假设有一个包含两列的简单审计表(在生产中有更多列):
ID | Date
处理请求时,我们会在此表中添加记录。 请求分批处理,批处理中可以有任意数量的项目。对于每个项目,我们将添加一条记录。批次之间至少有2秒的延迟(数量可配置)。
性能是通过每单位时间(例如每秒)处理请求的速度来衡量的。考虑这个样本数据(2个集群,项目数量仅用于演示目的):
--2016-01-29 10:27:25.603
--2016-01-29 10:27:25.620
--2016-01-29 10:27:25.637
--2016-01-29 10:27:25.653
--2016-01-29 10:27:25.723
--Avg time between requests = 24ms
--2016-01-29 10:27:34.647
--2016-01-29 10:27:34.667
--2016-01-29 10:27:34.680
--2016-01-29 10:27:34.690
--2016-01-29 10:27:34.707
--Avg time = 12ms
我们可以说,最糟糕的是,每秒可以处理41.67个请求,最多可以处理83.33个请求。很高兴知道平均批次性能。
问题。是否可以单独使用T-SQL获取这些指标以及如何使用?
编辑:要使结果具有统计显着性,丢弃批次可能比小于10个项目(可配置)更有用。
答案 0 :(得分:3)
也许我已经过度简化了您的请求,但请考虑以下内容
Declare @YourTable table (ID int,Date datetime)
Insert Into @YourTable values
( 1,'2016-01-29 10:27:25.603'),
( 2,'2016-01-29 10:27:25.620'),
( 3,'2016-01-29 10:27:25.637'),
( 4,'2016-01-29 10:27:25.653'),
( 5,'2016-01-29 10:27:25.723'),
( 6,'2016-01-29 10:27:34.647'),
( 7,'2016-01-29 10:27:34.667'),
( 8,'2016-01-29 10:27:34.680'),
( 9,'2016-01-29 10:27:34.690'),
(10,'2016-01-29 10:27:34.707')
Declare @BatchSecondsGap int = 2 -- Seconds Between Batches
Declare @MinObservations int = 5 -- Batch must n or greater
;with cte as (
Select *,Cnt = sum(1) over (Partition By Batch)
From (
Select *,Batch = sum(Flg) over (Order By Date)
From (
Select ID,Date
,Flg = case when DateDiff(SECOND,Lag(Date,1,null) over (Order By Date),Date)>=@BatchSecondsGap then 1 else 0 end
,MS = case when DateDiff(SECOND,Lag(Date,1,Date) over (Order By Date),Date)>=@BatchSecondsGap then 0 else DateDiff(MILLISECOND,Lag(Date,1,Date) over (Order By Date),Date) end
From @YourTable
) A
) B
)
Select Title = 'Total'
,DateR1 = min(Date)
,DateR2 = max(Date)
,BatchCnt = count(Distinct Batch)
,TransCnt = count(*)
,MS_Ttl = sum(MS)
,MS_Avg = avg(MS*1.0)
,MS_Std = stdev(MS)
From cte
Where Cnt>=@MinObservations
Union All
Select Title = concat('Batch ',Batch)
,DateR1 = min(Date)
,DateR2 = max(Date)
,BatchCnt = count(Distinct Batch)
,TransCnt = count(*)
,MS_Ttl = sum(MS)
,MS_Avg = avg(MS*1.0)
,MS_Std = stdev(MS)
From cte
Where Cnt>=@MinObservations
Group By Batch
返回
下图说明您不会因批次之间的时间而受到惩罚,因此它会成为最终结果的简单聚合