Question

有一个表（SQL Server 2008 R2）可以保存多个服务器的up / down日志。服务器定期进行ping操作，并将其状态（向上或向下）写入此表。它有这样的结构：

CREATE TABLE StatusLog
(
  LogID INT PRIMARY KEY,
  ServerID INT,
  QueryDate DATETIME,
  ServerStatus VARCHAR(50)
)

示例数据

INSERT INTO StatusLog
VALUES
(1, '1724', '2016-04-16 09:28:00.000', 'up'),
(2, '1724', '2016-04-16 09:29:00.000', 'up'),
(3, '1724', '2016-04-16 09:30:00.000', 'down'),
(6, '1724', '2016-04-16 09:31:00.000', 'down'),
(8, '1724', '2016-04-16 09:32:00.000', 'down'),
(9, '1724', '2016-04-16 09:33:00.000', 'down'),
(17, '1724', '2016-04-16 09:33:40.000', 'up'),
(18, '1724', '2016-04-16 09:34:00.000', 'up')

我试图在给定时间段内找到特定服务器的总停机时间。在上面的数据提取中，ID为1724的服务器的状态变为＆＃34; down＆＃34;在09:30:00然后回到＆＃34; up＆＃34;在09:33:40，这是220秒的总停机时间。

我的方法是：

对于每个＆＃34; down block＆＃34;，找到＆＃34; down＆＃34;记录并将其QueryDate设置为新列中的下行开始时间。这很快。
在另一个新专栏中，找到第一个＆＃34; up＆＃34;在该开始时间之后记录并将其QueryDate设置为停机时间的结束。这相当快。
但是，仅对下方块中的第一个向下记录执行此操作，而不对下方块中的其他向下执行此操作，否则您错误地多次计算相同的停机时间。现在要做到这一点，我需要查看行号，这是事情变得混乱和缓慢的地方。
最后，从彼此中提取它们，并且你有那段时间的停机时间
总结所有停机时间以查找总停机时间。

我编写了以下脚本，但速度非常慢（每台服务器都有数十万条日志记录）

DECLARE @StartDate DATE = '2016-04-01'
DECLARE @EndDate DATE = '2016-04-30'
DECLARE @ServerID INT = '1724' 

;WITH CTE_StatusLog AS 
(
SELECT LogID, QueryDate, ServerStatus, 
    ROW_NUMBER() OVER (ORDER BY QueryDate) AS RN 
FROM StatusLog 
WHERE ServerID = @ServerID
    AND QueryDate BETWEEN @StartDate AND @EndDate
)

SELECT LogID, 
       QueryDate,
       ServerStatus,
       RN,
       DownStarted = CASE WHEN s1.ServerStatus = 'down' 
                          THEN s1.QueryDate END,
       DownEnded = (SELECT TOP 1 QueryDate 
                    FROM CTE_StatusLog AS s2 
                    WHERE s2.QueryDate > s1.QueryDate
                    AND s1.ServerStatus = 'down'
                    AND s2.ServerStatus = 'up'
                    AND (SELECT s3.ServerStatus 
                    FROM CTE_StatusLog AS s3 
                    WHERE s3.RN = s1.RN-1) <> 'down'
                ORDER BY s2.QueryDate),
       DownDuration = DATEDIFF(SECOND, 
                CASE WHEN s1.ServerStatus = 'down' 
                    THEN s1.QueryDate END, 
                (SELECT TOP 1 QueryDate 
                FROM CTE_StatusLog AS s2 
                WHERE s2.QueryDate > s1.QueryDate
                AND s1.ServerStatus = 'down'
                AND s2.ServerStatus = 'up'
                AND (SELECT s3.ServerStatus 
                    FROM CTE_StatusLog AS s3 
                    WHERE s3.RN = s1.RN-1) <> 'down'
                ORDER BY s2.QueryDate))
FROM CTE_StatusLog AS s1
WHERE QueryDate BETWEEN @StartDate AND @EndDate
ORDER BY s1.RN

