forked from rook/rook
Merge pull request #8532 from travisn/reset-release-notes
docs: Reset the release notes for v1.8
This commit is contained in:
+1
-36
@@ -1,6 +1,6 @@
|
||||
# Major Themes
|
||||
|
||||
v1.7...
|
||||
v1.8...
|
||||
|
||||
## K8s Version Support
|
||||
|
||||
@@ -10,47 +10,12 @@ v1.7...
|
||||
|
||||
### Ceph
|
||||
|
||||
- The Operator configuration option `ROOK_ALLOW_MULTIPLE_FILESYSTEMS` has been removed in favor of simply verifying the Ceph version is at least Pacific.
|
||||
Multiple filesystems are stable since Ceph Pacific.
|
||||
So users who had `ROOK_ALLOW_MULTIPLE_FILESYSTEMS` enabled will need to update their Ceph version to Pacific.
|
||||
|
||||
## Features
|
||||
|
||||
### Core
|
||||
|
||||
### Ceph
|
||||
|
||||
- Official Ceph images have moved from docker.io to quay.io. Users running tags like `v14.2`, `v15.2`, `v16.2` must change the registry URL.
|
||||
So the CephCLuster spec field `image` must be updated to point to quay, like `image: quay.io/ceph/ceph:v16.2`.
|
||||
- Add user data protection when deleting Rook-Ceph Custom Resources
|
||||
- A CephCluster will not be deleted if there are any other Rook-Ceph Custom resources referencing
|
||||
it with the assumption that they are using the underlying Ceph cluster.
|
||||
- A CephObjectStore will not be deleted if there is a bucket present. In addition to protection
|
||||
from deletion when users have data in the store, this implicitly protects these resources from
|
||||
being deleted when there is a referencing ObjectBucketClaim present.
|
||||
- See [the design](https://github.com/rook/rook/blob/master/design/ceph/resource-dependencies.md)
|
||||
for detailed information.
|
||||
- Add support for creating Hybrid Storage Pools
|
||||
- Hybrid storage pool helps to create hybrid crush rule for choosing primary OSD for high performance
|
||||
devices (ssd, nvme, etc) and remaining OSD for low performance devices (hdd).
|
||||
- See [the design](Documentation/ceph-pool-crd.md#hybrid-storage-pools) for more details.
|
||||
- Checkout the [ceph docs](https://docs.ceph.com/en/latest/rados/operations/crush-map/#custom-crush-rules)
|
||||
for detailed information.
|
||||
- Add support cephfs mirroring peer configuration, refer to the [configuration](Documentation/ceph-filesystem-crd.md#mirroring) for more details
|
||||
- Add support for Kubernetes TLS secret for referring TLS certs needed for ceph RGW server.
|
||||
- Stretch clusters are considered stable
|
||||
- Ceph v16.2.5 or greater is required for stretch clusters
|
||||
- The use of peer secret names in CephRBDMirror is deprecated. Please use CephBlockPool CR to configure peer secret names and import peers. Checkout the `mirroring` section in the CephBlockPool [spec](Documentation/ceph-pool-crd.md#spec) for more details.
|
||||
- Update Ceph CSI to `v3.4.0` for more details read the [official release note](https://github.com/ceph/ceph-csi/releases/tag/v3.4.0)
|
||||
|
||||
### Cassandra
|
||||
|
||||
- CRDs converted from v1beta1 to v1
|
||||
- Schema is generated from the internal types for more complete validation
|
||||
- Minimum K8s version for the v1 CRDs is K8s 1.16
|
||||
|
||||
### NFS
|
||||
|
||||
- CRDs converted from v1beta1 to v1
|
||||
- Schema is generated from the internal types for more complete validation
|
||||
- Minimum K8s version for the v1 CRDs is K8s 1.16
|
||||
|
||||
Reference in New Issue
Block a user