mysql - 如何有效地选择基于 SQL 中不同时间戳计算的多个总和的平均总和？

Question

我有一个如下所示的数据库表:

id | macaddr | load | timestamp
=========================================
 1 | 0011111 |   17 | 2012-02-07 10:00:00
 1 | 0011111 |    6 | 2012-02-07 12:00:00
 2 | 0022222 |    3 | 2012-02-07 12:00:03
 3 | 0033333 |    9 | 2012-02-07 12:00:04
 4 | 0022222 |    4 | 2012-02-07 12:00:06
 5 | 0033333 |    8 | 2012-02-07 12:00:10
...

现在，我想计算不同时间段内所有设备(= mac 地址)的平均负载，例如，今天、昨天、本周、本月。

可以通过首先找出不同时间点(样本日期)的总负荷总和，然后计算这些样本日期的负荷总和的平均值来计算平均负荷。例如，如果我想要最后十秒的平均负载(现在是 2012-02-07 12:00:10)，我可以将样本日期定在 12:00:02、12:00: 04、12:00:06、12:00:08 和 12:00:10。然后，我将通过汇总每个设备的最新负载值来计算负载总和:

2012-02-07 12:00:02 |  6  [= load(id=2)]
2012-02-07 12:00:04 | 18  [= load(id=2) + load(id=3) + load(id=4)]
2012-02-07 12:00:06 | 19  [= load(id=2) + load(id=4) + load(id=5)]
2012-02-07 12:00:08 | 19  [= load(id=2) + load(id=4) + load(id=5)]
2012-02-07 12:00:10 | 18  [= load(id=2) + load(id=5) + load(id=6)]

如果设备的负载值早于例如一个小时(此处发生 id=1)，则该设备的负载值将被忽略。在这种情况下，平均值为 16。

目前，我生成了一个相当复杂的查询，其中包含许多“UNION ALL”语句，速度非常慢:

SELECT avg(l.load_sum) as avg_load
FROM (
    SELECT sum(so.load) AS load_sum 
    FROM (
        SELECT * 
        FROM (
            SELECT si.macaddr, si.load 
            FROM sensor_data si WHERE si.timestamp > '2012-02-07 11:00:10' AND si.timestamp < '2012-02-07 12:00:10'
            ORDER BY si.timestamp DESC 
        ) AS sm
        GROUP BY macaddr
    ) so
    UNION ALL
    [THE SAME THING AGAIN WITH OTHER TIMESTAMPS]
    UNION ALL
    [AND AGAIN]
    UNION ALL
    [AND AGAIN]
    ...
) l

现在假设我想计算整个月的平均负载。对于每小时采样日期，我需要使用 UNION ALL 语句加入 30x24=720 个查询。在我的机器上完成整个查询需要将近一分钟。我确信没有 UNION ALL 语句会有更好的解决方案。但是，我没有在网上找到任何有用的东西。因此，非常感谢您的帮助!

Answer 1

使用 unix 时间戳的一小部分:首先我们计算每小时(3600 秒)的平均值:

SELECT
  macaddr, 
  sum(CAST(load AS float))/CAST(count(*) AS float) AS loadavg,
  FLOOR(UNIX_TIMESTAMP(`timestamp`)/3600) AS hourbase
FROM sensor_data
GROUP BY macaddr,FLOOR(UNIX_TIMESTAMP(`timestamp`)/3600)

然后我们计算一个月的平均值

SELECT 
  avg(loadavg) as monthlyavg,
  macaddr
FROM (
    SELECT
      macaddr, 
      sum(CAST(load AS float))/CAST(count(*) AS float) AS loadavg,
      FLOOR(UNIX_TIMESTAMP(`timestamp`)/3600) AS hourbase
    FROM sensor_data
    WHERE `timestamp` BETWEEN '2012-01-07 12:00:00' AND '2012-02-07 11:59:59'
    GROUP BY macaddr,FLOOR(UNIX_TIMESTAMP(`timestamp`)/3600)
) AS hourlies
GROUP BY macaddr, hourbase