After thinking a little more on this, the more appropriate question might be , "does all the data get loaded into the buffer cache before the filter is applied?"
The filter will be applied in Oracle. OSCH does not push down the filter to Hadoop to execute and MapReduce code. So the predicate would be handled the same way as any other external table or I presume any (non-exadata) table would be.
Along the same lines of this question, does anybody know a way to connect from Oracle to Hadoop/HDFS in such a way as to push the query filter down into the Hadoop query?
I really don't want to pull ALL my hdfs data into the Oracle buffer cache for processing!