Github user zjffdu commented on the issue:
https://github.com/apache/zeppelin/pull/2986
For now yarn-cluster mode of spark interpreter is the only use case that
run interpreter process in remote host. And we use property to config the
portRange for the remote spark interpreter connect back to zeppelin server. In
future we may runt all the other interpreters in containers which is another
scenario of running interpreter in remote host. For this scenario, I think it
is easy to expose one port when starting this container which is not difficult
IIUC based on my limited knowledge on containers. BI directional RPC is very
common pattern in distributed computing, I think it is worth to use Bi
directional RPC in zeppelin with regarding its benefits mentioned int he design
doc. @Leemoonsoo What do you think ?
---
