加入视图包含PIVOT时查询速度慢

时间:2013-08-13 11:50:14

标签: sql sql-server

我有以下包含PIVOT的视图:

CREATE VIEW [vwPhoneNumberByPref]
AS
SELECT * FROM(
    SELECT ObjectKey, ObjectTypeName,r_id,MaskedNumber FROM (
        SELECT * ,ROW_NUMBER() OVER (PARTITION BY ObjectKey, ObjectTypeName ORDER BY preference) as r_id
        FROM vwPhoneNumberMasked  ) AS a WHERE r_id<=3
    )AS P
PIVOT(
    MAX(MaskedNumber) FOR r_id in ([1],[2],[3])
)AS Pvt

当我使用WHERE ObjectKey = 1234运行它时会立即返回结果。当我在连接中使用此查询时:

Select * 
from tblPerson p 
inner join 
    vwPhoneNumberByPref ph on p.PersonID = ph.ObjectKey

大约需要10秒钟,并且在执行计划中,正在对超过200,000行进行排序。

如果未在连接中使用,则会在单行上执行“排序”。我尝试使用CROSS APPLY进行连接,这使得它稍微快一些,但仍然需要太长时间,并且仍在排序200,000多行。

如何改善表现?


其他信息:

CREATE VIEW [vwPhoneNumberMasked] AS
SELECT ph.ObjectTypeName, ph.ObjectKey, dbo.ApplyMask(ph.PhoneNumber, pt.Mask), ph.Preference
FROM tblPhoneNumber ph inner join tblPhoneType pt on
ph.PhoneTypeName = pt.Name
WHERE ph.isDeleted = 0
GO

CREATE TABLE tblPhoneType
Name varchar(20) NOT NULL,
Mask varchar(20) NOT NULL
GO

tblPhoneNumber

上的ObjectTypeName, ObjectKey, Preference, isDeleted上有一个索引

函数dbo.ApplyMask将MaskType从PhoneType应用到PhoneNumber,掩码为(##) #### ####

我实际上并未运行Select * from tblPerson,但即使运行SELECT PersonID from tblPerson INNER JOIN ....也会导致查询耗时。这仅仅是例如。

1 个答案:

答案 0 :(得分:0)

在您提供更多信息后,我建议您尝试以下操作。在表tblPhoneNumber上添加索引:

create index ix_tblPhoneNumber_1
    on tblPhoneNumber (isDeleted, ObjectKey, ObjectTypeName, Preference)
        include (PhoneNumber, PhoneTypeName)

在表tblPhoneType上添加(如果没有)

create unique clustered index ix_tblPhoneType
    on tblPhoneType (PhoneTypeName)

create unique index ix_tblPhoneType
    on tblPhoneType (PhoneTypeName) include (Mask)

将您的vwPhoneNumberMasked视图更改为:

ALTER VIEW [vwPhoneNumberMasked] AS
SELECT ph.ObjectTypeName, ph.ObjectKey, mp.Value as MaskedNumber, ph.Preference
FROM tblPhoneNumber ph
    inner /*loop*/ join tblPhoneType pt on ph.PhoneTypeName = pt.Name
    cross apply ftMaskPhone(ph.PhoneNumber, pt.Mask) mp
WHERE ph.isDeleted = 0

其中ftMaskPhone来自 this answer (请使用方法2 版本)。