SQL 中连续行的 GROUP BY

Question

给出下表：

ID   State  Date
12   1      2009-07-16 10:00
45   2      2009-07-16 13:00
67   2      2009-07-16 14:40
77   1      2009-07-16 15:00
89   1      2009-07-16 15:30
99   1      2009-07-16 16:00

问题：
如何按“状态”字段进行分组，同时仍然保持状态更改之间的边界？

SELECT MIN(ID) AS ID, State, MIN(Date) AS Date, COUNT(ID) AS Count
FROM table GROUP BY State

结果如下：

ID   State  Date              Count
12   1      2009-07-16 10:00  4
45   2      2009-07-16 13:00  2

但这就是期望的输出：

ID   State  Date              Count
12   1      2009-07-16 10:00  1
45   2      2009-07-16 13:00  2
77   1      2009-07-16 15:00  3

Is this possible in SQL? I didn't find a solution so far...

Answer 1

SELECT  MIN(id) AS id, MIN(ts) AS ts, MIN(state) AS state, COUNT(*) cnt
FROM    (
        SELECT  @r := @r + (@state != state) AS gn,
                @state := state AS sn,
                s.*
        FROM    (
                SELECT  @r := 0,
                        @state := 0
                ) vars,
                t_state s
        ORDER BY
                ts
        ) q
GROUP BY
        gn

用于测试的表创建脚本：

CREATE TABLE t_state (id INT NOT NULL PRIMARY KEY, state INT NOT NULL, ts DATETIME NOT NULL);

INSERT
INTO  t_state
VALUES
(12,   1,      '2009-07-16 10:00'),
(45,   2,      '2009-07-16 13:00'),
(67,   2,      '2009-07-16 14:40'),
(77,   1,      '2009-07-16 15:00'),
(89,   1,      '2009-07-16 15:30'),
(99,   1,      '2009-07-16 16:00');

Answer 2

这是在 MSSQL 服务器上使用 CTE 执行此操作的方法

-- DROP TABLE MyLog
CREATE TABLE MyLog(
        ID          INT PRIMARY KEY
        , State     INT
        , Date      DATETIME
        )
INSERT MyLog
SELECT 12, 1, '2009-07-16 10:00' UNION ALL
SELECT 45, 2, '2009-07-16 13:00' UNION ALL
SELECT 67, 2, '2009-07-16 14:40' UNION ALL
SELECT 77, 1, '2009-07-16 15:00' UNION ALL
SELECT 89, 1, '2009-07-16 15:30' UNION ALL
SELECT 99, 1, '2009-07-16 16:00'

;WITH   CTE
AS      (
        SELECT  ROW_NUMBER() OVER(ORDER BY ID) AS RowNo
                , *
        FROM    MyLog
        )
, MyLogGroup
AS      (
        SELECT  l.*
                , ( SELECT  MAX(ID)
                    FROM    CTE c
                    WHERE   NOT EXISTS (SELECT * FROM CTE
                                        WHERE RowNo = c.RowNo-1 AND State = c.State)
                            AND c.ID <= l.ID) AS GroupID
        FROM    MyLog l
        )
SELECT  *
FROM    MyLogGroup

Answer 3

这里是一个更长的描述 Quassnoi 提供的解决方案如何工作

Answer 4

我可能在这里说的是显而易见的事情，但如果您愿意使用 Transact-SQL，您可以迭代表的行并构建您自己的结果集，这可能看起来很麻烦，但它肯定会起作用。 迭代可以不使用游标来完成。

Answer 5

我为 BigQuery 创建了一个解决方案：

WITH offset_state_table AS (
  SELECT
    id,
    date,
    state,
    LEAD(state)
      OVER(PARTITION BY id ORDER BY date ASC) AS offset_state,
  FROM
    `my_project.my_dataset.my_table`
),
grouped_table AS (
  SELECT
    id,
    date,
    state,
  FROM
    offset_state_table
  WHERE
    offset_state != state
  OR
    offset_state IS NULL
)
SELECT
  id,
  state,
  date AS start_date,
  LEAD(date)
    OVER(PARTITION BY id ORDER BY date ASC) AS end_date
FROM
  grouped_table