tsql合并速度慢下来

时间:2013-08-21 13:49:54

标签: database tsql indexing

TSql合并指令由于某种原因而变慢。

我批量合并数据批量大小等于10000条记录,我看到从一批到另一批合并需要更长时间。

这是合并指令:

MERGE [dbo].[SResult] AS target
            USING [dbo].[SResultTemp] AS source 
            ON (target.QSId = source.QSId 
                and target.ResultId = source.ResultId 
                and target.EngineId = source.EngineId)  
            WHEN NOT MATCHED THEN INSERT 
                (QSId, ResultId, EngineId, Position) 
            values 
                (source.QSId, source.ResultId, source.EngineId, source.Position);

源表如此声明

    CREATE TABLE [dbo].[SResultTemp](
        [QSId] int not null,
        [ResultId] int not null,
        [EngineId] int not null,
        [Position] int not null,
        CONSTRAINT [PK_SResultTemp] PRIMARY KEY CLUSTERED(
            [QSId], [ResultId], [EngineId], [Position] ASC
        )WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
        ) ON [PRIMARY]

目标是相同的,但它有另外的字段SResultId作为主键和另一组索引:

PK_SResult - 主要

IX_SResult_QSId_ResultId - 非唯一,非聚集

IX_SResult_EngineId - 非唯一,非聚集

UX_SResult_EngineId_QSId_Position - 唯一,非群集

以下是我在日志中看到的内容:

    Results Upload: SResult took 00:00:01.0008344 
    Results Upload: SResult took 00:00:18.1046734
    Results Upload: SResult took 00:00:17.9797846
    Results Upload: SResult took 00:00:27.7828817
    Results Upload: SResult took 00:01:30.4140091
    Results Upload: SResult took 00:03:17.6433416
    Results Upload: SResult took 00:03:21.3761251
    Results Upload: SResult took 00:06:07.2555342
    Results Upload: SResult took 00:06:56.2423653
    Results Upload: SResult took 00:06:57.1729179
    Results Upload: SResult took 00:07:09.7221083

另外,我使用多个表,因此对于其他表,没有这样的规律性。有人可以帮忙吗?

谢谢!

3 个答案:

答案 0 :(得分:1)

当然,对于具有四个字段的群集PK,您需要更长的时间吗?您在合并期间不断重新排序表中的所有记录。

答案 1 :(得分:1)

索引片段插入速度降低 正如你所看到的那样可以迅速降级 如果你可以按照PK的顺序插入,那么这将减少碎片 填充索引会减慢碎片的速度 是否可以在插入完成后删除所有非聚簇索引然后重建?

SResultId是一个身份吗?

为什么指数设计?
您有一个与SResultTemp上的PK不同的唯一约束。

索引加速选择但会减慢插入和更新 联接的索引将有所帮助,但现在索引被拆分。

我的建议是将其分解为唯一性和测试所需的索引。

答案 2 :(得分:0)

有两个问题:

  1. 随着越来越多的行被添加,增加索引和维护工作
  2. ON-CLAUSE

    ON (target.QSId = source.QSId 
        and target.ResultId = source.ResultId 
        and target.EngineId = source.EngineId) 
    
  3. 导致对目标中检查的需求增加,因为存在越来越多需要比较的行。

    删除索引会极大地减慢ON子句的检查速度。我们在这里存在利益冲突。