Question

我目前正在运行此查询：

SELECT  Time_ID,
    Site_Type_ID,
    Abandoned_ID,
    WorkType_ID,
    SUM (staging.dbo.incoming_measure.ring_time) AS Ring_Time,
    SUM (staging.dbo.incoming_measure.hold_time) As Hold_Time,
    SUM (staging.dbo.incoming_measure.talk_time) AS Talk_Time,
    SUM (staging.dbo.incoming_measure.acw_time) AS ACW_Time,
    COUNT(*) CallCount
FROM incoming_measure
INNER JOIN DataMartEnd.dbo.Time_Dim 
        ON incoming_measure.StartTimeDate BETWEEN Time_Dim.Time_Start and 
                                                  Time_Dim.Time_End
INNER JOIN datamartend.dbo.Site_Type_Dim 
        ON incoming_measure.DBID = Site_Type_Dim.Site_Type_Code
INNER JOIN datamartend.dbo.Abandoned_Call_Dim 
        ON incoming_measure.Abandoned = Abandoned_Call_Dim.abandoned_value
INNER JOIN DataMartEnd.dbo.Work_Type_Dim 
        ON incoming_measure.DBID = work_type_dim.MIG_Site_ID AND
           Work_Type_Dim.Work_Type_Code = incoming_measure.Queue AND
           incoming_measure.StartTimeDate BETWEEN Work_Type_Dim.DimEffectiveStartDtm AND
                                                  Work_Type_Dim.DimEffectiveEndDtm
group by time_id, Site_Type_ID, Abandoned_ID, WorkType_ID

它返回了正确的结果但是大约需要8分钟才能运行，我只是想知道是否有人对如何加快查询有任何建议？主要问题是，如果这是我必须向客户展示最终结果的项目的一部分，我只允许10分钟演示（大学规则），这个查询大约是项目的30％。

估计执行的大部分是SORT，即57％。

Answer 1

SELECT
    Time_ID,
    Site_Type_ID,
    Abandoned_ID,
    WorkType_ID,
    SUM (im.ring_time) AS Ring_Time,
    SUM (im.hold_time) As Hold_Time,
    SUM (im.talk_time) AS Talk_Time,
    SUM (im.acw_time) AS ACW_Time,
    COUNT(*) CallCount
FROM incoming_measure im
INNER JOIN DataMartEnd.dbo.Time_Dim td
        ON dateadd(mi,
                   15*floor(datediff(mi,
                                     dateadd(dd, datediff(dd,0,im.StartTimeDate), 0),
                                     im.StartTimeDate ) / 15),
                   dateadd(dd, datediff(dd,0,im.StartTimeDate), 0) 
                  ) = td.Time_Start
INNER JOIN datamartend.dbo.Site_Type_Dim std
        ON im.DBID = std.Site_Type_Code
INNER JOIN datamartend.dbo.Abandoned_Call_Dim acd
        ON im.Abandoned = acd.abandoned_value
INNER JOIN DataMartEnd.dbo.Work_Type_Dim wtd
        ON im.DBID = wtd.MIG_Site_ID AND
           im.Queue = wtd.Work_Type_Code AND
           im.StartTimeDate BETWEEN wtd.DimEffectiveStartDtm AND wtd.DimEffectiveEndDtm
group by time_id, Site_Type_ID, Abandoned_ID, WorkType_ID

- 以便时间维度在等于值上连接，而不是在一系列值之间的值上。

如果这不能显着提高性能，那么我建议您在现有查询上创建索引视图，并从索引视图中选择新查询 - 您可以找到有关创建索引视图here的更多信息，是关于其限制的一些信息here。

Answer 2

我认为性能问题是由于加入了：

FROM incoming_measure
INNER JOIN DataMartEnd.dbo.Time_Dim 
        ON incoming_measure.StartTimeDate BETWEEN Time_Dim.Time_Start and 
                                                  Time_Dim.Time_End

Time_Dim的颗粒度是多少？ StartTimeDate的粒度是多少？这些名称表明一个是以天为单位测量的，另一个是以小时，分钟或秒为单位。这可能会导致大量其他记录匹配。

