Feature #21305
closedJust discard changed data since snapshot in "rbd rollback" command
0%
Description
Rolling back images is painfully slow. Yes, I know, "rbd clone", but this creates another image and automatic-snapshotting-backuppers get failed.
Why not to bring image to the state of specified snapshot ?
Updated by Jason Dillaman over 6 years ago
- Project changed from Ceph to rbd
- Subject changed from Just discard changed data since snapshot in "rbd roollback" command to Just discard changed data since snapshot in "rbd rollback" command
Only possible if you are rolling back to the most recent snapshot. If you are rolling back to an older snapshot, you would need to copy the older data to the HEAD revision instead of just deleting the HEAD revision to expose the most recent snapshot.
Updated by Марк Коренберг over 6 years ago
Yes, ability to do FAST rollback to latest snapshot will be huge improvement, IMHO.
Updated by Jason Dillaman over 6 years ago
- Status changed from New to In Progress
- Assignee set to Jason Dillaman
Updated by Jason Dillaman over 6 years ago
- Status changed from In Progress to Rejected
I quickly hacked this up and it turns out the OSDs don't actually allow you to do this. The low-level, internal transactions within the OSD will still clone the HEAD object and then "delete" it, which results in a zeroed object at HEAD.
Updated by Марк Коренберг over 6 years ago
I don't understand cause why this would not be implemented. Maybe open this issue until someone be able to fix it ?
Updated by Jason Dillaman over 6 years ago
There is nothing RBD can do -- you can perhaps open a new ticket against RADOS to optimize snap rollback directly.