SPARKSQL中分组排序并提取前N行

select * from (select *,row_number() over (PARTITION BY col1,col2,col3 ORDER BY col4 desc) rank from df1) tmp where rank<=10
这个是提取前10的数据