mysql - 表中的哪一行（按某列排序）对应于另一个表中的一行？

Question

我有两个表，如下（从实际简化）：

mysql> desc small_table;
+-----------------+---------------+------+-----+-- --------+-------+
| 领域 | 类型 | 空 | 钥匙 | 默认 | 额外 |
+-----------------+---------------+------+-----+-- --------+-------+
| 事件时间 | 日期时间 | 否 | | 空 | |
| 用户 ID | 字符(15) | 否 | | 空 | |
| 其他数据 | 整数(11) | 否 | 穆尔 | 空 | |
+-----------------+---------------+------+-----+-- --------+-------+
3 行一组（0.00 秒）

mysql> desc large_table;
+-----------------+---------------+------+-----+-- --------+-------+
| 领域 | 类型 | 空 | 钥匙 | 默认 | 额外 |
+-----------------+---------------+------+-----+-- --------+-------+
| 事件时间 | 日期时间 | 否 | | 空 | |
| 用户 ID | 字符(15) | 否 | | 空 | |
| 其他数据 | 整数(11) | 否 | | 空 | |
+-----------------+---------------+------+-----+-- --------+-------+
3 行一组（0.00 秒）

现在，small_table嗯，很小：对于每个user_id通常只有一排（尽管有时会有更多）。large_table另一方面，在中，每个都出现user_id了无数次。

mysql> 从 small_table\G 中选择 count(1)
****************************** 1. 行 ************************ *******
计数（1）：20182
一组中的 1 行（0.00 秒）


mysql> 从 large_table\G 中选择 count(1)
****************************** 1. 行 ************************ *******
计数（1）：2870522
一组中的 1 行（0.00 秒）

但是，这很重要，对于中的每一行small_table，至少有一行large_table具有相同user_id、相同other_data和相似event_time（例如，几分钟内相同）。

我想知道是否small_table有对应于第一行、第二行或相同和相似的任何不同行的^行。也就是说，我想：large_tableuser_idevent_time

对于每个，按顺序排列user_id的不同行的计数，但仅限于例如三个小时内；也就是说，我只寻找相隔三小时之内的行数；和large_tableevent_timeevent_timeevent_time
对于每个这样的不同行集合，该列表中哪一行的标识（按顺序event_time）在small_table.

我似乎无法编写一个可以完成第一步的查询，更不用说一个可以完成第二步的查询了，并且希望得到任何指导。

score 1 · Accepted Answer

为此所需的 SQL 是残酷的；它会给你的优化器一个非常严肃的锻炼。

从问题和问题之后的评论来看，如果它们都落在相邻事件之间的某个固定间隔内，则希望将大表中给定用户 ID 的事件序列视为“连续的”。例如，固定间隔为 3 小时。我正在为 IBM Informix Dynamic Server 编码（为了争论，版本 11.70，但 11.50 也可以正常工作）。这意味着我需要解释一个特殊的符号。具体来说，符号3 UNITS HOUR表示3小时的间隔；它也可以用INTERVAL(3) HOUR TO HOURSQL 的 Informix 方言或INTERVAL '3' HOUR标准 SQL 编写。

生成 SQL 有几个关键技术，尤其是复杂的 SQL。一种是逐步、零碎地构建 SQL，组装最终结果。另一个是确保你清楚地说明你所追求的。

在下面的符号中，“for the same User_ID”的限定应被视为始终是表达式的一部分。

在大表中，在加入小表之前，我们要考虑三类范围。

既没有足够接近的事件之前的事件时间行也没有足够接近的事件之后的事件时间的行的条目。这是一个开始和结束时间相同的时间范围。
表中的一对条目，它们本身足够接近，但既没有事件早于足够接近的对的早期事件，也没有事件晚于足够接近的对的晚期事件，也没有事件在这对之间。
表中三个或更多条目的序列，其中有：
- 没有事件 E1 早于足够接近的最早事件
- 没有事件 E2 晚于足够接近的最新事件
- 晚于最早的事件 E3 与最早的足够接近
- 一个比最新事件更早的事件 E4 与最新事件足够接近（E4 可能与 E3 事件相同）
- 最早和最晚之间没有一对事件 E5、E6，其中 E5 和 E6 之间没有事件，但 E5 和 E6 之间的差距太大而无法计算。

