I have two paragraphs in a zeppelin spark notebook:
val a = 5
and:
println(a)
If I run the second immediately after the first, then all is ok. but if I wait a few seconds, then the interpreter shuts down, and the second paragraph fails
In the interpreter log, the interpeter is seen to shut down, after a few seconds:
INFO [2017-03-14 12:58:12,053] ({pool-1-thread-3} Logging.scala[logInfo]:54) - Stopped Spark web UI at http://10.0.0.2:4040
INFO [2017-03-14 12:58:12,065] ({pool-1-thread-3} Logging.scala[logInfo]:54) - Shutting down all executors
INFO [2017-03-14 12:58:12,066] ({dispatcher-event-loop-11} Logging.scala[logInfo]:54) - Asking each executor to shut down
INFO [2017-03-14 12:58:12,087] ({dispatcher-event-loop-25} Logging.scala[logInfo]:54) - MapOutputTrackerMasterEndpoint stopped!
INFO [2017-03-14 12:58:12,103] ({pool-1-thread-3} Logging.scala[logInfo]:54) - MemoryStore cleared
INFO [2017-03-14 12:58:12,104] ({pool-1-thread-3} Logging.scala[logInfo]:54) - BlockManager stopped
INFO [2017-03-14 12:58:12,115] ({pool-1-thread-3} Logging.scala[logInfo]:54) - BlockManagerMaster stopped
INFO [2017-03-14 12:58:12,120] ({dispatcher-event-loop-28} Logging.scala[logInfo]:54) - OutputCommitCoordinator stopped!
INFO [2017-03-14 12:58:12,124] ({pool-1-thread-3} Logging.scala[logInfo]:54) - Successfully stopped SparkContext
INFO [2017-03-14 12:58:12,141] ({pool-1-thread-3} InterpreterGroup.java[close]:145) - Close interpreter group 2CAQSK5DV::2CBFFWCNP
INFO [2017-03-14 12:58:14,247] ({Thread-3} Logging.scala[logInfo]:54) - Shutdown hook called
INFO [2017-03-14 12:58:14,249] ({Thread-3} Logging.scala[logInfo]:54) - Deleting directory /tmp/spark-b0cfee1f-e8ee-49dd-aae2-2d3446bfaaa1
INFO [2017-03-14 12:58:14,256] ({Thread-3} Logging.scala[logInfo]:54) - Deleting directory /tmp/spark-9c7c38d8-3e82-4833-8557-afb94f3c3cb7
I tried adding interpreter configuration such as:
zeppelin.interpreter.persistent
=true
spark.qubole.idle.timeout
= 900 (I'm not using qubole, but seemed worth trying ...)
... but nothing changes this behavior.
how to handle this?
Edit: note that the same problem seems to exist with %python
interpreter too, so seems not interpreter/spark-specific, but global to all zeppelin interpreters