Indexed Relations
Also in 2.1-SNAPSHOT we have added a new feature called Indexed Relations. The main idea is to allow efficient fetch of object(s) in a relation, by using one of their attributes.
Putting it simple, if an Address has a country (attribute in the class), and a Person is related to many Addresses, this allows easily obtaining the Address of a Person that has a given country. Not only it is more efficient with regard to single thread performance, but it is also transactional-friendly because it reduces the footprint that makes a transaction vulnerable to aborts.
These indexes also allow multiple values to collide in the same indexed key. In other words, a Person might have more than one Address with the same country. As a result, the indexes would return a set of Addresses for a given query. This is defined in the DML file according to the documentation. That allows the framework to generate additional code in the Domain Classes to support the queries through simple methods (once again, also described in the documentation).
Finally, the internal implementation is interestingly backend-independent by relying on internal DML-based collections that back up the relations. As a matter of fact, an index is implicitly stored in the collection that backs the relation. Therefore using a unique index poses no additional overhead. This has resulted in speedups of one order of magnitude across several benchmarks (synthetic and realistic) and backends.
Posted by 