[ZEPPELIN-3247] Restart grpc stream for each paragraph run

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 0.7.3, 0.8.0
• Component/s: Core, pySpark, python-interpreter, zeppelin-interpreter, zeppelin-server
• Assignee: Unassigned
• Reporter: Ruslan Dautkhanov
• Created: 19/Feb/18 23:36
• Updated: 19/Feb/18 23:36

Description

Please read https://github.com/grpc/grpc-java/issues/4086 for details

The RPC (a.k.a. the stream) cannot continue, but you can start a new RPC. The Channel will still work. You can start a new RPC on the existing channel. Without seeing the code I don't know why a failed RPC would fail the whole application, but that isn't the intended behavior.

 
It seems spark interpreter --> grpc --> ipython backend is currently somewhat brittle as any exception stops grpc stream [1].
 
Would it be possible to adjust code in ipython logic to restart grpc stream for each paragraph run ? To make new ipython logic more robust.
 
 
 
 
[1]
 

INFO [2018-02-14 10:39:10,923] ({grpc-default-worker-ELG-1-2} AbstractClientStream2.java[inboundDataReceived]:249) - Received data on closed stream
INFO [2018-02-14 10:39:10,924] ({grpc-default-worker-ELG-1-2} AbstractClientStream2.java[inboundDataReceived]:249) - Received data on closed stream
INFO [2018-02-14 10:39:10,925] ({grpc-default-worker-ELG-1-2} AbstractClientStream2.java[inboundDataReceived]:249) - Received data on closed stream