I'd say RethinkDB (http://rethinkdb.com). It's labelled as not production ready, but most of the things that matter have been solidified (like data integrity).
It takes very good cues from both the NoSQL and SQL worlds...it's document-based but supports server-side joins. It uses MVCC (no global write lock) and while I haven't scaled it yet, the server administration is a lot simpler than MongoDB (`rethinkdb join [any_server_in_cluster]` vs creating 3 config servers and a bunch of repsets and setting up sharding schemas that just don't work). Also, the query language is crazy good. It's the only database where I've felt like I'm programming against my data (map/reduce doesn't come close).
There are some gotchas, like it doesn't have a binary storage field type, and to upgrade versions you have to dump/restore your data. I'm sure there are more. Like I said, not "production ready" yet. Give it a year though.
It takes very good cues from both the NoSQL and SQL worlds...it's document-based but supports server-side joins. It uses MVCC (no global write lock) and while I haven't scaled it yet, the server administration is a lot simpler than MongoDB (`rethinkdb join [any_server_in_cluster]` vs creating 3 config servers and a bunch of repsets and setting up sharding schemas that just don't work). Also, the query language is crazy good. It's the only database where I've felt like I'm programming against my data (map/reduce doesn't come close).
There are some gotchas, like it doesn't have a binary storage field type, and to upgrade versions you have to dump/restore your data. I'm sure there are more. Like I said, not "production ready" yet. Give it a year though.