Bug #36074
Updated by Jason Dillaman over 5 years ago
If an image is being synced by an rbd-mirror daemon and is then blacklisted from the local cluster while it's deep-copying an object (specifically after it's written the last snapshot version but before it updates the object map), it might silently fail without passing the -ESHUTDOWN error code back to "ImageCopyRequest<I>::handle_object_copy". Without the failure error code, the sync point object position can get improperly updated so that the now the object map remains incorrect yet not flagged as invalid (due to the blacklisting).
http://qa-proxy.ceph.com/teuthology/jdillaman-2018-09-18_08:50:35-rbd-wip-rbd-mirror-distro-basic-smithi/3038017/teuthology.log