从描述中可以看出，这将是一些可怕的 SQL！

注意：代码现在已经过测试；一些（主要是小的）改变是必要的。一项不必要的小改动是在中间查询中添加了 ORDER BY 子句。另一个不必要的更改是从小表中选择其他数据进行验证。此修订是在没有研究msh210发布的修订版本的情况下进行的。

另请注意，我不确定这是一个最小的公式；使用单个 SELECT 语句而不是三个 SELECT 语句的 UNION 对所有范围进行分类可能是可行的（如果是这样的话会很好）。

单例范围

-- Ranges of exactly 1 event
SELECT lt1.user_id, lt1.event_time AS min_time, lt1.event_time AS max_time
  FROM Large_Table AS lt1
 WHERE NOT EXISTS -- an earlier event that is close enough
       (SELECT *
          FROM Large_Table AS lt3
         WHERE lt1.user_id = lt3.user_id
           AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
           AND lt3.event_time < lt1.event_time
       )
   AND NOT EXISTS -- a later event that is close enough
       (SELECT *
          FROM Large_Table AS lt4
         WHERE lt1.user_id = lt4.user_id
           AND lt4.event_time > lt1.event_time
           AND lt4.event_time < lt1.event_time + 3 UNITS HOUR
       )
ORDER BY User_ID, Min_Time;

道尔顿山脉

-- Ranges of exactly 2 events
SELECT lt1.user_id, lt1.event_time AS min_time, lt2.event_time AS max_time
  FROM Large_Table AS lt1
  JOIN Large_Table AS lt2
    ON lt1.user_id = lt2.user_id
   AND lt1.event_time < lt2.event_time
   AND lt2.event_time < lt1.event_time + 3 UNITS HOUR
 WHERE NOT EXISTS -- an earlier event that is close enough
       (SELECT *
          FROM Large_Table AS lt3
         WHERE lt1.user_id = lt3.user_id
           AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
           AND lt3.event_time < lt1.event_time
       )
   AND NOT EXISTS -- a later event that is close enough
       (SELECT *
          FROM Large_Table AS lt4
         WHERE lt1.user_id = lt4.user_id
           AND lt4.event_time > lt2.event_time
           AND lt4.event_time < lt2.event_time + 3 UNITS HOUR
       )
   AND NOT EXISTS -- any event in between
       (SELECT *
          FROM Large_Table AS lt5
         WHERE lt1.user_id = lt5.user_id
           AND lt5.event_time > lt1.event_time
           AND lt5.event_time < lt2.event_time
       )
ORDER BY User_ID, Min_Time;

在外部 WHERE 子句中添加了 3 小时标准。

多个事件范围

-- Ranges of 3 or more events
SELECT lt1.user_id, lt1.event_time AS min_time, lt2.event_time AS max_time
  FROM Large_Table AS lt1
  JOIN Large_Table AS lt2
    ON lt1.user_id = lt2.user_id
   AND lt1.event_time < lt2.event_time
 WHERE NOT EXISTS -- an earlier event that is close enough
       (SELECT *
          FROM Large_Table AS lt3
         WHERE lt1.user_id = lt3.user_id
           AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
           AND lt3.event_time < lt1.event_time
       )
   AND NOT EXISTS -- a later event that is close enough
       (SELECT *
          FROM Large_Table AS lt4
         WHERE lt1.user_id = lt4.user_id
           AND lt4.event_time > lt2.event_time
           AND lt4.event_time < lt2.event_time + 3 UNITS HOUR
       )
   AND NOT EXISTS -- a gap that's too big in the events between first and last
       (SELECT *
          FROM Large_Table AS lt5 -- E5 before E6
          JOIN Large_Table AS lt6
            ON lt5.user_id = lt6.user_id
           AND lt5.event_time < lt6.event_time
         WHERE lt1.user_id = lt5.user_id
           AND lt6.event_time < lt2.event_time
           AND lt5.event_time > lt1.event_time
           AND (lt6.event_time - lt5.event_time) > 3 UNITS HOUR
           AND NOT EXISTS -- an event in between these two
               (SELECT *
                  FROM Large_Table AS lt9
                 WHERE lt5.user_id = lt9.user_id
                   AND lt9.event_time > lt5.event_time
                   AND lt9.event_time < lt6.event_time
               )
       )
   AND EXISTS -- an event close enough after the start
       (SELECT *
          FROM Large_Table AS lt7
         WHERE lt1.user_id = lt7.user_id
           AND lt1.event_time < lt7.event_time
           AND lt7.event_time < lt1.event_time + 3 UNITS HOUR
           AND lt7.event_time < lt2.event_time
       )
   AND EXISTS -- an event close enough before the end
       (SELECT *
          FROM Large_Table AS lt8
         WHERE lt2.user_id = lt8.user_id
           AND lt2.event_time > lt8.event_time
           AND lt8.event_time > lt2.event_time - 3 UNITS HOUR
           AND lt8.event_time > lt1.event_time
       )
ORDER BY User_ID, Min_Time;

