Ceph

(Implicit in the above is that ECBackend drives application, and PG is merely notified, this may not be the best design and should be justified or discarded)

ECBackend will have discretion about when to actually apply a particular log entry to allow it to batch up many applications at once while allowing it to control the size of its volatile cache of unstable object data. We'll need plumbing in PGBackend to allow PGLog to update its metadata about which entries have been applied (and therefore cannot be rolled back).