根据 event_time 和 is_active 列计算 ended_at 列

Question

这是一个示例 table，其中包含数据库中单个拆分测试对象的事件。拆分测试可以打开和关闭，如 is_active.

中所述

EVENT_ID	SPLIT_TEST_ID	BRANCH	IS_ACTIVE	EVENT_TIME
1	595aa50	a	TRUE	2021-11-11 22:53:08.360
2	595aa50	b	TRUE	2021-11-11 22:53:08.360
3	595aa50	a	FALSE	2021-11-11 22:34:39.235
4	595aa50	b	FALSE	2021-11-11 22:34:39.235
5	595aa50	a	TRUE	2021-11-02 23:40:27.001
6	595aa50	b	TRUE	2021-11-02 23:40:27.001
7	595aa50	a	FALSE	2021-11-02 20:54:29.620
8	595aa50	b	FALSE	2021-11-02 20:54:29.620
9	595aa50	a	TRUE	2021-11-02 20:31:08.297
10	595aa50	b	TRUE	2021-11-02 20:31:08.297
11	595aa50	c	FALSE	2021-10-05 20:33:36.394
12	595aa50	b	FALSE	2021-10-05 20:33:36.394
13	595aa50	c	TRUE	2021-09-15 21:33:58.856
14	595aa50	b	TRUE	2021-09-15 21:33:58.856
15	595aa50	c	FALSE	2021-09-08 18:42:35.728
16	595aa50	b	FALSE	2021-09-08 18:42:35.728
17	595aa50	c	TRUE	2021-09-01 23:09:15.596
18	595aa50	b	TRUE	2021-09-01 23:09:15.596

我正在尝试转换 table 以便每次测试运行时每个分支都有一行，其中 started_at 和 ended_at值而不是 event_time。如果测试还没有关闭，ended_at 的值应该是 null.

这是我正在寻找的结果：

SPLIT_TEST_ID	BRANCH	STARTED_AT	ENDED_AT
595aa50	a	2021-11-11 22:53:08.360	null
595aa50	b	2021-11-11 22:53:08.360	null
595aa50	a	2021-11-02 23:40:27.001	2021-11-11 22:34:39.235
595aa50	b	2021-11-02 23:40:27.001	2021-11-11 22:34:39.235
595aa50	a	2021-11-02 20:31:08.297	2021-11-02 20:54:29.620
595aa50	b	2021-11-02 20:31:08.297	2021-11-02 20:54:29.620
595aa50	c	2021-09-15 21:33:58.856	2021-10-05 20:33:36.394
595aa50	b	2021-09-15 21:33:58.856	2021-10-05 20:33:36.394
595aa50	c	2021-09-01 23:09:15.596	2021-09-08 18:42:35.728
595aa50	b	2021-09-01 23:09:15.596	2021-09-08 18:42:35.728

我正在努力根据 is_active 拉取 ended_at。我尝试过

select 
    split_test_id, 
    branch, 
    event_time as started_at,
    last_value(started_at) over (partition by split_test_id, branch, is_active order by event_time) as ended_at
from example
where is_active = true

但我知道它们在逻辑上是有缺陷的。

SQL 创建上面的例子 table:

drop table if exists example;

create table example (
    event_id varchar, -- unique
    split_test_id varchar, 
    branch varchar,
    is_active boolean,
    started_at timestamp_ntz
);
 
insert into example values
    ('1', '595aa50', 'a', true, '2021-11-11 22:53:08.360'),
    ('2', '595aa50', 'b', true, '2021-11-11 22:53:08.360'),
    ('3', '595aa50', 'a', false, '2021-11-11 22:34:39.235'),
    ('4', '595aa50', 'b', false, '2021-11-11 22:34:39.235'),
    ('5', '595aa50', 'a', true, '2021-11-02 23:40:27.001'),
    ('6', '595aa50', 'b', true, '2021-11-02 23:40:27.001'),
    ('7', '595aa50', 'a', false, '2021-11-02 20:54:29.620'),
    ('8', '595aa50', 'b', false, '2021-11-02 20:54:29.620'),
    ('9', '595aa50', 'a', true, '2021-11-02 20:31:08.297'),
    ('10', '595aa50', 'b', true, '2021-11-02 20:31:08.297'),
    ('11', '595aa50', 'c', false, '2021-10-05 20:33:36.394'),
    ('12', '595aa50', 'b', false, '2021-10-05 20:33:36.394'),
    ('13', '595aa50', 'c', true, '2021-09-15 21:33:58.856'),
    ('14', '595aa50', 'b', true, '2021-09-15 21:33:58.856'),
    ('15', '595aa50', 'c', false, '2021-09-08 18:42:35.728'),
    ('16', '595aa50', 'b', false, '2021-09-08 18:42:35.728'),
    ('17', '595aa50', 'c', true, '2021-09-01 23:09:15.596'),
    ('18', '595aa50', 'b', true, '2021-09-01 23:09:15.596');

select 
    *
from example;

Answer 1

我在重新阅读问题时弄明白了。

with base as (
    select 
        *,
        lag(event_time, 1) over (partition by split_test_id, branch order by event_time desc) as ended_at
    from example
)

select 
    split_test_id, 
    branch, 
    event_time as started_at,
    ended_at
from base 
where is_active = true
order by started_at desc

根据 event_time 和 is_active 列计算 ended_at 列

Calculating ended_at column based on event_time and is_active columns

sql

window-functions