为什么在这个查询中没有使用主（集群）键？

Question

我有一个 SQL Server 2008 R2 表，其架构结构如下所示：

CREATE TABLE [dbo].[CDSIM_BE]
(
    [ID] [bigint] NOT NULL,
    [EquipmentID] [varchar](50) NOT NULL,
    [SerialNumber] [varchar](50) NULL,
    [PyrID] [varchar](50) NULL,
    [MeasMode] [varchar](50) NULL,
    [ReadTime] [datetime] NOT NULL,
    [SubID] [varchar](15) NULL,
    [ProbePosition] [float] NULL,
    [DataPoint] [int] NULL,

    CONSTRAINT [PK_CDSIM_BE] 
    PRIMARY KEY CLUSTERED ([ID] ASC, [EquipmentID] ASC, [ReadTime] ASC)
         WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, 
               IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, 
               ALLOW_PAGE_LOCKS = ON) ON [MonthlyArchiveScheme9]([ReadTime])
) ON [MonthlyArchiveScheme9]([ReadTime])

CREATE NONCLUSTERED INDEX [idx_CDSIM_BE__SubID_ProbePosition] 
ON [dbo].[CDSIM_BE] ([SubID] ASC, [ProbePosition] ASC)
INCLUDE ([EquipmentID], [ReadTime], [BECorr]) 
    WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, 
          SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, 
          ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [MonthlyArchiveScheme9]([ReadTime])

CREATE NONCLUSTERED INDEX [IX_CDSIM_BE_ProbePosition] 
ON [dbo].[CDSIM_BE] ([ProbePosition] ASC)
INCLUDE ([SerialNumber], [SubID]) 
    WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, 
          SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, 
          ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [MonthlyArchiveScheme9]([ReadTime])

CREATE NONCLUSTERED INDEX [IX_CSDIM_Readtime] 
ON [dbo].[CDSIM_BE]([ReadTime] ASC)
INCLUDE ([EquipmentID]) 
    WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, 
          SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, 
          ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [MonthlyArchiveScheme9]([ReadTime])

我正在执行这个简单的查询：

Select Max(Id)
From dbo.CDSIM_BE

表中有 ~2.5B 行。

查询计划显示正在对IX_CdSIM_BE_ProbePosition索引进行索引扫描。我想知道为什么 SQL Server 根本不使用聚集（和主）索引并立即转到表中的最后一行并检索 Id 值，因为它必须是最大值。

Answer 1

聚集索引被分区，ReadTime所以它不能像你描述的那样使用 PK。它需要Max(Id)为每个分区找到 ，然后找到它们的最大值。它是可以重写查询，但得到这样的计划。

使用基于本文的示例，可能的重写可能是

SELECT MAX(ID) AS ID
FROM   sys.partitions AS P
       CROSS APPLY (SELECT MAX(ID) AS ID
                    FROM   [dbo].[CDSIM_BE]
                    WHERE  $PARTITION.MonthlyArchiveFunction9(ReadTime) 
                                                    = P.partition_number) AS A
WHERE  P.object_id = OBJECT_ID('dbo.CDSIM_BE')
       AND P.index_id <= 1; 

依次处理每个分区。

请注意，该计划仍然有一个扫描（使用搜索谓词来选择分区），但这不是对分区的完整扫描。

扫描按索引顺序进行，方向为“BACKWARD”。在TOP收到后的第一个迭代器可以请求停止从扫描行。