SQL Server INSERT 到大表很慢

Question

我在 SQL Server 2008 数据库中有一个大表，它有大约 5.7 亿条记录。我们每天都会运行一个批处理作业，该作业获取包含 200,000 条左右交易记录的文件，对这些数据进行分组和求和，然后将其插入到大表中。

最近我尝试将大表的聚集索引更改为identity int列，这将插入时间从3小时缩短到1小时，但我仍然很困惑为什么这个简单的查询要花这么长时间来运行（无论桌子的大小）

这是有 5.7 亿行的表

CREATE TABLE [dbo].[POINTS_EARNED](
[POINTS_EARNED_ID]int identity not null,
[CARD_ID] [int] NOT NULL,
[CYCLE_ID] [int] NOT NULL,
[POINTS_CODE] [int] NOT NULL,
[NO_POINTS] [int] NULL,
[ACCOUNT_ID] [int] NOT NULL,
[CREATED_DATE] [datetime] NULL,
[CREATED_BY] [varchar](20) NULL,
[LAST_MODIFIED_DATE] [datetime] NULL,
[LAST_MODIFIED_BY] [varchar](20) NULL,
[DELETED] [bit] NULL,
CONSTRAINT [PK_POINTS_EARNED] PRIMARY KEY CLUSTERED 
(
[POINTS_EARNED_ID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, 
IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]

) ON [PRIMARY]

它还有一些约束（默认值和外键）和索引，我想知道这些是否是导致问题的原因。

实际需要一个小时运行的 SQL 是：

insert into points_earned (
    card_id,
    cycle_id,
    points_code,
    no_points,
    account_id
    ) 

select pe.card_id, pe.cycle_id, pe.points_code, sum(pe.no_points),pe.account_id
from #points_earned pe 
group by pe.card_id, pe.cycle_id, pe.points_code,pe.account_id

临时表 #points_earned 有大约 200,000 行，并具有以下结构（没有索引）

create table #points_earned (
        card_id           int,
        cycle_id    int,
        points_code int,
        card_type   varchar(5),
        no_points   int,
        account_id int
        )

