Question

我无法在Sybase 15.5上优化以下查询。有谁知道我怎么能改善它。在那里使用的每个表都有大约3000万行。我尽力优化它，但仍然耗费大量时间（1.5小时）。

create table #tmp1( f_id smallint, a_date smalldatetime )
create table #tmp2( f_id smallint, a_date smalldatetime )


insert #tmp1
select f_id, a_date = max( a_date )
    FROM  audit_table
    WHERE  i_date =  @pIDate
group by f_id

insert #tmp2
select f_id , a_date = max( a_date )
    FROM  n_audit_table
    WHERE  i_date =  @pIDate
group by f_id

    create table #tmp(
    t_account varchar(32) not null,
    t_id varchar(32) not null,
    product varchar(64) null
    )

insert into #tmp
select t_account,t_id, product
    FROM  audit_table nt, #tmp1 a
    WHERE  i_date   =  @pIDate
    and nt.a_date = a.a_date
    and nt.f_id = a.f_id
union
select t_account,t_id, product
    FROM  n_audit_table t, #tmp2 a
    WHERE  t.item_date =  @pIDate
    and t.a_date = a.a_date
    and t.f_id = a.f_id

这两个表都有i_date，a_date，f_id上的索引。请在下面的showplan中找到很长时间。声明2的查询计划（第24行）。使用串行模式优化

STEP 1
    The type of query is INSERT.

10 operator(s) under root

   |ROOT:EMIT Operator (VA = 10)
   |
   |   |INSERT Operator (VA = 9)
   |   |  The update mode is direct.
   |   |
   |   |   |HASH UNION Operator (VA = 8)  has 2 children.
   |   |   | Using Worktable1 for internal storage.
   |   |   |  Key Count: 3
   |   |   |
   |   |   |   |NESTED LOOP JOIN Operator (VA = 3) (Join Type: Inner Join)
   |   |   |   |
   |   |   |   |   |SCAN Operator (VA = 0)
   |   |   |   |   |  FROM TABLE
   |   |   |   |   |  #tmp1
   |   |   |   |   |  a
   |   |   |   |   |  Table Scan.
   |   |   |   |   |  Forward Scan.
   |   |   |   |   |  Positioning at start of table.
   |   |   |   |   |  Using I/O Size 2 Kbytes for data pages.
   |   |   |   |   |  With LRU Buffer Replacement Strategy for data pages.
   |   |   |   |
   |   |   |   |   |RESTRICT Operator (VA = 2)(5)(0)(0)(0)(0)
   |   |   |   |   |
   |   |   |   |   |   |SCAN Operator (VA = 1)
   |   |   |   |   |   |  FROM TABLE
   |   |   |   |   |   |  audit_table
   |   |   |   |   |   |  nt
   |   |   |   |   |   |  Index : IX_audit_table
   |   |   |   |   |   |  Forward Scan.
   |   |   |   |   |   |  Positioning by key.
   |   |   |   |   |   |  Keys are:
   |   |   |   |   |   |    i_date ASC
   |   |   |   |   |   |    a_date ASC
   |   |   |   |   |   |  Using I/O Size 2 Kbytes for index leaf pages.
   |   |   |   |   |   |  With LRU Buffer Replacement Strategy for index leaf pages.
   |   |   |   |   |   |  Using I/O Size 2 Kbytes for data pages.
   |   |   |   |   |   |  With LRU Buffer Replacement Strategy for data pages.
   |   |   |
   |   |   |   |NESTED LOOP JOIN Operator (VA = 7) (Join Type: Inner Join)
   |   |   |   |
   |   |   |   |   |SCAN Operator (VA = 4)
   |   |   |   |   |  FROM TABLE
   |   |   |   |   |  #tmp2
   |   |   |   |   |  a
   |   |   |   |   |  Table Scan.
   |   |   |   |   |  Forward Scan.
   |   |   |   |   |  Positioning at start of table.
   |   |   |   |   |  Using I/O Size 2 Kbytes for data pages.
   |   |   |   |   |  With LRU Buffer Replacement Strategy for data pages.
   |   |   |   |
   |   |   |   |   |RESTRICT Operator (VA = 6)(5)(0)(0)(0)(0)
   |   |   |   |   |
   |   |   |   |   |   |SCAN Operator (VA = 5)
   |   |   |   |   |   |  FROM TABLE
   |   |   |   |   |   |  n_audit_table
   |   |   |   |   |   |  t
   |   |   |   |   |   |  Index : IX_n_audit_table
   |   |   |   |   |   |  Forward Scan.
   |   |   |   |   |   |  Positioning by key.
   |   |   |   |   |   |  Keys are:
   |   |   |   |   |   |    i_date ASC
   |   |   |   |   |   |    a_date ASC
   |   |   |   |   |   |  Using I/O Size 2 Kbytes for index leaf pages.
   |   |   |   |   |   |  With LRU Buffer Replacement Strategy for index leaf pages.
   |   |   |   |   |   |  Using I/O Size 2 Kbytes for data pages.
   |   |   |   |   |   |  With LRU Buffer Replacement Strategy for data pages.
   |   |
   |   |  TO TABLE
   |   |  #tmp
   |   |  Using I/O Size 2 Kbytes for data pages.


 Total estimated I/O cost for statement 2 (at line 24): 29322945.

Answer 1

我怀疑它是一个工会问题。查询是更可能的麻烦制造者。

我想你应该从临时表上添加索引开始：

create table #tmp1( f_id smallint, a_date smalldatetime ) 
Create clustered index IX1Temp on #tmp1(f_id ) 
Create clustered index IX2Temp on #tmp1(a_date ) 
...

另外，我认为在＃tmp1，＃tmp2中使用它们的方式没什么意义。你可以改为调用CTE。也。我建议你尝试使用PARTITION BY而不是GROUP BY语句。

Answer 2

根据查询执行计划，问题是表扫描临时表。

请获取以下查询的执行计划：

insert into #tmp
select t_account,t_id, product
FROM
    audit_table nt,
    (
        select f_id, a_date = max(a_date)
        FROM  audit_table
        WHERE  i_date =  @pIDate
        group by f_id
    ) a
WHERE
    i_date   =  @pIDate
    and nt.a_date = a.a_date
    and nt.f_id = a.f_id
union
select t_account,t_id, product
FROM
    n_audit_table t,
    (
        select f_id , a_date = max( a_date )
        FROM  n_audit_table
        WHERE  i_date =  @pIDate
        group by f_id
    ) a
WHERE
    t.item_date =  @pIDate
    and t.a_date = a.a_date
    and t.f_id = a.f_id

Answer 3

每个临时表中最多有多少行？看起来临时表可以使用HAVING替换，我需要测试它，当你的group by在一个列上并且输出中需要更多列时，它总是很复杂。

尝试使用SET STATISTICS PLANCOST ON和SET STATISTICS IO ON运行此语句，因为这样可以很好地了解扫描的页数以及Sybase在优化查询时是否出错。

优化Sybase 15.5联合查询

3 个答案: