If query a non-exist column against a parquet file, DRILL currently raises ExecutionSetupException:
SELECT NON_EXIT_COLUMN from cp.`tpch/nation.parquet`;
This will cause problem, when we have multiple parquet files, and the first one does not have the column, while the rest of them have the column.
It would be better to return NULL expression during execution stage, if the columns does not exist in the parquet file.
Later on, if DRILL adds a new option to verify column existence before executing a query ( just like what a schema-based system will do), we could throw PlanException in planning phase, once the parquet footer information is available to planner.