所以，我想听听大家的意见，我是否应该

在临时表上添加索引
添加数据之前在大表上删除非聚集索引，然后重新创建它们
还有其他选择吗？

更新 - 根据要求提供更多信息

select 语句在 2 秒内运行，没有插入，所以这似乎不是问题，所以可能不需要担心临时表的索引

索引、（更新）触发器和约束是：

    CREATE NONCLUSTERED INDEX [IDX_CYCLE_ID] ON [dbo].[POINTS_EARNED] 
(
    [CYCLE_ID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [IX_ACCOUNT_ID] ON [dbo].[POINTS_EARNED] 
(
    [ACCOUNT_ID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [IX_ACCOUNT_ID_POINTS_CODE] ON [dbo].[POINTS_EARNED] 
(
    [ACCOUNT_ID] ASC,
    [POINTS_CODE] ASC
)
INCLUDE ( [CARD_ID],
[CYCLE_ID],
[NO_POINTS]) WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [RELATION_151_FK] ON [dbo].[POINTS_EARNED] 
(
    [CARD_ID] ASC,
    [CYCLE_ID] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
CREATE NONCLUSTERED INDEX [RELATION_152_FK] ON [dbo].[POINTS_EARNED] 
(
    [POINTS_CODE] ASC
)WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, SORT_IN_TEMPDB = OFF, IGNORE_DUP_KEY = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
GO
/****** Object:  Trigger [update_points_earned]    Script Date: 09/13/2013 13:20:54 ******/
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
CREATE TRIGGER [dbo].[update_points_earned] ON [dbo].[POINTS_EARNED]
FOR UPDATE
AS
 BEGIN
  UPDATE points_earned 
  SET Last_Modified_By = USER,
   Last_Modified_Date = GETDATE()  
  FROM
   points_earned t,
   inserted i
  WHERE
   t.card_id = i.card_id AND
   t.cycle_id = i.cycle_id AND
   t.points_code = i.points_code AND
   t.account_id = i.account_id
 END
GO
/****** Object:  Default [DF_POINTS_EARNED_ACCOUNT_ID]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED] ADD  CONSTRAINT [DF_POINTS_EARNED_ACCOUNT_ID]  DEFAULT ((0)) FOR [ACCOUNT_ID]
GO
/****** Object:  Default [DF_POINTS_EARNED_CREATED_DATE]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED] ADD  CONSTRAINT [DF_POINTS_EARNED_CREATED_DATE]  DEFAULT (getdate()) FOR [CREATED_DATE]
GO
/****** Object:  Default [DF_POINTS_EARNED_CREATED_BY]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED] ADD  CONSTRAINT [DF_POINTS_EARNED_CREATED_BY]  DEFAULT (user_name()) FOR [CREATED_BY]
GO
/****** Object:  Default [DF_POINTS_EARNED_DELETED]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED] ADD  CONSTRAINT [DF_POINTS_EARNED_DELETED]  DEFAULT ((0)) FOR [DELETED]
GO
/****** Object:  ForeignKey [FK_POINTS_E_REFERENCE_CYCLE_CA]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED]  WITH CHECK ADD  CONSTRAINT [FK_POINTS_E_REFERENCE_CYCLE_CA] FOREIGN KEY([CARD_ID], [CYCLE_ID])
REFERENCES [dbo].[CYCLE_CARD] ([CARD_ID], [CYCLE_ID])
GO
ALTER TABLE [dbo].[POINTS_EARNED] CHECK CONSTRAINT [FK_POINTS_E_REFERENCE_CYCLE_CA]
GO
/****** Object:  ForeignKey [FK_POINTS_E_REFERENCE_POINTS_C]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED]  WITH NOCHECK ADD  CONSTRAINT [FK_POINTS_E_REFERENCE_POINTS_C] FOREIGN KEY([POINTS_CODE])
REFERENCES [dbo].[POINTS_CODE] ([POINTS_CODE])
GO
ALTER TABLE [dbo].[POINTS_EARNED] CHECK CONSTRAINT [FK_POINTS_E_REFERENCE_POINTS_C]
GO
/****** Object:  ForeignKey [FK_POINTS_EARNED_REF_ACCOUNT]    Script Date: 09/13/2013 13:20:54 ******/
ALTER TABLE [dbo].[POINTS_EARNED]  WITH NOCHECK ADD  CONSTRAINT [FK_POINTS_EARNED_REF_ACCOUNT] FOREIGN KEY([ACCOUNT_ID])
REFERENCES [dbo].[ACCOUNT] ([ACCOUNT_ID])
GO
ALTER TABLE [dbo].[POINTS_EARNED] CHECK CONSTRAINT [FK_POINTS_EARNED_REF_ACCOUNT]

