Currently we could retry updating the on disk "state" for a resharding operation and hit an invariant that the previous state was not correct. This is because the in memory and on disk state are not synchronized.

Either update the invariant to handle this, or synchronize the in memory state.