csi volumes. See the flex conversion tool.
|
|
- Min supported version of K8s is now 1.16. If running on an older version of K8s it is recommended
|
|
to update to a newer version before updating to Rook v1.8.
|
|
|
|
## Features
|
|
|
|
- The Rook Operator does not use "tini" as an init process. Instead, it uses the "rook" and handles
|
|
signals on its own.
|
|
- Rook adds a finalizer `ceph.rook.io/disaster-protection` to resources critical to the Ceph cluster
|
|
(rook-ceph-mon secrets and configmap) so that the resources will not be accidentally deleted.
|
|
- Add support for [Kubernetes Authentication when using HashiCorp Vault Key Management Service](Documentation/ceph-kms.md##kubernetes-based-authentication).
|