Description of problem: When backuping a cinder volume with Ceph as a backend, the operation terminates successfully but Cinder backup logs an error about rbd import-diff Result is having full volume backups dump instead of incrementals. Version-Release number of selected component (if applicable): 10 - Might be on 11 too haven't tried How reproducible: Always Steps to Reproduce: 1. Create a 2G cinder volume cinder create --name vol2 2 2. Backup the volume (volume is available) openstack volume backup create vol2 3. Create an incr backup (volume is available) openstack volume backup create (--increment) vol2 Actual results: Cinder backup fails with: /var/log/cinder/backup.log Importing image diff: 0% complete...failed.backup.drivers.ceph [req-08d753ea-8ce1-44e6-ac3b-72e4c7d6ab01 b1362809137541ce8a3de5ff55eedc27 9fd36bcb07884224b3ecd2e22d730931 - default default] RBD diff op failed - (ret=33 stderr= rbd: import-diff failed: (33) Numerical argument out of domain The error occurs during both backups steps (full and incremental) If we looks at rbd images they all have the cinder volume size (2G) # rbd du -p backups PROVISIONED USED volume-389ce208-9c67-483c-8128-46c43adcf436.backup.44977689-8287-4cdb-a551-6b15bf661e50 2048M 2048M volume-389ce208-9c67-483c-8128-46c43adcf436.backup.f7cd81fa-0c55-4ee5-9b52-bdf29b1ea76d 2048M 2044M <TOTAL> 4096M 4092M Expected results: Backups should use rbd-diff mechanism, this bug has an impact in terms of time to backup and space used. The full backup should looks like : NAME PROVISIONED USED volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base.snap.1497363445.68 2048M 32768k volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base 2048M 0 Additional info: Looking for upstream bugs I found one[1] that matches this issue for mitaka with users still complaining for Newton/Ocata. Using patch from comment #18 on file /usr/lib/python2.7/site-packages/os_brick/initiator/connectors/rbd.py seems to solves the issue. Redoing the full backup + one incremental gives : rbd du -p backups warning: fast-diff map is not enabled for volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base. operation may be slow. NAME PROVISIONED USED volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base.snap.1497361566.5 2048M 32768k volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base.snap.1497361619.98 2048M 16384k volume-389ce208-9c67-483c-8128-46c43adcf436.backup.base 2048M 0 <TOTAL> 2048M 49152k However I still have the same error if the volume is in-use or if using --snapshot [1] https://bugs.launchpad.net/cinder/+bug/1578036
I came across https://bugzilla.redhat.com/show_bug.cgi?id=1375207 which is similar but when using a different Ceph cluster for backups. In this BZ the backups and the volumes are in the same cluster (default config) so this bug affects both cases (same and different clusters)
*** This bug has been marked as a duplicate of bug 1375207 ***
We are running into the same issue and this is a separate bug from 1375207.