Best refresh approach when defining pre-aggregations with big tables and mutable data?