SQL Server基线算法（同比和环比）
基线介绍
基线为历史数据统计而成的数据，具有参考价值，并利用基线数据与当前值的对比，通过一定的报警机制，形成实时监控架构。SQL Server计数器采用同比和环比两种方式。
同比：可以计算未来一周的基线数据，取近八周同一天同一时刻的均值。
1 /****************************** 功能描述：<同比基线算法> 2 * 创建者：<HuangCH〉 3 * 创建日期：<2014-11-05> 4 * 备注说明：<每天执行一次> 5 ########## 6 Change Log 7 ########## 8 Date Changer Description 9 -------------------------------------------------- 10 <2014-11-05> <HuangCH> <新建> 11 说明： 12 1、同比算法取近八周数据可以统计一周后的数据 13 2、考虑当天前一周0点之前的数据已经归档SQLPerfDataStat_Histroy.dbo.CounterData_OneMinute_Bak，因此要计算前一天之前的数据。 14 15 2、计数器新建要超过一周才能开始计算 16 17 3、开始时间当天0点与结束时间不能超过一周--CONVERT(VARCHAR(26),DATEADD(WEEK,1,@LastCBaseEnd),23)<=@LastOneMinute 18 19 -------------------------------------------------- 20 ***************************/ 21 ALTER PROC [dbo].[spb_CBaseCounterData] 22 AS 23 SET NOCOUNT ON 24 DECLARE @LastOneMinute DATETIME 25 DECLARE @LastCBaseEnd DATETIME --计算一周之后的数据 26 DECLARE @RunCBaseEnd DATETIME 27 DECLARE @RunCBaseEtart DATETIME 28 DECLARE @CounterID INT 29 DECLARE @MachineName VARCHAR(128) 30 31 --若时间未超过一周，则取当前时间 32 UPDATE SQLPerfDataStat.dbo.CounterDetails_Dts 33 SET LastCBaseEnd=DATEADD(D,-1,LastOneMinute) 34 WHERE DATEADD(WEEK,-1,DATEADD(D,-1,LastOneMinute))<CreateTime 35 36 DECLARE CBaseCur CURSOR FOR 37 --计数器新建超过一周； 38 --只能计算当天0点之前的数据：CONVERT(DATETIME,CONVERT(VARCHAR(16),LastOneMinute,23)) 39 --计算当前的时间的前天之前 40 SELECT DISTINCT MachineName,DATEADD(D,-1,LastOneMinute) LastOneMinute,CONVERT(DATETIME,CONVERT(VARCHAR(16),LastCBaseEnd,120)) LastCBaseEnd 41 FROM SQLPerfDataStat.dbo.CounterDetails_Dts WITH(NOLOCK) 42 WHERE DATEDIFF(MI,CONVERT(DATETIME,CONVERT(VARCHAR(16),LastCBaseEnd,120)),DATEADD(D,-1,LastOneMinute))>=1 and DATEADD(WEEK,-1,DATEADD(D,-1,LastOneMinute))>=CreateTime 43 44 OPEN CBaseCur 45 46 FETCH NEXT FROM CBaseCur INTO @MachineName,@LastOneMinute,@LastCBaseEnd 47 48 WHILE @@FETCH_STATUS=0 49 BEGIN 50 51 IF CONVERT(VARCHAR(26),DATEADD(WEEK,1,@LastCBaseEnd),23)<=@LastOneMinute--超过一周未收集则时间为上次收集时间推迟一周 52 BEGIN 53 SET @RunCBaseEnd=CONVERT(VARCHAR(26),DATEADD(WEEK,1,@LastCBaseEnd),23) 54 SET @RunCBaseEtart=@LastCBaseEnd 55 END 56 ELSE 57 BEGIN 58 SET @RunCBaseEnd=@LastOneMinute 59 SET @RunCBaseEtart=@LastCBaseEnd 60 END 61 62 UPDATE SQLPerfDataStat.dbo.CounterDetails_Dts 63 SET LastCBaseEnd=@RunCBaseEnd 64 WHERE MachineName=@MachineName 65 66 BEGIN TRY 67 68 INSERT INTO dbo.CBaseCounterData_OneMinute(CounterID, CounterDateTime, CounterValue,MaxCounterValue) 69 SELECT