Efficiency improvements in OpenOA - ENTR warehouse constructor, use PySpark and Arrow in OpenOA ENTR Constructor

[ejsimley]

We found that using PyHive and HiveThriftServer2 to load data into OpenOA is unacceptably slow (~1 minute to load La Haute Borne data into a PlantData). This ticket aims to overcome this issue by switching the ENTR constructor to use PySpark directly instead of PyHive, and ensure reasonably correct optimizations are made (e.g., Apache Arrow, number and size of Spark executors which makes sense for the demo).

Impacted Repositories:

• entr/entr_runtime
  • remove automatic pyhive at startup
  • tune spark standalone cluster
• entr/openoa
  • switch entr constructor to use pyspark dataframes instead of pyhive