Github user Tagar commented on the issue:
Thanks for the heads up, yep I figured out that I have to tune up
zeppelin.ipython.grpc.framesize to a large number.
I looked over the PR. Two quick suggestions
1) Would it be possible to make spark interpreter keep and not close the
stream if such an exception happens? We can see a higher limit, but I am sure
users will have cases when they will try to go higher. The Spark interpreter is
then in a bad state and only way to fix this is to try increase a limit again..
Not sure if this problem belongs to Zeppelin or to grpc, so provisionally
opened an issue in grpc too - https://github.com/grpc/grpc-java/issues/4086
2) Should we increase the default? .. 4Mb isn't that hard to hit when
ipython returns a mid-size dataset / table.