CounterID 70 ,DATEADD(MI,-DATEDIFF(MI,CounterDateTime,Dateadd(week,1,@RunCBaseEnd))%(7*24*60),Dateadd(week,1,@RunCBaseEnd)) CounterDateTime 71 ,AVG(CounterValue) CounterValue,MAX(CounterValue) MaxCounterValue 72 FROM ( 73 SELECT CounterID,CounterDateTime,CounterValue 74 FROM SQLPerfDataStat_Histroy.dbo.CounterData_OneMinute_Bak a (NOLOCK) 75 WHERE 76 ( 77 (CounterDateTime>=Dateadd(week,-7,@RunCBaseEtart) 78 AND CounterDateTime<Dateadd(week,-7,@RunCBaseEnd)) 79 OR 80 (CounterDateTime>=Dateadd(week,-6,@RunCBaseEtart) 81 AND CounterDateTime<Dateadd(week,-6,@RunCBaseEnd)) 82 OR 83 (CounterDateTime>=Dateadd(week,-5,@RunCBaseEtart) 84 AND CounterDateTime<Dateadd(week,-5,@RunCBaseEnd)) 85 OR 86 (CounterDateTime>=Dateadd(week,-4,@RunCBaseEtart) 87 AND CounterDateTime<Dateadd(week,-4,@RunCBaseEnd)) 88 OR 89 (CounterDateTime>=Dateadd(week,-3,@RunCBaseEtart) --取近四周 90 AND CounterDateTime<Dateadd(week,-3,@RunCBaseEnd)) 91 OR 92 (CounterDateTime>=Dateadd(week,-2,@RunCBaseEtart) 93 AND CounterDateTime<Dateadd(week,-2,@RunCBaseEnd)) 94 OR 95 (CounterDateTime>=Dateadd(week,-1,@RunCBaseEtart) 96 AND CounterDateTime<Dateadd(week,-1,@RunCBaseEnd)) 97 ) AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName=@MachineName) 98 UNION ALL 99 SELECT CounterID,CounterDateTime,CounterValue 100 FROM dbo.CounterData_OneMinute a (NOLOCK) 101 WHERE (CounterDateTime>=@RunCBaseEtart 102 AND CounterDateTime<@RunCBaseEnd) 103 AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName=@MachineName) 104 ) AA 105 GROUP BY CounterID,DATEADD(MI,-DATEDIFF(MI,CounterDateTime,Dateadd(week,1,@RunCBaseEnd))%(7*24*60),Dateadd(week,1,@RunCBaseEnd)) 106 107 END TRY 108 BEGIN CATCH 109 IF @@TRANCOUNT >0 110 BEGIN 111 Rollback; 112 END; 113 THROW 114 END CATCH 115 116 FETCH NEXT FROM CBaseCur INTO @MachineName,@LastOneMinute,@RunCBaseEnd 117 END 118 CLOSE CBaseCur 119 DEALLOCATE CBaseCur
同比可以消除由于业务变动带来的波动，具有报警对比价值。下图红线同比数据，表示历史上该时段存在异常升高。
环比：可以计算未来一天的基线数据，取近两周每天同一时刻的均值。
1 /****************************** 功能描述：<环比基线算法> 2 * 创建者：<HuangCH〉 3 * 创建日期：<2014-11-05> 4 * 备注说明：<每小时执行一次> 5 ########## 6 Change Log 7 ########## 8 Date Changer Description 9 -------------------------------------------------- 10 <2014-11-05> <HuangCH> <新建> 11 说明： 12 1、环比算法可以统计一天后的数据 13 14 2、如果开始时间与当前时间对比超过一天，则只取开始时间当天的数据进行计算--CONVERT(VARCHAR(26),DATEADD(D,1,@LastRBaseEnd),23)< = @LastOneMinute 15 16 3、如果时间分别为两天的0点，则需要单独处理当天0点的数据--DATEDIFF(MI,@RunRBaseStart,@RunRBaseEnd)=1440 17 18 4、环比算法，分离工作日和周末，本别计算，统计两周内的数据--DATEPART(WEEKDAY,@RunRBaseStart) IN(2,3,4,5,6)--非周末 19 20 -------------------------------------------------- 21 ***************************/ 22 ALTER PROC [dbo].