Major Description
When impersonation is disabled, hadoop UGI user will always be livy server user, there is no way to get actually owner user in spark application.
This issue is to fix this problem by add a "spark.livy.owner" property to spark context. Spark application code can get this property if needed.
Actually, we use this way to control Spark SQL permissions. We use a spark.sql.extensions to get owner and decide whether operation is allowed. Spark application is submitted to YARN using hive user, this will make Spark SQL permission model exactly the same with hive(Sentry) permission model.
Attachments
Issue Links
- links to
