UPD

Question

我有一个python应用程序，它将设备作为测试来检查它们是否存活。

以下是我用来对每个设备进行分组的查询，并根据PingResults中该设备的最后一条记录确定设备是否在线

SELECT     
    c.ID, c.DeviceName, c.GroupID, c.DeviceIP, p1.Status, 
    p1.DateTime AS LastUpdate, DeviceGroups.GroupName
FROM         
    Devices AS c 
INNER JOIN
    PingResults AS p1 ON c.ID = p1.DeviceID 
INNER JOIN
    DeviceGroups ON c.GroupID = DeviceGroups.ID 
LEFT OUTER JOIN
    PingResults AS p2 ON c.ID = p2.DeviceID 
                      AND (p1.DateTime < p2.DateTime OR
                           p1.DateTime = p2.DateTime AND p1.DeviceID < p2.DeviceID)
WHERE     
    (p2.ID IS NULL)

查询有效，但目前在pingResults表和28个设备中使用44000条记录运行1:23分钟。

查询输出

DeviceName  DeviceIP    GroupID ID  Status        LastUpdate           GroupName

Machine 25  192.168.0.226   1   114 True    2018-02-20 09:46:40.717 Machine Terminals
Machine 2   192.168.0.199   1   100 True    2018-02-20 09:48:09.113 Machine Terminals
Machine 3   192.168.0.229   1   101 True    2018-02-20 09:48:12.710 Machine Terminals
Machine 4   192.168.0.224   1   102 True    2018-02-20 09:48:15.123 Machine Terminals
Machine 5   192.168.0.218   1   103 True    2018-02-20 09:48:17.763 Machine Terminals
Machine 6   192.168.0.219   1   104 True    2018-02-20 09:48:19.823 Machine Terminals
Machine 7   192.168.0.217   1   105 False   2018-02-20 09:48:23.763 Machine Terminals
Machine 8   192.168.0.220   1   106 False   2018-02-20 09:48:26.763 Machine Terminals

当前执行计划

有没有什么方法可以优化或更改此查询以提高效率，因为在目前的速度下，当数据库填满时运行时间太长。

Answer 1

LEFT OUTER JOIN  PingResults AS p2 
   ON c.ID = p2.DeviceID AND (p1.DateTime < p2.DateTime 
      OR p1.DateTime = p2.DateTime AND p1.DeviceID < p2.DeviceID)
WHERE     (p2.ID IS NULL)

-->

WHERE NOT EXISTS(SELECT 1 FROM PingResults AS p2 
    WHERE c.ID = p2.DeviceID AND (p1.DateTime < p2.DateTime 
      OR p1.DateTime = p2.DateTime AND p1.DeviceID < p2.DeviceID)
)

为每个表创建聚簇索引！

UPD

接着看了一遍：

SELECT  c.ID,
        c.DeviceName,
        c.GroupID,
        c.DeviceIP,
        p1.[Status],
        p1.LastUpdate,
        DeviceGroups.GroupName
FROM    Devices                  AS c   
        INNER JOIN  DeviceGroups
            ON  c.GroupID = DeviceGroups.ID   
        CROSS APPLY(
            SELECT TOP 1 p1.[Status], p1.[DateTime] AS LastUpdate
            FROM PingResults  AS p1
            WHERE p1.DeviceID  = c.ID
            ORDER BY p1.DateTime DESC
        ) p1

无需对PingResults进行两次“读取”。

另一件事是：

ON c.ID = p2.DeviceID 
AND (p1.DateTime < p2.DateTime 
   OR p1.DateTime = p2.DateTime 
     AND p1.DeviceID < p2.DeviceID ---<<<<<<<<<<<<<<<<<<<<
)

真的？ c.ID = p1.DeviceID和c.ID = p2.DeviceID之后？

Answer 2

对于此查询：

SELECT c.ID, c.DeviceName, c.GroupID, c.DeviceIP, p1.Status, 
       p1.DateTime AS LastUpdate, dg.GroupName
FROM Devices c INNER JOIN
     PingResults p1
     ON c.ID = p1.DeviceID INNER JOIN
     DeviceGroups dg
     ON c.GroupID = dg.ID LEFT OUTER JOIN
     PingResults p2
     ON c.ID = p2.DeviceID AND
        (p1.DateTime < p2.DateTime OR
         p1.DateTime = p2.DateTime AND p1.DeviceID < p2.DeviceID
        )
WHERE p2.ID IS NULL;

您希望确保您拥有索引：

PingResults(deviceId, datetime)
DeviceGroups(id, groupname)（次要优化）

您可能还会发现使用row_number()编写此内容会加快查询速度：

SELECT c.ID, c.DeviceName, c.GroupID, c.DeviceIP, p1.Status, 
       p1.DateTime AS LastUpdate, dg.GroupName
FROM Devices c INNER JOIN
     (select p1.*, row_number() over (partition by deviceid order by datetime) as seqnum
      from PingResults p1
     ) p1
     ON c.ID = p1.DeviceID AND seqnum = 1 INNER JOIN
     DeviceGroups dg
     ON c.GroupID = dg.ID ;

Answer 3

您计划中的最大成本是哈希匹配（内部联接）。

要加快速度，请将JOIN（设备和设备组）中涉及的两个表索引到它们加入的列上。

这应该将哈希匹配更改为内循环，这比您在计划中看到的要快得多。

Answer 4

对我来说，你应该使用row_number：

;WITH CTE as
(
  SELECT  
    c.ID, 
    c.DeviceName, 
    c.GroupID, 
    c.DeviceIP, 
    p1.Status, 
    p1.DateTime,
    DeviceGroups.GroupName,
    row_number() over (partition by c.ID order by p1.DateTime desc) rn
  FROM
    Devices AS c
  INNER JOIN
    PingResults AS p1
  ON c.ID = p1.DeviceID
  INNER JOIN
    DeviceGroups
  ON c.GroupID = DeviceGroups.ID
)
SELECT 
  ID, 
  DeviceName, 
  GroupID, 
  DeviceIP, 
  Status, 
  DateTime LastUpdate,
  DeviceGroups.GroupName
FROM CTE
WHERE rn = 1

优化我的SQL Server查询

4 个答案:

UPD