[spb_RBaseCounterData] 23 AS 24 SET NOCOUNT ON 25 SET DATEFIRST 7--由于需要考虑工作日与周末，因此需要确认默认设置是以周日开始 26 DECLARE @LastOneMinute DATETIME 27 DECLARE @LastRBaseEnd DATETIME --计算一周之后的数据 28 DECLARE @RunRBaseEnd DATETIME 29 DECLARE @RunRBaseStart DATETIME 30 DECLARE @SqlCmd VARCHAR(MAX) 31 DECLARE @WHERE VARCHAR(2000) 32 DECLARE @GROUPBY VARCHAR(MAX) 33 DECLARE @CYLE INT 34 DECLARE @MachineName VARCHAR(128) 35 DECLARE @ERRORMSG VARCHAR(MAX) 36 SET @ERRORMSG='' 37 38 --更新未超过一周的数据 39 UPDATE SQLPerfDataStat.dbo.CounterDetails_Dts 40 SET LastRBaseEnd=CONVERT(DATETIME,CONVERT(VARCHAR(16),GETDATE(),120)) 41 WHERE DATEADD(WEEK,-1,LastOneMinute)<CreateTime 42 43 --取最后计算小于当前统计时间的数据 44 DECLARE RBaseCur CURSOR FOR 45 SELECT MachineName,LastOneMinute,CONVERT(DATETIME,CONVERT(VARCHAR(16),LastRBaseEnd,120)) LastRBaseEnd 46 FROM SQLPerfDataStat.dbo.CounterDetails_Dts WITH(NOLOCK) 47 WHERE DATEDIFF(MI,LastRBaseEnd,LastOneMinute)>1 AND DATEADD(WEEK,-1,LastOneMinute)>=CreateTime--不超过新建时间一周，则不处理 48 49 OPEN RBaseCur 50 51 FETCH NEXT FROM RBaseCur INTO @MachineName,@LastOneMinute,@LastRBaseEnd 52 53 WHILE @@FETCH_STATUS=0 54 BEGIN 55 56 IF CONVERT(VARCHAR(26),DATEADD(D,1,@LastRBaseEnd),23)< = @LastOneMinute--如果第二天0点小于当前时间；即不能跨天处理 57 BEGIN 58 SET @RunRBaseEnd=CONVERT(VARCHAR(26),DATEADD(D,1,@LastRBaseEnd),23) 59 SET @RunRBaseStart=@LastRBaseEnd 60 END 61 ELSE 62 BEGIN--在同一天 63 SET @RunRBaseEnd=@LastOneMinute 64 SET @RunRBaseStart=@LastRBaseEnd 65 END 66 67 IF DATEPART(WEEKDAY,@RunRBaseStart) IN(2,3,4,5,6)--非周末 68 BEGIN 69 SET @WHERE='WHERE ((CounterDateTime>'+CASE WHEN CONVERT(VARCHAR(24),@RunRBaseStart,120)=CONVERT(VARCHAR(24),@RunRBaseStart,23)+' 00:00:00' THEN ''ELSE '=' END +''''+CONVERT(VARCHAR(24),@RunRBaseStart,120)+''' AND CounterDateTime<'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120)+''')' 70 SET @CYLE =1 71 WHILE(@CYLE<14) 72 BEGIN 73 IF DATEPART(WEEKDAY,DATEADD(D,-@CYLE,@RunRBaseStart)) IN(2,3,4,5,6) 74 BEGIN 75 SET @WHERE=@WHERE+CHAR(10)+'OR (CounterDateTime>'++CASE WHEN CONVERT(VARCHAR(24),@RunRBaseStart,120)=CONVERT(VARCHAR(24),@RunRBaseStart,23)+' 00:00:00' THEN ''ELSE '=' END ++''''+CONVERT(VARCHAR(24),DATEADD(D,-@CYLE,@RunRBaseStart),120)+''' AND