SQL Presto 中的 row_number() 表达式问题

Question

样品表：

object_id	event_time	event_type	event_subtype	stage
1	2022-10-01	create	name, stage	A
1	2022-10-02	update	stage	B
1	2022-10-03	update	stage	C
1	2022-10-04	update	stage	A
2	2022-10-01	create	name, stage	A
2	2022-10-02	update	stage	C
2	2022-10-03	update	stage	A
2	2022-10-04	update	stage	B
2	2022-10-05	update	stage	C
2	2022-10-06	update	stage	A

所以我需要的是一个根据阶段对行进行编号的列——在 object_id 到达阶段 C 之后，相同 object_id 的行号应该递增。它看起来像这样：

object_id	event_time	event_type	event_subtype	stage	row_number
1	2022-10-01	create	name, stage	A	1
1	2022-10-02	update	stage	B	1
1	2022-10-03	update	stage	C	1
1	2022-10-04	update	stage	A	2
2	2022-10-01	create	name, stage	A	1
2	2022-10-02	update	stage	C	1
2	2022-10-03	update	stage	A	2
2	2022-10-04	update	stage	B	2
2	2022-10-05	update	stage	C	2
2	2022-10-06	update	stage	A	3

该表必须按 object_id、event_time 排序。我在编写执行此操作的窗口函数时遇到了麻烦，这是我尝试过的：

row_number() over (partition by object_id, stage order by event_time)

它只是不适用于所有情况。此外，当我没有将 stage = C 定义为任何地方的分隔符时，我也很难理解这是如何工作的。有任何想法吗？

谢谢！

Answer 1

给你最终“正确”的顺序不是 row_number 的问题。

对于你雾使用ORDER BY

SELECT
"object_id", "event_time", "event_type", "event_subtype", "stage",
ROW_NUMBER() OVER(PARTITION BY "object_id","stage" ORDER BY "event_time") rn
  FROM tab1
  ORDER BY "object_id",rn,"stage"

object_id	event_time	event_type	event_subtype	stage	rn
1	2022-10-01	create	name, stage	A	1
1	2022-10-02	update	stage	B	1
1	2022-10-03	update	stage	C	1
1	2022-10-04	update	stage	A	2
2	2022-10-01	create	name, stage	A	1
2	2022-10-04	update	stage	B	1
2	2022-10-02	update	stage	C	1
2	2022-10-03	update	stage	A	2
2	2022-10-05	update	stage	C	2
2	2022-10-06	update	stage	A	3

Answer 2

我建议对 stage 的先前值使用总和：

-- sample data
with dataset(object_id, event_time, event_type, event_subtype, stage) as (
    values    (1, '2022-10-01', 'create',   'name, stage', 'A'),
    (1, '2022-10-02', 'update', 'stage', 'B'),
    (1, '2022-10-03', 'update', 'stage', 'C'),
    (1, '2022-10-04', 'update', 'stage', 'A'),
    (2, '2022-10-01', 'create', 'name, stage',' A'),
    (2, '2022-10-02', 'update', 'stage', 'C'),
    (2, '2022-10-03', 'update', 'stage', 'A'),
    (2, '2022-10-04', 'update', 'stage', 'B'),
    (2, '2022-10-05', 'update', 'stage', 'C'),
    (2, '2022-10-06', 'update', 'stage', 'A')
)

-- query
select object_id,
       event_time,
       event_type,
       event_subtype,
       stage,
       1 + sum(counter) over (partition by object_id order by event_time) as num
from (select *,
             if(lag(stage) over (partition by object_id order by event_time) = 'C', 1, 0) counter
      from dataset);

输出：

object_id	event_time	event_type	event_subtype	stage	num
1	2022-10-01	create	name, stage	A	1
1	2022-10-02	update	stage	B	1
1	2022-10-03	update	stage	C	1
1	2022-10-04	update	stage	A	2
2	2022-10-01	create	name, stage	A	1
2	2022-10-02	update	stage	C	1
2	2022-10-03	update	stage	A	2
2	2022-10-04	update	stage	B	2
2	2022-10-05	update	stage	C	2
2	2022-10-06	update	stage	A	3