在“big gaps”子查询中添加了省略的嵌套 NOT EXISTS 子句。

大表中的所有范围

显然，最后一个表中的完整范围列表是上述三个查询的并集。

查询因为不够有趣而被删除。它只是上面单独查询的 3 路 UNION。

最终查询

最后的查询在与小表中的条目足够接近的可怕的 3 部分 UNION 的结果中找到范围（如果有）。例如，小表中的单个条目可能落在 13:00，大表中可能有一个范围在 11:00 结束，另一个从 15:00 开始。大表中的两个范围是分开的（它们之间的间隔是 4 小时），但小表中的条目足够接近，可以计算。[测试涵盖了这种情况。]

SELECT S.User_id, S.Event_Time, L.Min_Time, L.Max_Time, S.Other_Data
  FROM Small_Table AS S
  JOIN (
        -- Ranges of exactly 1 event
        SELECT lt1.user_id, lt1.event_time AS min_time, lt1.event_time AS max_time
          FROM Large_Table AS lt1
         WHERE NOT EXISTS -- an earlier event that is close enough
               (SELECT *
                  FROM Large_Table AS lt3
                 WHERE lt1.user_id = lt3.user_id
                   AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
                   AND lt3.event_time < lt1.event_time
               )
           AND NOT EXISTS -- a later event that is close enough
               (SELECT *
                  FROM Large_Table AS lt4
                 WHERE lt1.user_id = lt4.user_id
                   AND lt4.event_time > lt1.event_time
                   AND lt4.event_time < lt1.event_time + 3 UNITS HOUR
               )
        UNION
        -- Ranges of exactly 2 events
        SELECT lt1.user_id, lt1.event_time AS min_time, lt2.event_time AS max_time
          FROM Large_Table AS lt1
          JOIN Large_Table AS lt2
            ON lt1.user_id = lt2.user_id
           AND lt1.event_time < lt2.event_time
           AND lt2.event_time < lt1.event_time + 3 UNITS HOUR
         WHERE NOT EXISTS -- an earlier event that is close enough
               (SELECT *
                  FROM Large_Table AS lt3
                 WHERE lt1.user_id = lt3.user_id
                   AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
                   AND lt3.event_time < lt1.event_time
               )
           AND NOT EXISTS -- a later event that is close enough
               (SELECT *
                  FROM Large_Table AS lt4
                 WHERE lt1.user_id = lt4.user_id
                   AND lt4.event_time > lt2.event_time
                   AND lt4.event_time < lt2.event_time + 3 UNITS HOUR
               )
           AND NOT EXISTS -- any event in between
               (SELECT *
                  FROM Large_Table AS lt5
                 WHERE lt1.user_id = lt5.user_id
                   AND lt5.event_time > lt1.event_time
                   AND lt5.event_time < lt2.event_time
               )
        UNION
        -- Ranges of 3 or more events
        SELECT lt1.user_id, lt1.event_time AS min_time, lt2.event_time AS max_time
          FROM Large_Table AS lt1
          JOIN Large_Table AS lt2
            ON lt1.user_id = lt2.user_id
           AND lt1.event_time < lt2.event_time
         WHERE NOT EXISTS -- an earlier event that is close enough
               (SELECT *
                  FROM Large_Table AS lt3
                 WHERE lt1.user_id = lt3.user_id
                   AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
                   AND lt3.event_time < lt1.event_time
               )
           AND NOT EXISTS -- a later event that is close enough
               (SELECT *
                  FROM Large_Table AS lt4
                 WHERE lt1.user_id = lt4.user_id
                   AND lt4.event_time > lt2.event_time
                   AND lt4.event_time < lt2.event_time + 3 UNITS HOUR
               )
           AND NOT EXISTS -- a gap that's too big in the events between first and last
               (SELECT *
                  FROM Large_Table AS lt5 -- E5 before E6
                  JOIN Large_Table AS lt6
                    ON lt5.user_id = lt6.user_id
                   AND lt5.event_time < lt6.event_time
                 WHERE lt1.user_id = lt5.user_id
                   AND lt6.event_time < lt2.event_time
                   AND lt5.event_time > lt1.event_time
                   AND (lt6.event_time - lt5.event_time) > 3 UNITS HOUR
                   AND NOT EXISTS -- an event in between these two
                       (SELECT *
                          FROM Large_Table AS lt9
                         WHERE lt5.user_id = lt9.user_id
                           AND lt9.event_time > lt5.event_time
                           AND lt9.event_time < lt6.event_time
                       )
               )
           AND EXISTS -- an event close enough after the start
               (SELECT *
                  FROM Large_Table AS lt7
                 WHERE lt1.user_id = lt7.user_id
                   AND lt1.event_time < lt7.event_time
                   AND lt7.event_time < lt1.event_time + 3 UNITS HOUR
                   AND lt7.event_time < lt2.event_time
               )
           AND EXISTS -- an event close enough before the end
               (SELECT *
                  FROM Large_Table AS lt8
                 WHERE lt2.user_id = lt8.user_id
                   AND lt2.event_time > lt8.event_time
                   AND lt8.event_time > lt2.event_time - 3 UNITS HOUR
                   AND lt8.event_time > lt1.event_time
               )
       ) AS L
    ON S.User_ID = L.User_ID
 WHERE S.Event_Time > L.Min_Time - 3 UNITS HOUR
   AND S.Event_Time < L.Max_Time + 3 UNITS HOUR
ORDER BY User_ID, Event_Time, Min_Time;