CounterDateTime<'''+CONVERT(VARCHAR(24),DATEADD(D,-@CYLE,@RunRBaseEnd),120)+''')' 76 END 77 SET @CYLE = @CYLE + 1 78 END 79 SET @GROUPBY='DATEADD(MI,-DATEDIFF(MI,CounterDateTime,'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120)+''')%(24*60),'''+CASE WHEN DATEPART(WEEKDAY,@RunRBaseStart) =6 THEN CONVERT(VARCHAR(24),DATEADD(D,3,@RunRBaseEnd),120) ELSE CONVERT(VARCHAR(24),DATEADD(D,1,@RunRBaseEnd),120) END+''')'--周五要计算周一的数据 80 END 81 ELSE--周末数据，取近两周周末数据库 82 BEGIN 83 SET @WHERE='WHERE ((CounterDateTime>'+CASE WHEN CONVERT(VARCHAR(24),@RunRBaseStart,120)=CONVERT(VARCHAR(24),@RunRBaseStart,23)+' 00:00:00' THEN ''ELSE '=' END +''''+CONVERT(VARCHAR(24),@RunRBaseStart,120)+''' AND CounterDateTime<'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120) +''')' 84 +CHAR(10)+'OR (CounterDateTime>'+CASE WHEN CONVERT(VARCHAR(24),@RunRBaseStart,120)=CONVERT(VARCHAR(24),@RunRBaseStart,23)+' 00:00:00' THEN ''ELSE '=' END +''''+CONVERT(VARCHAR(24),DATEADD(WEEK,-1,@RunRBaseStart),120)+''' AND CounterDateTime<'''+CONVERT(VARCHAR(24),DATEADD(WEEK,-1,@RunRBaseEnd),120)+''')' 85 SET @GROUPBY='DATEADD(MI,-DATEDIFF(MI,CounterDateTime,'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120)+''')%(24*60),'''+CONVERT(VARCHAR(24),DATEADD(WEEK,1,@RunRBaseEnd),120)+''')' 86 END 87 88 UPDATE SQLPerfDataStat.dbo.CounterDetails_Dts 89 SET LastRBaseEnd=@RunRBaseEnd 90 WHERE MachineName=@MachineName 91 92 BEGIN TRY 93 --插入数据 94 SET @SqlCmd='INSERT INTO dbo.RBaseCounterData_OneMinute(CounterID, CounterDateTime, CounterValue,MaxCounterValue)' 95 +CHAR(10)+'SELECT CounterID,'+@GROUPBY+',AVG(CounterValue),MAX(CounterValue) FROM' 96 +CHAR(10)+'(SELECT CounterID,CounterDateTime,CounterValue' 97 +CHAR(10)+'FROM CounterData_OneMinute a (NOLOCK)' 98 +CHAR(10)+@WHERE+') ' 99 +CHAR(10)+'AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName='''+@MachineName+''')' 100 +CHAR(10)+'Union all' 101 +CHAR(10)+'SELECT CounterID,CounterDateTime,CounterValue' 102 +CHAR(10)+'FROM SQLPerfDataStat_Histroy.dbo.CounterData_OneMinute_bak a (NOLOCK)' 103 +CHAR(10)+@WHERE+') ' 104 +CHAR(10)+'AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName='''+@MachineName+''')' 105 +CHAR(10)+')AA' 106 +CHAR(10)+ 'GROUP BY CounterID,'+@GROUPBY 107 exec (@SqlCmd) 108 109 --0点数据处理 110 IF CONVERT(VARCHAR(24),@RunRBaseStart,120)=CONVERT(VARCHAR(24),@RunRBaseStart,23)+' 00:00:00' 111 BEGIN 112 IF DATEPART(WEEKDAY,@RunRBaseStart) IN(2,3,4,5,6)--非周末 113 BEGIN 114 SET @WHERE='WHERE ((CounterDateTime='''+CONVERT(VARCHAR(24),@RunRBaseStart,120)+''')' 115 SET @CYLE =1 116 WHILE(@CYLE<14) 117 BEGIN 118 IF DATEPART(WEEKDAY,DATEADD(D,-@CYLE,@RunRBaseStart)) IN(2,3,4,5,6) 119 BEGIN 120 SET @WHERE=@WHERE+CHAR(10)+'OR (CounterDateTime='''+CONVERT(VARCHAR(24),DATEADD(D,-@CYLE,@RunRBaseStart),120)+''')' 121 END 122 SET @CYLE = @CYLE + 1 123 END 124 SET @GROUPBY='DATEADD(MI,-DATEDIFF(MI,CounterDateTime,'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120)+''')%(24*60),'''+CASE WHEN DATEPART(WEEKDAY,@RunRBaseStart) =6 THEN CONVERT(VARCHAR(24),DATEADD(D,3,@RunRBaseEnd),120) ELSE CONVERT(VARCHAR(24),DATEADD(D,1,@RunRBaseEnd),120) END+''')' 125 END 126 ELSE--周末数据，取近两周周末数据库 127 BEGIN 128 SET @WHERE='WHERE ((CounterDateTime='''+CONVERT(VARCHAR(24),@RunRBaseStart,120)+''')' 129 +CHAR(10)+'OR (CounterDateTime='''+CONVERT(VARCHAR(24),DATEADD(WEEK,-1,@RunRBaseStart),120)+''')' 130 SET @GROUPBY='DATEADD(MI,-DATEDIFF(MI,CounterDateTime,'''+CONVERT(VARCHAR(24),@RunRBaseEnd,120)+''')%(24*60),'''+CONVERT(VARCHAR(24),DATEADD(D,-1,DATEADD(WEEK,1,@RunRBaseEnd)),120)+''')' 131 END 132 SET @SqlCmd='INSERT INTO dbo.RBaseCounterData_OneMinute(CounterID, CounterDateTime, CounterValue,MaxCounterValue)' 133 +CHAR(10)+'SELECT CounterID,'+@GROUPBY+',AVG(CounterValue),MAX(CounterValue) FROM' 134 +CHAR(10)+'(SELECT CounterID,CounterDateTime,CounterValue' 135 +CHAR(10)+'FROM CounterData_OneMinute a (NOLOCK)' 136 +CHAR(10)+@WHERE+') ' 137 +CHAR(10)+'AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName='''+@MachineName+''')' 138 +CHAR(10)+'Union all' 139 +CHAR(10)+'SELECT CounterID,CounterDateTime,CounterValue' 140 +CHAR(10)+'FROM SQLPerfDataStat_Histroy.dbo.CounterData_OneMinute_bak a (NOLOCK)' 141 +CHAR(10)+@WHERE+') ' 142 +CHAR(10)+'AND EXISTS (SELECT TOP 1 1 FROM CounterDetails_Collect WITH(NOLOCK) WHERE CounterID = a.CounterID AND MachineName='''+@MachineName+''')' 143 +CHAR(10)+')AA' 144 +CHAR(10)+ 'GROUP BY CounterID,'+@GROUPBY 145 exec (@SqlCmd) 146 END 147 148 END TRY 149 BEGIN CATCH 150 SET @ERRORMSG=@ERRORMSG+ERROR_MESSAGE() 151 IF @@TRANCOUNT >0 152 BEGIN 153 ROLLBACK; 154 END 155 END CATCH 156 157 FETCH NEXT FROM RBaseCur INTO @MachineName,@LastOneMinute,@LastRBaseEnd 158 END 159 CLOSE RBaseCur 160 DEALLOCATE RBaseCur 161 162 163 IF @ERRORMSG<>'' 164 BEGIN 165 ;THROW 50000,@ERRORMSG,1 166 END
环比数据可以说明数据近期走势。具有优化对比价值。从上同比图可以看出117-27-139-238服务器cpu在历史上存在异常升高，下图红线环比比数据，却与蓝线同步，说明此异常在近期已经正常。
