Array_Agg 不支持 Snowflake 中的 Window Frame - 如何实现？

Question

我需要跑步：

select arrayagg(o_clerk) 
  within group (order by o_orderkey desc) 
  OVER (PARTITION BY o_orderkey order by o_orderkey 
     ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) AS RESULT
from sample_data

但是 Snowflake 返回错误

Sliding window frame unsupported for function ARRAYAGG

。如果我尝试在没有滑动窗口的情况下累积所有内容，则会收到错误

Cumulative window frame unsupported for function ARRAY_AGG

。

我怎样才能实现这个目标？

样本数据：

create or replace table sample_data as (
    with data as (
        select 1 a, [1,3,2,4,7,8,10] b
        union all select 2, [1,3,2,4,7,8,10]
    )

    select 'Ord'||a o_orderkey, 'c'||value o_clerk, index
    from data, table(flatten(b))
)
;

想要的结果：

（源，用于 BigQuery 迁移）

Answer 1

谢谢@Felipe Hoffa，这是我最初发布的解决方案：

https://community.snowflake.com/s/question/0D73r000006upCECAY/detail?fromEmail=1&s1oid=00Di0000000hZh2& s1nid=0DB3100000001Fq&s1uid=0050Z000009Xcck&s1ext=0&emkind=chatterCommentNotification&emtm=1669252526852

select o_orderkey, 
    array_compact([
        lag(o_clerk, 3) over(partition by o_orderkey order by index)
        , lag(o_clerk, 2) over(partition by o_orderkey order by index)
        , lag(o_clerk, 1) over(partition by o_orderkey order by index)
        , o_clerk
    ])
from sample_data

Answer 2

使用

ARRAY_AGG

和

ARRAY_SLICE

。受到Rajat的回答

的启发

SELECT *
   ,IFF(ROW_NUMBER() OVER(PARTITION BY o_orderkey ORDER BY INDEX) <= 4, 0, 
        ROW_NUMBER() OVER(PARTITION BY o_orderkey ORDER BY INDEX)-4) AS start_index
   ,IFF(ROW_NUMBER() OVER(PARTITION BY o_orderkey ORDER BY INDEX) <= 4, 
        ROW_NUMBER() OVER(PARTITION BY o_orderkey ORDER BY INDEX),4) AS num_elem

   ,ARRAY_SLICE(
         ARRAY_AGG(o_clerk) WITHIN GROUP (ORDER BY INDEX)
                            OVER(PARTITION BY o_orderkey)
         ,start_index
         ,start_index + num_elem) 
FROM sample_data
ORDER BY O_ORDERKEY, INDEX;

输出：

---

start_index 和 num_elem 计算可以进一步简化/推广以模拟窗口滑动框架

ROWS BETWEEN PRECEDING prec AND FOLLOWING foll

。

SELECT *
   ,ROW_NUMBER() OVER(PARTITION BY o_orderkey ORDER BY INDEX) AS rn
   ,3 AS prec
   ,0 AS foll
   ,ARRAY_SLICE(
         ARRAY_AGG(o_clerk) WITHIN GROUP (ORDER BY INDEX) 
                            OVER(PARTITION BY o_orderkey)
         ,IFF(rn <= prec+1, 0, rn-(prec+1))
         ,IFF(rn <= prec+1, 0, rn-(prec+1)) + IFF(rn <= prec+1, rn+foll,prec+1+foll)
   ) 
FROM sample_data
ORDER BY O_ORDERKEY, INDEX;

显示 ARRAY_SLICE 的每个参数以进行调试：

Answer 3

由于我们无法通过

ARRAY_AGG()

获得这些结果，我们可以通过自连接。

首先我们给每一行一个行号，然后我们选择其编号和+3之间的所有内容：

with numbered as (
    select o_orderkey, o_clerk, index
    from sample_data
), crossed as (
    select a.o_orderkey, a.index ai, b.index bi, b.o_clerk
    from numbered a
    join numbered b
    on a.o_orderkey = b.o_orderkey
    and a.index between b.index and b.index+3
)

select o_orderkey, array_agg(o_clerk) within group (order by bi)
from crossed
group by o_orderkey, ai
order by o_orderkey, max(bi)

请注意，对于最初的问题，我必须添加一个

index

字段 - 这样我们就可以在行内有一个明确的排序顺序。

Answer 4

从9.1.0版本开始原生支持ARRAY_AGG窗口框架：

SELECT *,ARRAY_AGG(o_clerk) OVER (PARTITION BY o_orderkey order by o_orderkey 
                                  ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) RESULT
FROM sample_data;

输出：