~~OK - 公平警告；SQL 实际上并没有靠近 SQL DBMS。~~

代码现在已经过测试。极小的机会实际上是零。有一个语法错误和几个或多或少的小问题需要解决。

在设计了测试数据后，我分阶段进行了实验。我在验证和修复查询时使用了“Alpha”数据（见下文），并添加了 Beta 数据只是为了确保不同的 User_ID 值之间没有串扰。

我使用了显式<和>操作而不是BETWEEN ... AND排除端点；如果您希望恰好相隔 3 小时的事件算作“足够接近”，那么您需要查看每个不等式，可能将它们更改为BETWEEN ... AND或可能仅使用>=或<=酌情使用。

有一个松散相似但相当简单的问题的答案，（a）我写的和（b）对上面的复杂处理提供了一些有用的想法（特别是，'没有事件早但足够接近'和'没有事件稍后但接近“足够”的标准。“足够接近”的标准肯定会使这个问题复杂化。

测试数据

大桌子

CREATE TABLE Large_Table
(
    Event_Time  DATETIME YEAR TO MINUTE NOT NULL,
    User_ID     CHAR(15) NOT NULL,
    Other_Data  INTEGER NOT NULL,
    PRIMARY KEY(User_ID, Event_Time)
);

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 09:15', 'Alpha',  1) { R4 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 11:15', 'Alpha',  2) { R4 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 13:15', 'Alpha',  3) { R4 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 15:15', 'Alpha',  4) { R4 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 12:17', 'Beta',   1) { R4 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 09:15', 'Alpha',  5) { R1 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 10:17', 'Beta',   2) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 09:15', 'Alpha',  6) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 11:15', 'Alpha',  7) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 10:17', 'Beta',   3) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 09:15', 'Alpha',  8) { R3 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 11:15', 'Alpha',  9) { R3 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 13:15', 'Alpha', 10) { R3 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 10:17', 'Beta',   4) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 09:15', 'Alpha', 11) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 11:15', 'Alpha', 12) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 10:17', 'Beta',   5) { R1 };
{ Probe here }
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 15:15', 'Alpha', 13) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 17:15', 'Alpha', 14) { R2 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 16:17', 'Beta',   6) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 09:15', 'Alpha', 15) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 11:15', 'Alpha', 16) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 13:15', 'Alpha', 17) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 15:15', 'Alpha', 18) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 17:15', 'Alpha', 19) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 19:15', 'Alpha', 20) { R6 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 16:17', 'Beta',   7) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 09:15', 'Alpha', 21) { R1 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 11:17', 'Beta',   8) { R1 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 12:15', 'Alpha', 22) { R1 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 13:17', 'Beta',   9) { R1 };

INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 09:15', 'Alpha', 23) { R5 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 11:15', 'Alpha', 24) { R5 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 13:15', 'Alpha', 25) { R5 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 15:15', 'Alpha', 26) { R5 };
INSERT INTO Large_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 17:15', 'Alpha', 27) { R5 };

小桌子

注意：为了测试的目的，小表实际上比大表包含更多的行。小表中 Other_Data 值大于 100 的行不应出现在结果中（也不应该出现）。这里的测试确实戳到了边缘条件。

CREATE TABLE Small_Table
(
    Event_Time  DATETIME YEAR TO MINUTE NOT NULL,
    User_ID     CHAR(15) NOT NULL,
    Other_Data  INTEGER NOT NULL,
    PRIMARY KEY(User_ID, Event_Time)
);

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 06:15', 'Alpha', 131) { XX };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 06:20', 'Alpha',  31) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 10:20', 'Alpha',  32) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 13:20', 'Alpha',  33) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 15:20', 'Alpha',  34) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 18:15', 'Alpha', 134) { XX };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 06:15', 'Alpha', 135) { XX };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 06:16', 'Alpha',  35) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 10:20', 'Alpha',  35) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 12:14', 'Alpha',  35) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 12:15', 'Alpha', 135) { XX };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 09:20', 'Alpha',  36) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 11:20', 'Alpha',  37) { YY };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 09:20', 'Alpha',  38) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 11:20', 'Alpha',  39) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 13:20', 'Alpha',  40) { YY };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 09:20', 'Alpha',  41) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 11:20', 'Alpha',  42) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 13:20', 'Alpha',  42) { 22 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 15:20', 'Alpha',  43) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 17:20', 'Alpha',  44) { YY };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 09:20', 'Alpha',  45) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 11:20', 'Alpha',  46) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 13:20', 'Alpha',  47) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 15:20', 'Alpha',  48) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 17:20', 'Alpha',  49) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 19:20', 'Alpha',  50) { YY };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 09:20', 'Alpha',  51) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 10:20', 'Alpha',  51) { 22 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 12:20', 'Alpha',  52) { YY };

INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 09:20', 'Alpha',  53) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 11:20', 'Alpha',  54) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 13:20', 'Alpha',  55) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 15:20', 'Alpha',  56) { YY };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-08 17:20', 'Alpha',  57) { YY };


INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 13:27', 'Beta',   9) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-07 11:27', 'Beta',   8) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-06 16:27', 'Beta',   7) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 16:27', 'Beta',   6) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-05 10:27', 'Beta',   5) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-04 10:27', 'Beta',   4) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-03 10:27', 'Beta',   3) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-02 10:27', 'Beta',   2) { R1 };
INSERT INTO Small_Table(Event_Time, User_ID, Other_Data) VALUES('2012-01-01 12:27', 'Beta',   1) { R4 };

