为了将这个问题放在上下文中，我试图根据事件日志计算“应用程序时间”。

假设下表:

user_id   event_time
2         2012-05-09 07:03:38
3         2012-05-09 07:03:42
4         2012-05-09 07:03:43
2         2012-05-09 07:03:44
2         2012-05-09 07:03:45
4         2012-05-09 07:03:52
2         2012-05-09 07:06:30

我想从彼此相差 2 分钟以内(并按用户分组)的一组时间戳中获取最高和最低 event_time 之间的差异。如果时间戳超出集合的 2 分钟间隔，则应将其视为另一集合的一部分。

期望的输出:

user_id  seconds_interval
2        7     (because 07:03:45 - 07:03:38 is 7 seconds)
3        0     (because 07:03:42)
4        9     (because 07:03:52 - 2012-05-09 07:03:43)
2        0     (because 07:06:30 is outside 2 min interval of 1st user_id=2 set)

这是我尝试过的方法，虽然我不能在 seconds_interval 上分组(即使我可以，我不确定这是正确的方向):

SELECT (max(tr.event_time)-min(tr.event_time)) as seconds_interval
FROM some_table tr
INNER JOIN TrackingRaw tr2 ON (tr.event_time BETWEEN 
   tr2.event_time - INTERVAL 2 MINUTE AND tr2.event_time + INTERVAL 2 MINUTE) 
GROUP BY seconds_interval

最佳答案

我认为没有一种非常直接的方法可以查询您现有的表来生成您想要的数据。但是，您可以维护第二个用户 session 表(当然这有一个缺点，如果您以后想要一个使用不同 session 超时期限的报告，您将需要从头开始重新填充该表):

CREATE TABLE Sessions (
  user_id INT,
  session_start TIMESTAMP,
  session_end   TIMESTAMP,
  PRIMARY KEY (user_id, session_start),
  FOREIGN KEY (user_id, session_start) REFERENCES TrackingRaw(user_id, event_time),
  FOREIGN KEY (user_id, session_end  ) REFERENCES TrackingRaw(user_id, event_time)
);

您可以使用使用 INSERT ... SELECT ... ON DUPLICATE KEY UPDATE 的触发器自动填充/更新此类表:

CREATE TRIGGER after_insert_TrackingRaw AFTER INSERT ON TrackingRaw FOR EACH ROW
  INSERT INTO Sessions (user_id, session_start, session_end)
    SELECT NEW.user_id,
           IFNULL(MAX(session_start), NEW.event_time),
           NEW.event_time
    FROM   Sessions
    WHERE  user_id = NEW.user_id
       AND session_end >= NEW.event_time - INTERVAL 2 MINUTE
  ON DUPLICATE KEY UPDATE
    session_start = session_start,
    session_end   = NEW.event_time;

然后，获取你想要的查询结果:

SELECT user_id, session_end - session_start AS seconds_interval FROM Sessions;

查看 sqlfiddle .

更新

经过进一步思考，你当然可以在存储过程中构建这样一个Sessions表:

CREATE PROCEDURE getSessions(IN secs INT) READS SQL DATA BEGIN
  DECLARE no_more_rows BOOLEAN;
  DECLARE cur CURSOR FOR
    SELECT user_id, event_time FROM TrackingRaw ORDER BY event_time ASC;
  DECLARE CONTINUE HANDLER FOR NOT FOUND SET no_more_rows = TRUE;

  DROP   TEMPORARY TABLE IF EXISTS Sessions;
  CREATE TEMPORARY TABLE Sessions (
    user_id INT,
    session_start TIMESTAMP,
    session_end   TIMESTAMP,
    PRIMARY KEY(user_id,session_start),
    FOREIGN KEY(user_id,session_start) REFERENCES TrackingRaw(user_id,event_time),
    FOREIGN KEY(user_id,session_end  ) REFERENCES TrackingRaw(user_id,event_time)
  );

  OPEN cur;
  the_loop: LOOP
    FETCH cur INTO @u, @t;
    IF no_more_rows THEN
      CLOSE cur;
      LEAVE the_loop;
    END IF;

    INSERT INTO Sessions
      SELECT @u, IFNULL(MAX(session_start), @t), @t
      FROM   Sessions
      WHERE  user_id = @u AND session_end >= @t - secs
    ON DUPLICATE KEY UPDATE
      session_start = session_start, session_end = @t
  END LOOP the_loop;

  DEALLOCATE PREPARE stmt;
  SELECT user_id, session_end - session_start AS seconds_interval FROM Sessions;
  DROP TEMPORARY TABLE Sessions;
END;;

然后获取输出:

CALL getSessions(120); -- for a 2 minute (120 second) timeout

关于mysql - 按时间戳之间的间隔对时间戳进行分组，然后从 MySQL 组计算，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/11255858/