Yeah, I have talked a lot with TD (Spark), Job H(HBase), Stacks(HBase) about this. Nether thing HBase or Spark is the right project to put it in.
Right now the code is in Cloudera Labs and a github and works for CDH 5.3 and 5.4 we have a number of clients on it.
There is talk to make it an apache project. It is apache listened but it would be nice to put it under apache totally. The problem is it is soooooo simple some times it feels to small to be it's own project.
The design is just to have a HBase connection in a static location in the executor.
I know other NoSql brag about local gets, but HBase already had that even without SparkOnHBase. The Table input format already gives you local gets.
All Spark on HBase gives you is an active connection that can be accessed in the distributed function of Spark. Which is very important to some use cases. Like Spark Streaming and complex graph local.
Let me know. We are open to ideas.