I think it'd be great if databases had a "back in time" feature, ability to view the data in the form how it looked like in the past. Instead you have to always develop it again and again in the schemas - very often keeping complete history is desired. Storage is generally no more an issue in 2015, especially if compression is applied.
Even if you keep full query logs, getting to the state the database was 10th of February 2014 05:34:23 UTC is rather inconvenient and slow.
You wouldn't need bitemporal schemas either with a feature like this.
Another pet peeve in relational databases is a no-compromise hierarchy (and graph) storage. You can usually get fast reads or fast inserts, but there always seems to be a compromise.
As I understand it that is one of the selling points of the product Datomic. The database in its entirety effectively can be worked with as a value type.
Even if you keep full query logs, getting to the state the database was 10th of February 2014 05:34:23 UTC is rather inconvenient and slow.
You wouldn't need bitemporal schemas either with a feature like this.
Another pet peeve in relational databases is a no-compromise hierarchy (and graph) storage. You can usually get fast reads or fast inserts, but there always seems to be a compromise.