最终查询结果

使用上面的数据，得到的结果是：

Alpha   2012-01-01 06:20   2012-01-01 09:15   2012-01-01 15:15   31
Alpha   2012-01-01 10:20   2012-01-01 09:15   2012-01-01 15:15   32
Alpha   2012-01-01 13:20   2012-01-01 09:15   2012-01-01 15:15   33
Alpha   2012-01-01 15:20   2012-01-01 09:15   2012-01-01 15:15   34
Alpha   2012-01-02 06:16   2012-01-02 09:15   2012-01-02 09:15   35
Alpha   2012-01-02 10:20   2012-01-02 09:15   2012-01-02 09:15   35
Alpha   2012-01-02 12:14   2012-01-02 09:15   2012-01-02 09:15   35
Alpha   2012-01-03 09:20   2012-01-03 09:15   2012-01-03 11:15   36
Alpha   2012-01-03 11:20   2012-01-03 09:15   2012-01-03 11:15   37
Alpha   2012-01-04 09:20   2012-01-04 09:15   2012-01-04 13:15   38
Alpha   2012-01-04 11:20   2012-01-04 09:15   2012-01-04 13:15   39
Alpha   2012-01-04 13:20   2012-01-04 09:15   2012-01-04 13:15   40
Alpha   2012-01-05 09:20   2012-01-05 09:15   2012-01-05 11:15   41
Alpha   2012-01-05 11:20   2012-01-05 09:15   2012-01-05 11:15   42
Alpha   2012-01-05 13:20   2012-01-05 09:15   2012-01-05 11:15   42
Alpha   2012-01-05 13:20   2012-01-05 15:15   2012-01-05 17:15   42
Alpha   2012-01-05 15:20   2012-01-05 15:15   2012-01-05 17:15   43
Alpha   2012-01-05 17:20   2012-01-05 15:15   2012-01-05 17:15   44
Alpha   2012-01-06 09:20   2012-01-06 09:15   2012-01-06 19:15   45
Alpha   2012-01-06 11:20   2012-01-06 09:15   2012-01-06 19:15   46
Alpha   2012-01-06 13:20   2012-01-06 09:15   2012-01-06 19:15   47
Alpha   2012-01-06 15:20   2012-01-06 09:15   2012-01-06 19:15   48
Alpha   2012-01-06 17:20   2012-01-06 09:15   2012-01-06 19:15   49
Alpha   2012-01-06 19:20   2012-01-06 09:15   2012-01-06 19:15   50
Alpha   2012-01-07 09:20   2012-01-07 09:15   2012-01-07 09:15   51
Alpha   2012-01-07 09:20   2012-01-07 12:15   2012-01-07 12:15   51
Alpha   2012-01-07 10:20   2012-01-07 09:15   2012-01-07 09:15   51
Alpha   2012-01-07 10:20   2012-01-07 12:15   2012-01-07 12:15   51
Alpha   2012-01-07 12:20   2012-01-07 12:15   2012-01-07 12:15   52
Alpha   2012-01-08 09:20   2012-01-08 09:15   2012-01-08 17:15   53
Alpha   2012-01-08 11:20   2012-01-08 09:15   2012-01-08 17:15   54
Alpha   2012-01-08 13:20   2012-01-08 09:15   2012-01-08 17:15   55
Alpha   2012-01-08 15:20   2012-01-08 09:15   2012-01-08 17:15   56
Alpha   2012-01-08 17:20   2012-01-08 09:15   2012-01-08 17:15   57
Beta    2012-01-01 12:27   2012-01-01 12:17   2012-01-01 12:17    1
Beta    2012-01-02 10:27   2012-01-02 10:17   2012-01-02 10:17    2
Beta    2012-01-03 10:27   2012-01-03 10:17   2012-01-03 10:17    3
Beta    2012-01-04 10:27   2012-01-04 10:17   2012-01-04 10:17    4
Beta    2012-01-05 10:27   2012-01-05 10:17   2012-01-05 10:17    5
Beta    2012-01-05 16:27   2012-01-05 16:17   2012-01-05 16:17    6
Beta    2012-01-06 16:27   2012-01-06 16:17   2012-01-06 16:17    7
Beta    2012-01-07 11:27   2012-01-07 11:17   2012-01-07 13:17    8
Beta    2012-01-07 13:27   2012-01-07 11:17   2012-01-07 13:17    9

中间结果

效果略有不同的格式。

单例范围

Alpha|2012-01-02 09:15|2012-01-02 09:15
Alpha|2012-01-07 09:15|2012-01-07 09:15
Alpha|2012-01-07 12:15|2012-01-07 12:15
Beta|2012-01-01 12:17|2012-01-01 12:17
Beta|2012-01-02 10:17|2012-01-02 10:17
Beta|2012-01-03 10:17|2012-01-03 10:17
Beta|2012-01-04 10:17|2012-01-04 10:17
Beta|2012-01-05 10:17|2012-01-05 10:17
Beta|2012-01-05 16:17|2012-01-05 16:17
Beta|2012-01-06 16:17|2012-01-06 16:17

