Github user mebelousov commented on the issue:
https://github.com/apache/zeppelin/pull/2733
@zjffdu, yes, I've got this error in production.
Schedulers aren't shared, there is one scheduler per each interpreter
processes. If user starts new interpreter processes, Zeppelin creates new
scheduler object.
ExecutorFactory limits the number of schedulers and therefore the number of
java processes.
In description I mean that interpreter is java process.
How correctly to say about started interpreter process/JVM?
I wrote about this bug in mailing lists:
[1](https://lists.apache.org/thread.html/30966875b50b5ac8b4326c23a012d702ea7cac24d75f540ae58f31b3@%3Cusers.zeppelin.apache.org%3E),
[2](https://lists.apache.org/thread.html/a5e4de950d74a858e249af9e07fb17f63c94a136661ca78fc0c26c90@%3Cdev.zeppelin.apache.org%3E)
@felixcheung
Our team uses interpreters in isolated mode per user/per note.
Every night we restart Zeppelin.
The analysts run notes and zeppelin starts more than a 100 JVMs in the
evening.
---