输出：

LogID   QueryDate               ServerStatus   RN   DownStarted             DownEnded               DownDuration
1       2016-04-16 09:28:00.000 up             1    NULL                    NULL                    NULL
2       2016-04-16 09:29:00.000 up             2    NULL                    NULL                    NULL
3       2016-04-16 09:30:00.000 down           3    2016-04-16 09:30:00.000 2016-04-16 09:33:40.000 220
6       2016-04-16 09:31:00.000 down           4    2016-04-16 09:31:00.000 NULL                    NULL
8       2016-04-16 09:32:00.000 down           5    2016-04-16 09:32:00.000 NULL                    NULL
9       2016-04-16 09:33:00.000 down           6    2016-04-16 09:33:00.000 NULL                    NULL
17      2016-04-16 09:33:40.000 up             7    NULL                    NULL                    NULL
18      2016-04-16 09:34:00.000 up             8    NULL                    NULL                    NULL

如何改进此脚本或是否有更好的方法来计算此表结构的停机时间？

Answer 1

如果您只需要总停机时间，您可以弄清楚每行代表什么：假设每个下行代表自上次检查该服务器以来停机时间的秒数。然后SUM那些行：

DECLARE @StartDate DATE = '2016-04-01'
DECLARE @EndDate DATE = '2016-04-30'
DECLARE @ServerID INT = '1724'

SELECT
individualRows.ServerId,
individualRows.ServerStatus,
SUM(secondsInState) AS TotalTime
FROM
(Select
statusLog.ServerId,
statusLog.QueryDate,
statusLog.ServerStatus,
DateDiff(second, PreviousStatus.QueryDate, statusLog.QueryDate) as secondsInState
FROM
StatusLog
left outer join
StatusLog AS PreviousStatus
ON StatusLog.ServerId = PreviousStatus.ServerId
AND PreviousStatus.QueryDate < StatusLog.QueryDate
AND PreviousStatus.QueryDate = ( SELECT Max(QueryDate) FROM statusLog sl2 where sl2.ServerId= StatusLog.ServerId and sl2.QueryDate < StatusLog.QueryDate)
WHERE StatusLog.QueryDate > @StartDate
AND StatusLog.QueryDate < @EndDate
AND StatusLog.ServerId = @ServerID ) AS individualRows
GROUP BY
individualRows.ServerId,
individualRows.ServerStatus

如果你确实需要每次中断的时间，我可能会尝试一个临时表，每行与前一行以及相反状态的前一行连接。与您的结果类似。然后我会过滤并聚合那个临时表。

我的经验是，一旦表格获得多行数据，临时表就会比CTE快得多。

Answer 2

我会通过获得每个下记录的下一个上升时间来实现此目的。在SQL Server 2008中，它使用outer apply：

select sl.*, slup.querydate as next_update,
       datediff(second, sl.querydate, slup.querydate) as down_in_seconds
from statuslog sl outer apply
     (select top 1 sl2.*
      from statuslog sl2
      where sl2.serverid = sl.serverid and
            sl2.querydate >= sl.querydate and
            sl2.serverstatus = 'up'
      order by sl2.querydate asc
     ) slup
where sl.serverstatus = 'down';

如果你想通过停机时间得到摘要，那么我会使用聚合：

select servid, min(querydate) as down_date, next_update, 
       max(down_in_seconds)
from (select sl.*, slup.querydate as next_update,
             datediff(second, sl.querydate, slup.querydate) as down_in_seconds
      from statuslog sl outer apply
           (select top 1 sl2.*
            from statuslog sl2
            where sl2.serverid = sl.serverid and
                  sl2.querydate >= sl.querydate and
                  sl2.serverstatus = 'up'
            order by sl2.querydate asc
           ) slup
      where sl.serverstatus = 'down'
     ) slud
group by serverid, next_update;

从服务器日志表

2 个答案: