Since fields that explicitly aren't nullable are added to the filters (for pruning), the MongoSamplePartitioner needs to perform a full scan to count the filtered rows.

In some cases, it is very slow, especially on large data set. I notice that mongo-connector 2.0.0 uses an inaccurate count instead of performing a count, so that the problem does not exists.

Is it possible to add some warning information or document somewhere to prevent someone from misusing the InNullable schema qualifier.