[CEPH] cinder-backup service blocked by creating volume from backup

[Jeremy Liu]

When creating a large volume(1000GB) from a backup, cinder-backup service no longer send heartbeat to cinder-scheduler in a perioid of time. Further debug shows the backup restore process hangs at librbd diff_iterate function when it tries to calcuate whether the given rbd volume has extents.

volume driver: rbd
backup driver: ceph

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/cinder/+/804470

[Sofia Enriquez (lsofia-enriquez)]

Greetings Jeremy Liu,
Do you mind sharing with us more detailed information. In this way we could reproduce or help with this.
- steps to reproduce the problem.
- logs: c-api c-bak and c-vol and any other information you think It may help us.- cinder conf
- from the patch proposed in master I guess you're facing this issue on master but I need to ask: which cinder version are you referring to?
Thanks in advance,Sofia

[Jeremy Liu (liujiong)]

I'm testing against cinder rocky release, but I suppose this also affect master.
volume driver: rbd
backup driver: ceph

steps to reproduce:
1. create a large volume, cinder create --name large-volume --availability-zone az01 1000
2. attach the volume to instance, mount volume to instance file system, write data into volume, make volume usage to about 50%
3. create a backup for the volume
4. create volume from the backup, cinder create --backup-id <backup-uuid> --name volume-from-backup 1000

Then watch cinder service-list command output.