Sparkling water External Backend Support in kerberized cluster
Additionally expose H2O params:
principal, keytab, run_as_user
In internal backend, we use spark.yarn.principal and spark.yarn.keytab.
However in external backend, automatic mode, we also need to set up kerberos parameters to h2o driver.
Therefore we need to expose principal and keytab as params. We can be smart and if we see that spark.yarn.keytab and spark.yarn.principal is set, automatically set principal and keytab in External backend
After discussion with , we came to a conclusion that Spark as the main application should be handling kerberised access
Actually, principal & key tab are available just in case of the H2ODriver.
However it makes more sense to use kerberos principal and & keytab on the sparkl level as H2O is just embedded application