编辑2、insert语句的查询计划

  |--Sequence
   |--Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[IDX_CYCLE_ID]), SET:([POINTS_EARNED_ID1040] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID],[CYCLE_ID1041] = [Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID]) WITH ORDERED PREFETCH)
   |    |--Sort(ORDER BY:([Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID] ASC, [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID] ASC))
   |         |--Table Spool
   |              |--Clustered Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[PK_POINTS_EARNED]), SET:([Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID] = RaiseIfNullInsert([tempdb].[dbo].[#points_earned].[card_id] as [pe].[card_id]),[Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID] = RaiseIfNullInsert([tempdb].[dbo].[#points_earned].[cycle_id] as [pe].[cycle_id]),[Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE] = RaiseIfNullInsert([tempdb].[dbo].[#points_earned].[points_code] as [pe].[points_code]),[Progressive_Points].[dbo].[POINTS_EARNED].[NO_POINTS] = [Expr1006],[Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID] = RaiseIfNullInsert([tempdb].[dbo].[#points_earned].[account_id] as [pe].[account_id]),[Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID] = [Expr1007],[Progressive_Points].[dbo].[POINTS_EARNED].[CREATED_DATE] = [Expr1008],[Progressive_Points].[dbo].[POINTS_EARNED].[CREATED_BY] = [Expr1009],[Progressive_Points].[dbo].[POINTS_EARNED].[DELETED] = [Expr1010],[Progressive_Points].[dbo].[POINTS_EARNED].[LAST_MODIFIED_DATE] = NULL,[Progressive_Points].[dbo].[POINTS_EARNED].[LAST_MODIFIED_BY] = NULL) WITH UNORDERED PREFETCH)
   |                   |--Compute Scalar(DEFINE:([Expr1008]=getdate(), [Expr1009]=CONVERT_IMPLICIT(varchar(20),user_name(),0), [Expr1010]=(0)))
   |                        |--Compute Scalar(DEFINE:([Expr1007]=getidentity((1243867498),(8),NULL)))
   |                             |--Top(ROWCOUNT est 0)
   |                                  |--Parallelism(Gather Streams)
   |                                       |--Compute Scalar(DEFINE:([Expr1006]=CASE WHEN [Expr1062]=(0) THEN NULL ELSE [Expr1063] END))
   |                                            |--Hash Match(Aggregate, HASH:([pe].[card_id], [pe].[cycle_id], [pe].[points_code], [pe].[account_id]), RESIDUAL:([tempdb].[dbo].[#points_earned].[card_id] as [pe].[card_id] = [tempdb].[dbo].[#points_earned].[card_id] as [pe].[card_id] AND [tempdb].[dbo].[#points_earned].[cycle_id] as [pe].[cycle_id] = [tempdb].[dbo].[#points_earned].[cycle_id] as [pe].[cycle_id] AND [tempdb].[dbo].[#points_earned].[points_code] as [pe].[points_code] = [tempdb].[dbo].[#points_earned].[points_code] as [pe].[points_code] AND [tempdb].[dbo].[#points_earned].[account_id] as [pe].[account_id] = [tempdb].[dbo].[#points_earned].[account_id] as [pe].[account_id]) DEFINE:([Expr1062]=COUNT_BIG([tempdb].[dbo].[#points_earned].[no_points] as [pe].[no_points]), [Expr1063]=SUM([tempdb].[dbo].[#points_earned].[no_points] as [pe].[no_points])))
   |                                                 |--Parallelism(Repartition Streams, Hash Partitioning, PARTITION COLUMNS:([pe].[card_id], [pe].[cycle_id], [pe].[points_code], [pe].[account_id]))
   |                                                      |--Clustered Index Scan(OBJECT:([tempdb].[dbo].[#points_earned] AS [pe]))
   |--Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[IX_ACCOUNT_ID]), SET:([POINTS_EARNED_ID1042] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID],[ACCOUNT_ID1043] = [Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID]) WITH ORDERED PREFETCH)
   |    |--Sort(ORDER BY:([Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID] ASC, [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID] ASC))
   |         |--Table Spool
   |--Assert(WHERE:(CASE WHEN [Expr1050] IS NULL THEN (0) ELSE NULL END))
   |    |--Nested Loops(Left Semi Join, OUTER REFERENCES:([Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID], [Expr1068]) WITH UNORDERED PREFETCH, DEFINE:([Expr1050] = [PROBE VALUE]))
   |         |--Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[IX_ACCOUNT_ID_POINTS_CODE]), SET:([POINTS_EARNED_ID1044] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID],[CARD_ID1045] = [Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID],[CYCLE_ID1046] = [Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID],[POINTS_CODE1047] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE],[NO_POINTS1048] = [Progressive_Points].[dbo].[POINTS_EARNED].[NO_POINTS],[ACCOUNT_ID1049] = [Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID]) WITH UNORDERED PREFETCH)
   |         |    |--Table Spool
   |         |--Clustered Index Seek(OBJECT:([Progressive_Points].[dbo].[ACCOUNT].[PK_ACCOUNT]), SEEK:([Progressive_Points].[dbo].[ACCOUNT].[ACCOUNT_ID]=[Progressive_Points].[dbo].[POINTS_EARNED].[ACCOUNT_ID]) ORDERED FORWARD)
   |--Assert(WHERE:(CASE WHEN [Expr1054] IS NULL THEN (0) ELSE NULL END))
   |    |--Nested Loops(Left Semi Join, OUTER REFERENCES:([Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID], [Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID], [Expr1070]) WITH UNORDERED PREFETCH, DEFINE:([Expr1054] = [PROBE VALUE]))
   |         |--Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[RELATION_151_FK]), SET:([POINTS_EARNED_ID1051] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID],[CARD_ID1052] = [Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID],[CYCLE_ID1053] = [Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID]) WITH ORDERED PREFETCH)
   |         |    |--Sort(ORDER BY:([Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID] ASC, [Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID] ASC, [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID] ASC))
   |         |         |--Table Spool
   |         |--Row Count Spool
   |              |--Index Seek(OBJECT:([Progressive_Points].[dbo].[CYCLE_CARD].[IDX_NCLST_CARD_ID_CYCLE_ID]), SEEK:([Progressive_Points].[dbo].[CYCLE_CARD].[CARD_ID]=[Progressive_Points].[dbo].[POINTS_EARNED].[CARD_ID] AND [Progressive_Points].[dbo].[CYCLE_CARD].[CYCLE_ID]=[Progressive_Points].[dbo].[POINTS_EARNED].[CYCLE_ID]) ORDERED FORWARD)
   |--Assert(WHERE:(CASE WHEN [Expr1057] IS NULL THEN (0) ELSE NULL END))
        |--Merge Join(Left Semi Join, MANY-TO-MANY MERGE:([Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE])=([Progressive_Points].[dbo].[POINTS_CODE].[POINTS_CODE]), RESIDUAL:([Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE]=[Progressive_Points].[dbo].[POINTS_CODE].[POINTS_CODE]))
             |--Index Insert(OBJECT:([Progressive_Points].[dbo].[POINTS_EARNED].[RELATION_152_FK]), SET:([POINTS_EARNED_ID1055] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID],[POINTS_CODE1056] = [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE]) WITH ORDERED PREFETCH)
             |    |--Sort(ORDER BY:([Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_CODE] ASC, [Progressive_Points].[dbo].[POINTS_EARNED].[POINTS_EARNED_ID] ASC))
             |         |--Table Spool
             |--Index Scan(OBJECT:([Progressive_Points].[dbo].[POINTS_CODE].[POINTS_CODES_PK]), ORDERED FORWARD)

