将一系列事件分组到一行中，并带有开始和结束时间

Question

这个问题是关于 SQL（特别是使用 Presto）。

我遇到了代理与任务交互（评论、获取所有权、添加标签等）的情况，我试图将每一系列交互分组到一行中，并汇总开始和结束时间。

基本上，表格目前看起来像这样：

Task Name	Agent Name	Event Type	Time
Computer Repair Help	Dave	Claim	2:35PM
Computer Repair Help	Dave	Comment	2:37PM
Computer Repair Help	Dave	Tag	2:39PM
Computer Repair Help	Dave	Pass	2:45PM
Computer Repair Help	Susie	Claim	2:56PM
Computer Repair Help	Susie	Tag	2:58PM
Computer Repair Help	Susie	Pass	3:02PM
Computer Repair Help	Dave	Claim	3:05PM
Computer Repair Help	Dave	Comment	3:08PM
Computer Repair Help	Dave	Comment	3:11PM
Computer Repair Help	Dave	Close	3:16PM

但我希望得到这样的东西：

Task Name	Agent Name	Start Time	End Time
Computer Repair Help	Dave	2:35PM	2:45PM
Computer Repair Help	Susie	2:56PM	3:02PM
Computer Repair Help	Dave	3:05PM	3:16PM

我尝试按任务名称和代理名称进行分区以获得每次交互的第一行和最后一行，但主要问题是，如果代理有两个交互窗口，它会给我所有交互的总范围而不是像示例中那样分开。不确定如何告诉查询在每次代理名称更改后将其视为新交互。

任何帮助都将不胜感激......几周以来我一直在为此苦苦挣扎，似乎必须有一些我没有看到的简单解决方案。谢谢！

Answer 1

您可以使用MATCH_RECOGNIZE 解决此问题，Trino (formerly known as Presto SQL) 的最新版本中提供了此功能：

WITH data(task, agent, event, ts) AS (
    VALUES ('Computer Repair Help', 'Dave', 'Claim', '2:35PM'),
           ('Computer Repair Help', 'Dave', 'Comment', '2:37PM'),
           ('Computer Repair Help', 'Dave', 'Tag', '2:39PM'),
           ('Computer Repair Help', 'Dave', 'Pass', '2:45PM'),
           ('Computer Repair Help', 'Susie', 'Claim', '2:56PM'),
           ('Computer Repair Help', 'Susie', 'Tag', '2:58PM'),
           ('Computer Repair Help', 'Susie', 'Pass', '3:02PM'),
           ('Computer Repair Help', 'Dave', 'Claim', '3:05PM'),
           ('Computer Repair Help', 'Dave', 'Comment', '3:08PM'),
           ('Computer Repair Help', 'Dave', 'Comment', '3:11PM'),
           ('Computer Repair Help', 'Dave', 'Close', '3:16PM')
)
SELECT task, agent, start_time, end_time
FROM data MATCH_RECOGNIZE (
        PARTITION BY task
        ORDER BY ts
        MEASURES
            FIRST(M.agent) AS agent,
            FIRST(M.ts) AS start_time,
            LAST(M.ts) AS end_time
        ONE ROW PER MATCH
        AFTER MATCH SKIP PAST LAST ROW
        PATTERN (M+)
        DEFINE M AS agent = FIRST(agent)
    )

=>

         task         | agent | start_time | end_time
----------------------+-------+------------+----------
 Computer Repair Help | Dave  | 2:35PM     | 2:45PM
 Computer Repair Help | Susie | 2:56PM     | 3:02PM
 Computer Repair Help | Dave  | 3:05PM     | 3:16PM
(3 rows)

基本思路是：

• 按任务对数据集进行分区，按时间排序
• 匹配代理等于序列中第一行代理的行序列
• 提取序列中第一个和最后一个事件的时间

您可以在此处找到有关 MATCH_RECOGNIZE 的更多详细信息：

• 文档：https://trino.io/docs/current/sql/match-recognize.html
• 博文：https://trino.io/blog/2021/05/19/row_pattern_matching.html
• 播客：https://trino.io/episodes/23.html

Answer 2

这可以作为“差距和孤岛”问题来解决。根据您的数据集，我假设每组新的交互都以 Claim 开头。

因此，CTE（也可以重写为子查询）使用SUM 窗口函数在案例表达式的帮助下创建组。最终投影使用此GroupNum（其中RANGE UNBOUNDED PRECEDING 是可选的）、任务和事件类型对数据进行分组，以便为​​每个组找到StartTime 作为MIN(Time) 和EndTime 作为MAX(Time)。

您可以尝试以下方法：

WITH CTE AS (
    SELECT
        *,
        SUM(
            CASE
                WHEN EventType='Claim' THEN 1
                ELSE 0
            END
        ) OVER (
            PARTITION BY TaskName, AgentName 
            ORDER BY Time 
            RANGE UNBOUNDED PRECEDING
        ) as GroupNum
    FROM
        my_table
)
SELECT
    TaskName,
    AgentName,
    MIN(Time) as StartTime,
    MAX(Time) as EndTime
FROM
    CTE
GROUP BY
    TaskName,
    AgentName,
    GroupNum

TaskName	AgentName	StartTime	EndTime
Computer Repair Help	Dave	2:35PM	2:45PM
Computer Repair Help	Susie	2:56PM	3:02PM
Computer Repair Help	Dave	3:05PM	3:16PM

---

View on DB Fiddle

让我知道这是否适合你。

Answer 3

这是间隙和孤岛问题的变体。使用lag()，您可以找到值变化的位置，然后进行累积求和以分配组。

WITH dataset AS (
    SELECT * FROM (VALUES   
('Computer Repair Help' ,'Dave','   Claim'  ,'2:35PM'),
('Computer Repair Help' ,'Dave','   Comment'    ,'2:37PM'),
('Computer Repair Help' ,'Dave','   Tag',   '2:39PM'),
('Computer Repair Help' ,'Dave','   Pass'   ,'2:45PM'),
('Computer Repair Help' ,'Susie','  Claim'  ,'2:56PM'),
('Computer Repair Help' ,'Susie','  Tag','2:58PM'),
('Computer Repair Help' ,'Susie','  Pass'   ,'3:02PM'),
('Computer Repair Help' ,'Dave  ','Claim'   ,'3:05PM'),
('Computer Repair Help' ,'Dave  ','Comment' ,'3:08PM'),
('Computer Repair Help' ,'Dave  ','Comment' ,'3:11PM'),
('Computer Repair Help' ,'Dave  ','Close'   ,'3:16PM')
                   
 ) AS t (task, name, event_type, time))


select task, name, min(time) as start_time, max(time) as end_time
from (
         select *,
                sum(case when name = prev_id then 0 else 1 end) over (partition by task, name order by time) as grp
         from (
                  select task, name, time,
                         lag(name) over (order by time) as prev_id
                  from dataset
              )
     )
group by task, name, grp

输出：

task	name	start_time	end_time
Computer Repair Help	Dave	2:35PM	2:45PM
Computer Repair Help	Susie	2:56PM	3:02PM
Computer Repair Help	Dave	3:05PM	3:16PM

请注意，您可能需要将时间解析为实际时间。