So a common thing in tech seems to be "oh this new tool will solve everything!". I think this phenomenon is good in the sense that it means that "religious" adherents to a tech will push its boundaries, find out where it is well suited and where it lacks, etc. I think this is just a natural outgrowth of excitement about the new shiny thing -- software people are surprisingly passionate about their craft. Of course this has a tendency to lead to annoying religious wars (sometimes at the level of pointless minutia like semi-colons :) ).

But for me the exciting part is when the fervor dies down a bit, and there is a clear understanding of "opposing" tools' strengths and weaknesses. That is when interesting tech and techniques are developed to handle both based on the strengths.

I think we are starting to get to that point with relational and nosql. Tools are starting to emerge on the relational DB side to allow more nosql-like structures within traditional DBMS systems. I think in the coming year or two we'll start seeing a lot more interesting tools to mix relational and non-relational datastores. Currently the common way of mixing is to use kv or document stores as a caching layer for relational data, or to have relational systems separate from nosql systems, with custom code to select from each. My hope tho is that there will be more uniform access methods developing, allowing the relational and non-relation data to have a common, well understood access API layer.

For example I have a project where the core data consists of users, certain user objects, and relations between users, between users and their or others objects, and between objects, basically shaped like a graph. Using neo4j + gremlin makes dealing with this portion of the app trivial - the graphdb literall solves problems for me. But another aspect of the app is relational data on objects related to user records, basically it is shaped as rows where n objects' rows need to be merged, filtered and then operated on (aka JOINed). SQL makes this set of operations trivial, in this case a DBMS literally solves problems for me. But, as it stands, I need to query one DB, get the object record pointers I need, then connect to a different DB, and query the relevant tables if I want to let the relevant DBs do the work. This involves different APIs, weird connection management, and sysadminning two different database environments.

Essentially I'm hoping someone smarter then me will come up with a way for me to:

   SELECT MY_COLLATE(d1.a, d2.b) FROM
      GREMLIN{find object o1}.data as d1 JOIN
      GREMLIN{find object o2}.data as d2 ON
          d1.col == d2.col
   WHERE conditions