Set (configurable) limit to number of keys fetched in updateAttributes

[hafen]

To scale to very large numbers of subsets, it is unrealistic to expect to be able to retrieve all the keys in updateAttributes. When the number of subsets is very large, it is most likely that Hadoop is being used, in which case it is not as important to have all the keys at once.