I've been working on improvements to the SQL interpreter, but I'm not quite sure why it works the way it currently does. Right now, the interpreter is associated with a SQL service that is a BrokerService. BrokerService says that it is a high-level interface between the JVM and another process, like R or Python, so it isn't clear why SQL uses it.
The version I have working is considerably simpler and uses the lazy `kernel.sparkSession` accessor. Is there a reason why the current one doesn't use this approach? rb -- Ryan Blue Software Engineer Netflix