MySQL 较大记录集中的组求和

Question

我有一个有趣的情况。我将发布 SQL 然后解释。

SELECT 
    `c`.`clusterName` AS `clusterName`,
    DATE_FORMAT(`p`.`insertedTS`, '%Y-%m-%d') AS `insertedDate`,
    MAX(`p`.`maxMemory`) AS `peakMaxMemory`,
    MIN(`p`.`maxMemory`) AS `minMaxMemory`,
    MAX(`p`.`usedMem`) AS `peakUsedMem`,
    MIN(`p`.`usedMem`) AS `minUsedMem`,
    MAX(`p`.`availMem`) AS `peakAvailMem`,
    MIN(`p`.`availMem`) AS `minAvailMem`,
    MAX(`p`.`numPoweredOnVms`) AS `peakPoweredOnVms`,
    MIN(`p`.`numPoweredOnVms`) AS `minPoweredOnVms`,
    (MAX(`p`.`maxMemory`) - MIN(`p`.`maxMemory`)) AS `deltaMaxMemory`,
    (MAX(`p`.`usedMem`) - MIN(`p`.`usedMem`)) AS `deltaUsedMem`,
    (MAX(`p`.`availMem`) - MIN(`p`.`availMem`)) AS `deltaAvailMem`,
    (MAX(`p`.`numPoweredOnVms`) - MIN(`p`.`numPoweredOnVms`)) AS `deltaPoweredOnVms`
FROM
    (`vtrend`.`clusters` `c`
    JOIN `vtrend`.`capacityDataRawPOSH` `p` ON ((`c`.`clusterID` = `p`.`clusterID`)))
GROUP BY `c`.`clusterName` , DATE_FORMAT(`p`.`insertedTS`, '%Y-%m-%d')
ORDER BY `c`.`clusterName`

结果如下所示:

clusterName,insertedDate,peakMaxMemory,minMaxMemory,peakUsedMem,minUsedMem,peakAvailMem,minAvailMem,peakPoweredOnVms,minPoweredOnVms,deltaMaxMemory,deltaUsedMem,deltaAvailMem,deltaPoweredOnVms
Some-Cluster,2016-09-20,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-21,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-22,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-23,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-24,221,221,111,111,111,111,17,16,0,0,0,1
Some-Cluster,2016-09-25,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-26,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-27,221,221,111,111,111,111,17,17,0,0,0,0
Some-Cluster,2016-09-28,221,221,111,111,111,111,17,15,0,0,0,2
Some-Cluster,2016-09-29,221,221,111,111,111,111,17,17,0,0,0,0

因此，我想要完成(但失败)的是在末尾添加一列，该列将显示 numPoweredOnVms 计数在一天中增加或减少。由于上面的每条记录代表一整天的分组，每 15 分钟输入一次数据，因此我希望能够查看 numPoweredOnVms 中的第一条记录和当天，并从中减去当天的最后一条记录然后得到一个正数(增加)或负数(减少)计数，我可以使用 IF 语句来呈现“增加”或“减少”文本。

这是原始数据的示例:

idcapacityDataRaw,insertedTS,clusterID,totalRamGb,maxMemory,availMem,usedMem,actualUsageRamGb,totalCpuMhz,maxMhz,usedMhz,numPoweredOnVms
739,"2016-09-20 16:50:12",29,384,221,111,111,111,268032,178688,3394,17
976,"2016-09-20 21:50:42",29,384,221,111,111,111,268032,178688,4072,17
1074,"2016-09-20 22:13:07",29,384,221,111,111,111,268032,178688,4683,17
1172,"2016-09-20 22:35:36",29,384,221,111,111,111,268032,178688,3916,17
1270,"2016-09-20 22:58:01",29,384,221,111,111,111,268032,178688,3857,17
1365,"2016-09-20 23:21:06",29,384,221,111,111,111,268032,178688,7867,17
1463,"2016-09-20 23:43:37",29,384,221,111,111,111,268032,178688,3971,17
1575,"2016-09-21 00:11:05",29,384,221,111,111,111,268032,178688,8196,17
2149,"2016-09-21 19:13:33",29,384,221,111,111,111,268032,178688,3758,17
2495,"2016-09-21 21:01:48",29,384,221,111,111,111,268032,178688,3927,16

所以基本上我希望能够从当天的第一条记录和最后一条记录中提取 numPoweredOnVms 并减去它们。现在有趣的部分是如何在多天内对多个集群执行此操作？我想我可能需要对原始数据进行单独的 View 并按记录计算该记录，然后使用该 View 作为此查询的支持，只是不确定。

Answer 1

使用原始查询作为子查询，并让它返回每天的第一个和最后一个时间戳。然后连接回表以获取每次该列的值。

SELECT c.clusterName, p.*,
       CASE SIGN(plast.numPoweredOnVms  > pfirst.numPoweredOnVms)
            WHEN -1 THEN 'Decrease'
            WHEN 0 THEN 'No change'
            ELSE 'Increase'
       END AS changePoweredOnVms
FROM vtrend.clusters AS c
JOIN (
    SELECT 
        clusterID,
        DATE(`insertedTS`) AS `insertedDate`,
        MAX(`maxMemory`) AS `peakMaxMemory`,
        MIN(`maxMemory`) AS `minMaxMemory`,
        MAX(`usedMem`) AS `peakUsedMem`,
        MIN(`usedMem`) AS `minUsedMem`,
        MAX(`availMem`) AS `peakAvailMem`,
        MIN(`availMem`) AS `minAvailMem`,
        MAX(`numPoweredOnVms`) AS `peakPoweredOnVms`,
        MIN(`numPoweredOnVms`) AS `minPoweredOnVms`,
        (MAX(`maxMemory`) - MIN(`maxMemory`)) AS `deltaMaxMemory`,
        (MAX(`usedMem`) - MIN(`usedMem`)) AS `deltaUsedMem`,
        (MAX(`availMem`) - MIN(`availMem`)) AS `deltaAvailMem`,
        (MAX(`numPoweredOnVms`) - MIN(`numPoweredOnVms`)) AS `deltaPoweredOnVms`
        MIN(insertedTS) AS firstTS,
        MAX(insertedTS) AS lastTS
    FROM `vtrend`.`capacityDataRawPOSH`
    GROUP BY clusterID, insertedDate) AS p ON p.clusterID = c.clusterID
JOIN vtrend.capacityDataRawPOSH AS pfirst ON pfirst.clusterID = p.clusterID AND pfirst.insertedTS = p.firstTS
JOIN vtrend.capacityDataRawPOSH AS plast ON plast.clusterID = p.clusterID AND plast.insertedTS = p.lastTS

此外，使用 DATE(insertedTS) 而不是 DATE_FORMAT(insertedTS, '%Y-%m-%d') 将时间戳转换为日期。 %Y-%m-%d 是显示日期时使用的默认格式。