Github user egorklimov commented on the issue:
https://github.com/apache/zeppelin/pull/3102
@zjffdu If I'm not mistaken according to `interpreter.sh`:
```
HOSTNAME=$(hostname)
ZEPPELIN_SERVER=org.apache.zeppelin.interpreter.remote.RemoteInterpreterServer
INTERPRETER_ID=$(basename "${INTERPRETER_DIR}")
ZEPPELIN_PID="${ZEPPELIN_PID_DIR}/zeppelin-interpreter-${INTERPRETER_ID}-${ZEPPELIN_IDENT_STRING}-${HOSTNAME}-${PORT}.pid"
ZEPPELIN_LOGFILE="${ZEPPELIN_LOG_DIR}/zeppelin-interpreter-${INTERPRETER_SETTING_NAME}-"
```
Pid file generates every time (except case when interpreter process didn't
start successfully). But in yarn scenario it will be hard to analyze CPU and
memory load because resources will be consumed on each node, maybe someone will
add one more template for that case.
In case of container, I suppose we could generate other info in run folder.
Now my tests fail third Travis test, trying to fix that scenario ;)
---
