Very likely in Spark 1.4. Hortonworks has been helping out with this, we just need some internal refacotring to the API to make it work.