Answer 1

好的，这就是我要做的：

检查您是否需要两个索引 [IX_ACCOUNT_ID_POINTS_CODE] 和 [IX_ACCOUNT_ID]，因为它们可能是多余的。
在执行
```
INSERT
```
之前，禁用触发器并删除外键。
执行
```
INSERT
```
设置通常由触发器设置的字段，并确保 FK 列的值有效。
重新启用触发器，并使用 NOCHECK 重新创建外键。

当您插入的行数少于总行数的 0.2% 时，我会保留索引，因此就地更新它们可能比删除并重建它们更快。

Answer 2

与其一次性从庞大的表中删除 20 万行，不如尝试对其进行分块。例如：

while (1=1)
begin
  delete top(1000) from #points_earned
  output deleted.* into points_earned
  if @@rowcount=0 break
end

Answer 3

对于您的表，有 3 个注意事项会影响您添加的每条记录的性能： (1) 您的索引 (2) 你的触发点 (3) 你的外键

如果您负担得起，请为您的表应用适当的架构，例如适当的 SAS 驱动器中的分区表和分区索引。

否则，在类似的情况下，每分钟更新数十万条记录，我使用 BULK/INSERT 技术，并使用另一个表（例如 [POINTS_EARNED_TMP] 位于同一实例 (*) 内的单独数据库上。

使用触发器 [POINTS_EARNED_TMP] 添加记录。然后，从另一个 BULK 中，您设置没有触发器的过程，并将数据从 [POINTS_EARNED_TMP] 批量/插入到 [POINTS_EARNED]（包括用户和日期更新）。至少，避免了 Trigger 性能，也避免了同一个 Instance 内的#TMP。

(*) 使用另一个数据库主要是出于维护原因。到目前为止，与 INSERT TO 相比，BULK 给出了惊人的结果。

SQL Server INSERT 到大表很慢

问题描述投票：0回答：3

3个回答

最新问题

SQL Server INSERT 到大表很慢

问题描述 投票：0回答：3

3个回答

最新问题

问题描述投票：0回答：3