numRecords()/dataSize() would reset to 0 after each restart, but for some reason the size-based oplog truncation test still passed when it shouldn't have. After SERVER-131592 fixed numRecords()/dataSize() to return the correct value, the test still passes but we should investigate why the test was passing when it shouldn't have been. 

From yuhong.zhang@mongodb.com : One possible theory is that the test was short living so writes were not checkpointed yet and oplog entries got replayed after restart. That way the sizes/counts would have been tracked in memory