mirror of
https://github.com/ceph/ceph-csi.git
synced 2025-06-14 02:43:36 +00:00
ci: fix mdl related failures
This commit address the issue- https://github.com/ceph/ceph-csi/issues/3448. Signed-off-by: riya-singhal31 <rsinghal@redhat.com>
This commit is contained in:
committed by
mergify[bot]
parent
d721ed6c5c
commit
539686329f
@ -79,13 +79,13 @@ volume is present in the pool.
|
||||
## Problems with volumeID Replication
|
||||
|
||||
* The clusterID can be different
|
||||
* as the clusterID is the namespace where rook is deployed, the Rook might
|
||||
* as the clusterID is the namespace where rook is deployed, the Rook might
|
||||
be deployed in the different namespace on a secondary cluster
|
||||
* In standalone Ceph-CSI the clusterID is fsID and fsID is unique per
|
||||
* In standalone Ceph-CSI the clusterID is fsID and fsID is unique per
|
||||
cluster
|
||||
|
||||
* The poolID can be different
|
||||
* PoolID which is encoded in the volumeID won't remain the same across
|
||||
* PoolID which is encoded in the volumeID won't remain the same across
|
||||
clusters
|
||||
|
||||
To solve this problem we need to have a new mapping between clusterID's and the
|
||||
|
Reference in New Issue
Block a user