I would not recommend user to do that to add extra library. Because most of time zeppelin user don't know the cluster details like which/where extra library are located in the cluster machines. I would suggest livy to support dynamic library loading as native spark interpreter.

@zjffdu - Livy doesn't support dynamic library loading right now. So, I don't think there's a ZeppelinContext available in Livy. Would this be acceptable until dynamic library loading is added to Livy and a Zeppelin Context becomes available? As of now, Livy is not that much use if you cannot add libraries to your Spark app.

@jimdowling Right, there's no ZeppelinContext in livy right now. Would allowing user to specify jars/packages when creating livy session more general ? (like --jars/--packages in spark-submit)

Do you mean by modifying the UI on notebook creation or using a param to selecting the interpreter like
'%livy --jars [path]

No, I mean adding interpreter property to allow user to specify jars/packages.

I think that the livy itself doesn't support add (--jars/--packages property in spark-submit)

You can use --conf spark.jars=<jars> and --conf spark.jars.packages=<packages> instead

@zjffdu I added spark.jars.packages property in the recent commit, and it works correctly.

@zjffdu do you have any further comments ?

could you update the title of this PR and JIRA ZEPPELIN-1258, if we are changing how we would approach this?

@felixcheung I suggests this title "Adding packages to livy interpreter", Is it Ok ?

how about "Add Spark packages support to Livy interpreter"?

@felixcheung done.

@felixcheung do you have any further comments ?

There is a test failure?

@felixcheung : This PR does not change the code. It is just a change in the configurations. So, it should not affect the tests

That's a fair point.

I'll merge if there is no more discussion.