Ceph restore from backup can attempt to discard unbounded amount of data

[Jason Dillaman]

This is semi-related to an upstream Ceph ticket [1] where a discard of more than 2GB was attempted against an RBD image and it resulted in a failure. As an extreme example, if a 1GB image is restored onto a 1TB RBD image, the discard request from the Cinder backup driver would result in 261,888 discard operations being sent concurrently to the Ceph cluster.

I suggest that the discard could be restricted to some amount like the existing chunk size parameter -- or avoid discarding at all and simply shrink the RBD image when restoring a smaller backup since librbd automatically throttles the discard operations associated with a (shrink) resize.

[1] http://tracker.ceph.com/issues/21966