[AtlasBarfed]

Why isn't it horizontally scalable? Are there tons of joins?

I wrote a JSON/Document/Property Graph database front end with Cassandra and DynamoDB backends, so that is ... kinda ... schemaless.

I ended up doing a lot of things that RethinkDB and OrientDB did but didn't scale like cassandra. I was trying to do a streaming document layer. Also a JSON document database with simple relations is a property graph database.

But Elassandra and similar projects kinda did the same thing so I dropped it. But I came out of it thinking it shouldn't be hard to scale JSON, which is basically schemaless.

I had schema-optional things too where certain named properties/elements of a JSON doc would be rendered onto a fixed table and schema, are you.guys doing anything like that? I think a schema optional or options for partial schemas for documents would enable the transition of schemaless data to schema'd stores in enterprise system evolution.

Do you guys intermix all documents like graph databases do, which seems a bit insane if you have to do some large scale table scan or migration of data and have to sift through data? Or build an index on a heap of documents with only some of them needing indexing?