SparkSQL中纵向迭代计算的3种方式，枚举、python的for循环、pandas_UDF函数（附案例）

需求：计算c4的逻辑：当 c2 = 1，则 c4 = 1；否则 c4 = (上一个 c4 + 当前的 c3) / 2

1、用枚举方式：

--准备数据
create or replace view test2(c1,c2,c3,c4) as values
(1,1,6,1),
(1,2,23,null),
(1,3,8,null),
(1,4,4,null),
(2,1,32,1),
(2,2,9,null),
(2,3,15,null),
(2,4,8,null);

select * from test2;
set spark.sql.shuffle.partitions=4;
select c1, c2, c3,
       case when c2=1 then 1
           else (lag(c4) over (partition by c1 order by c2)+c3)/2
       end    as c4
from test2;
--使用枚举的方式
--计算当c2=2时 ，c4=?
create or replace temporary view test_c2 as
    select c1, c2, c3,
           if(c2=2,(lag(c4) over (partition by c1 order by c2)+c3)/2,c4) as c4