道尔顿山脉

Alpha|2012-01-03 09:15|2012-01-03 11:15
Alpha|2012-01-05 09:15|2012-01-05 11:15
Alpha|2012-01-05 15:15|2012-01-05 17:15
Beta|2012-01-07 11:17|2012-01-07 13:17

多个事件范围

Alpha|2012-01-01 09:15|2012-01-01 15:15
Alpha|2012-01-04 09:15|2012-01-04 13:15
Alpha|2012-01-06 09:15|2012-01-06 19:15
Alpha|2012-01-08 09:15|2012-01-08 17:15

score 0 · Accepted Answer

select count(s.user_id), s.event_time, s.other_data from small_table s
where s.user_id IN (select distinct user_id from big_table where event_time between 'StartDate' and 'EndDate')
order by s.event_time

我不确定您在提到的小范围内要求什么。

还：

select * from large_table t1, large_table t2 
where t1.event_time <= date_sub(t2.event_time, INTERVAL 3 hour)

所以，试试：

  select count(s.user_id), s.event_time, s.other_data from small_table s
    where s.user_id IN ( select * from large_table t1, large_table t2 
    where t1.event_time <= date_sub(t2.event_time, INTERVAL 3 hour))
order by s.event_time

score 0 · Accepted Answer

这也许应该是对Jonathan Leffler的详细而有用的答案的评论，但是（a）它太长了，（b）它确实有助于回答我的问题，所以我将它作为答案发布。

Jonathan Leffler 的答案中标题为“Multiple Event Ranges”的代码找到了第二个实例在第一个实例之后不久，倒数第二个实例在最后一个实例之前的范围，并且没有出现大的中断，但是内部实例之间没有任何大的差距，甚至如果他们之间有其他情况。因此，例如，如果限制为 3 小时，则 1、2、4、6 和 7 处的实例将被禁止，因为 2 和 6 之间的差距。我认为正确的代码应该是（直接建立在 Jonathan Leffler 的):

SELECT lt1.user_id, lt1.event_time AS min_time, lt2.event_time AS max_time
  FROM Large_Table AS lt1
  JOIN Large_Table AS lt2
    ON lt1.user_id = lt2.user_id
   AND lt1.event_time < lt2.event_time
 WHERE NOT EXISTS -- an earlier event that is close enough
       (SELECT *
          FROM Large_Table AS lt3
         WHERE lt1.user_id = lt3.user_id
           AND lt3.event_time > lt1.event_time - 3 UNITS HOUR
           AND lt3.event_time < lt1.event_time
       )
   AND NOT EXISTS -- a later event that is close enough
       (SELECT *
          FROM Large_Table AS lt4
         WHERE lt1.user_id = lt4.user_id
           AND lt4.event_time > lt2.event_time
           AND lt4.event_time < lt2.event_time + 3 UNITS HOUR
       )
   AND NOT EXISTS -- a gap that's too big in the events between first and last
       (SELECT *
          FROM Large_Table AS lt5 -- E5 before E6
          JOIN Large_Table AS lt6
            ON lt5.user_id = lt6.user_id
           AND lt1.user_id = lt5.user_id
           AND lt5.event_time < lt6.event_time
           AND lt6.event_time <= lt2.event_time
           AND lt5.event_time >= lt1.event_time
           AND (lt6.event_time - lt5.event_time) > 3 UNITS HOUR
           and not exists (
             select * from large_table as lt9 
               where lt9.event_time > lt5.event_time
                 and lt6.event_time > lt9.event_time
             )
       )

这消除了and existsJonathan Leffler 的答案中标题为“Multiple Event Ranges”的代码中最后两个 s 的需要，并且实际上消除了他的答案中对“Singleton Ranges”和“Doubleton Ranges”代码的需要。

除非我错过了什么。

mysql - 表中的哪一行（按某列排序）对应于另一个表中的一行？

3 回答 3

单例范围

道尔顿山脉

多个事件范围

大表中的所有范围

最终查询

测试数据

大桌子

小桌子

最终查询结果

中间结果

Related

Reference