我在cassandra中有数据，其中key（id）是二进制格式的，我正在尝试在hive表（与longtype具有相同的id字段）之间进行查询。因此，当我在spark sql 2.0.1中运行此查询时：

SELECT a.key
FROM cassandra_table a
WHERE a.key in(cast('829404561643414' as binary));

它会在几秒钟内返回结果。另外，当我运行这个查询时，spark sql：

SELECT cast(cast(id AS string) as binary)
     FROM hive_table c LATERAL VIEW EXPLODE (c.field) pixTable AS pix LATERAL VIEW EXPLODE (pix.fieldList) olTable AS olc
     WHERE (c.hit_date_utc = 20161001) LIMIT 1;

它还会在几秒钟内返回结果。
但是当我做这样的子查询时：

SELECT a.key
    FROM cassandra_table a
    WHERE a.key in(SELECT cast(cast(id AS string) as binary)
         FROM hive_table c LATERAL VIEW EXPLODE (c.field) pixTable AS pix LATERAL VIEW EXPLODE (pix.fieldList) olTable AS olc
         WHERE (c.hit_date_utc = 20161001) LIMIT 1);

它创建了大约40000个spark任务，而job只是继续运行。
知道为什么会这样吗？