Spark in SQL select count (1) the from tableName each map tasks can be carried within the partition data loading in the count, or strips traversal execution count?

In addition to shuffle, the SQL in which stage will cause the memory leak?

A great god guide stay about