Github user jongyoul commented on the issue:
https://github.com/apache/zeppelin/pull/2899
@felixcheung That's a proper position and I've tried to implement the way
you mentioned. But as you're worried about, users might be confused that kind
of slow starting of Spark Interpreter, and even we have no way to show a proper
message that we are downloading spark to users. In a current implementation,
RuntimeException will be shown to FE side. I've talked with Jeff and am trying
to find a proper position to locate code to download Spark.
BTW, while I've struggled this issue, I thought it might take almost same
time to implement 'dynamic download function' :)
---
