I was looking for a filestore; embeddable, document oriented with versioning, index heavy that could support data augmentation and enrichment and have settled on (for now) apache cassandra.
I wrote the basic building blocks myself as a custom B+tree log append file which was intellectually interesting (and from initial tests fast) but not so much for work towards a first product. I may clean up and open source when I get some time. The thing that sold me was the need for immediacy. I could spend another week fixing bugs etc, but I think cassandra will do at the moment and seems well supported.
Got it up and running with a basic client storing data in some tables in about a day so easy enough to get started. The worse bit so far is that to me the thrift API feels awkward. I’ll probably either write a quick layer myself or go try find a nice one for scala, seems to be a few about, maybe hector