如果您有时间维度，为什么要存储常规日期？如果您有数据库日期时间，为什么使用时间维度表？

此外，您应该为每个表提供可读的别名。试图弄清楚像：

SUM (staging.dbo.incoming_measure.ring_time) AS Ring_Time,

比以下更难：

SUM (im.ring_time) AS Ring_Time,

其中im是incoming_message的简短别名。

Answer 3

试试这可能对你有帮助。

SELECT * FROM 
(SELECT  
ROW_NUMBER() OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID ORDER BY time_id) No,
Time_ID,
    Site_Type_ID,
    Abandoned_ID,
    WorkType_ID,
    SUM (staging.dbo.incoming_measure.ring_time) OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID) AS Ring_Time,
    SUM (staging.dbo.incoming_measure.hold_time) OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID) As Hold_Time,
    SUM (staging.dbo.incoming_measure.talk_time) OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID) AS Talk_Time,
    SUM (staging.dbo.incoming_measure.acw_time) OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID) AS ACW_Time,
    COUNT(1)  OVER( PARTITION BY time_id, Site_Type_ID, Abandoned_ID, WorkType_ID) CallCount
FROM incoming_measure
INNER JOIN DataMartEnd.dbo.Time_Dim 
        ON incoming_measure.StartTimeDate BETWEEN Time_Dim.Time_Start and 
                                                  Time_Dim.Time_End
INNER JOIN datamartend.dbo.Site_Type_Dim 
        ON incoming_measure.DBID = Site_Type_Dim.Site_Type_Code
INNER JOIN datamartend.dbo.Abandoned_Call_Dim 
        ON incoming_measure.Abandoned = Abandoned_Call_Dim.abandoned_value
INNER JOIN DataMartEnd.dbo.Work_Type_Dim 
        ON incoming_measure.DBID = work_type_dim.MIG_Site_ID AND
           Work_Type_Dim.Work_Type_Code = incoming_measure.Queue AND
           incoming_measure.StartTimeDate BETWEEN Work_Type_Dim.DimEffectiveStartDtm AND
                                                  Work_Type_Dim.DimEffectiveEndDtm
                                                 ) AS T1 WHERE No = 1

或

SELECT  Time_ID,
    Site_Type_ID,
    Abandoned_ID,
    WorkType_ID,
    SUM (staging.dbo.incoming_measure.ring_time) AS Ring_Time,
    SUM (staging.dbo.incoming_measure.hold_time) As Hold_Time,
    SUM (staging.dbo.incoming_measure.talk_time) AS Talk_Time,
    SUM (staging.dbo.incoming_measure.acw_time) AS ACW_Time,
    COUNT(1) CallCount
FROM incoming_measure
INNER JOIN DataMartEnd.dbo.Time_Dim 
        ON incoming_measure.StartTimeDate BETWEEN Time_Dim.Time_Start and 
                                                  Time_Dim.Time_End
INNER JOIN datamartend.dbo.Site_Type_Dim 
        ON incoming_measure.DBID = Site_Type_Dim.Site_Type_Code
INNER JOIN datamartend.dbo.Abandoned_Call_Dim 
        ON incoming_measure.Abandoned = Abandoned_Call_Dim.abandoned_value
INNER JOIN DataMartEnd.dbo.Work_Type_Dim 
        ON incoming_measure.DBID = work_type_dim.MIG_Site_ID AND
           Work_Type_Dim.Work_Type_Code = incoming_measure.Queue AND
           incoming_measure.StartTimeDate BETWEEN Work_Type_Dim.DimEffectiveStartDtm AND
                                                  Work_Type_Dim.DimEffectiveEndDtm
group by time_id, Site_Type_ID, Abandoned_ID, WorkType_ID

Answer 4

您需要以下字段的索引：

Time_Dim.Time_ID

incoming_measure.DBID
incoming_measure.Queue
incoming_measure.Abandoned
incoming_measure.StartTimeDate

Site_Type_Dim.Site_Type_ID
Site_Type_Dim.Site_Type_Code

Abandoned_Call_Dim.Abandoned_ID
Abandoned_Call_Dim.abandoned_value

Work_Type_Dim.WorkType_ID
Work_Type_Dim.Work_Type_Code
work_type_dim.MIG_Site_ID

我不完全确定索引中字段的顺序，因此您可能需要进行实验，但我建议：

create index Time_Dim_Time_ID on Time_Dim (Time_ID)
create index incoming_measure_index on (DBID, Queue, Abandoned, StartTimeDate)
create index Site_Type_index on Site_Type_Dim (Site_Type_ID, Site_Type_Code)
create index Abandoned_Call_index on Abandoned_Call_Dim (Abandoned_ID, abandoned_value)
create index Work_Type_index on Work_Type_Dim (WorkType_ID, Work_Type_Code, MIG_Site_ID)

我怎样才能加快我的SQL查询速度

4 个答案: