Question

我有两个问题：

SELECT SELECT NamesRecord.NameID, NamesRecord.FulfillmentAddressID NameFulfillmentAddressID, ContractRecord.FulfillmentAddressID, ContractRecord.BillingAddressId
FROM Magnet.dbo.ContractRecord ContractRecord
    INNER JOIN Magnet.dbo.NamesRecord NamesRecord
        ON NamesRecord.NameId = ContractRecord.DonorId
WHERE NameID > -1
AND (EXISTS (
        SELECT 1
        FROM Magnet.dbo.AddressRecord AddressRecord
        WHERE AddressRecord.AddressId = ContractRecord.FulfillmentAddressId
        AND BuildingFloor LIKE 'M%')
    OR  EXISTS (
        SELECT 1
        FROM Magnet.dbo.AddressRecord AddressRecord
        WHERE AddressRecord.AddressId = ContractRecord.BillingAddressId
        AND BuildingFloor LIKE 'M%'))


SELECT SELECT NamesRecord.NameID, NamesRecord.FulfillmentAddressID NameFulfillmentAddressID, ContractRecord.FulfillmentAddressID, ContractRecord.BillingAddressId
FROM Magnet.dbo.ContractRecord ContractRecord
    INNER JOIN Magnet.dbo.NamesRecord NamesRecord
        ON NamesRecord.NameId = ContractRecord.DonorId
WHERE NameID > -1
AND (EXISTS (SELECT 1
        FROM Magnet.dbo.AddressRecord AddressRecord
        WHERE AddressRecord.AddressId IN (ContractRecord.FulfillmentAddressId, ContractRecord.BillingAddressId)
        AND BuildingFloor LIKE 'M%'))

第一个查询运行速度比第二个快10多倍。根据执行计划，第一个查询使用两个Clustered Index Scans，其中“BuildingFloor LIKE'M％'”作为谓词，并在ContractRecord上使用WHERE子句中的每个子选项进行索引搜索（每个成本40％）子选择）。

第二个查询使用Clustered Index Seek，其中“BuildingFloor LIKE'M％'”作为谓词，并使用AddressId约束的搜索谓词（96％成本）。估计行数也完全没有（实际为250，估计为1）。

如何改善第二个查询的性能？我可以强制SQL Server选择替代策略，还是必须修改表上的索引？

Answer 1

每行子查询和分离（or）过滤条件一样慢。完全摆脱子查询，如果您在过滤器中使用or谓词，则可以考虑将其替换为union。在内部，in被转换为or。

select
    NamesRecord.NameId
from (
    select
        ContractRecord.DonorId,
        ContractRecord.FulfillmentAddressId as AddressId
    from Magnet.dbo.ContractRecord ContractRecord
    union
    select
        ContractRecord.DonorId,
        ContractRecord.BillingAddressId as AddressId
    from Magnet.dbo.ContractRecord ContractRecord
) ContractRecordInfo
join Magnet.dbo.NamesRecord NamesRecord on 1=1
    and NamesRecord.NameId = ContractRecordInfo.DonorId
    and NamesRecord.NameId > -1
join Magnet.dbo.AddressRecord AddressRecord on 1=1
    and AddressRecord.AddressId = ContractRecordInfo.AddressId
    and AddressRecord.BuildingFloor like 'M%'

不良的聚集索引寻求表现？

1 个答案: