如何加速Oracle SQLDeveloper上的REGEXP LEVEL查询

时间:2017-11-16 19:41:31

标签: sql regex oracle performance insert-into

我有一个INSERT INTO SELECT语句,用源表中的;解析的值填充表:

INSERT INTO PC_MATERIALS_BRIDGE (MATERIAL_BRIDGE_ID, VARIABLE_ID, MATERIAL_NAME)
   SELECT PC_VAR_MATERIALS_BRIDGE_SEQ.NEXTVAL, VARIABLE_ID, MATERIAL_NAME FROM (SELECT DISTINCT E.VARIABLE_ID, LOWER(TRIM(REGEXP_SUBSTR(e.MATERIALS, '[^;]+', 1, LEVEL))) MATERIAL_NAME
        FROM (SELECT VARIABLE_ID, MATERIALS FROM SRC_VARS_OCEAN_ALL WHERE MATERIALS IS NOT NULL AND MATERIALS != 'N/A) e
        CONNECT BY LOWER(TRIM(REGEXP_SUBSTR(e.MATERIALS, '[^;]+', 1, LEVEL))) IS NOT NULL);

所以,源表中的数据

ID     MATERIAL_NAME
1      paper
2      paper; plastic

将显示为

MATERIAL_BRIDGE_ID     MATERIAL_NAME   
1                      paper
2                      paper
3                      plastic

在目标表中。

脚本运行良好;然而,它非常昂贵,因为源表有近40,000条记录,有些有三个值,例如paper; plastic; rubber。我知道LEVEL很贵。我将MATERIAL_NAME设置为VARCHAR2(255 BYTE)。不知道除了编写其他类型的查询之外如何改进(例如,递归但它可能很难)。 DISTINCT也导致它减速吗?可能不再需要DISTINCT,因为e.VARIABLE_ID现在是主键。

1 个答案:

答案 0 :(得分:2)

这是一种非常低效的方法。在下面的简单演示中删除DISTINCT时,您可以观察它导致问题的原因:

create table SRC_VARS_OCEAN_ALL(
  VARIABLE_ID int, 
  MATERIALS varchar2(200)
);

insert into SRC_VARS_OCEAN_ALL values( 1, 'ala;ma;kota' );
insert into SRC_VARS_OCEAN_ALL values( 2, 'as;to;pies' );
insert into SRC_VARS_OCEAN_ALL values( 3, 'baba;jaga' );
insert into SRC_VARS_OCEAN_ALL values( 4, 'zupa;obiad' );

SELECT  E.VARIABLE_ID, level,
        LOWER(TRIM(REGEXP_SUBSTR(e.MATERIALS, '[^;]+', 1, LEVEL))) MATERIAL_NAME
FROM (
    SELECT VARIABLE_ID, MATERIALS 
    FROM SRC_VARS_OCEAN_ALL 
    WHERE MATERIALS IS NOT NULL 
    AND MATERIALS != 'N/A'
) e
CONNECT BY LOWER(TRIM(REGEXP_SUBSTR(e.MATERIALS, '[^;]+', 1, LEVEL))) IS NOT NULL
order by 1,2;

VARIABLE_ID      LEVEL MATERIAL_NAME     
----------- ---------- -----------------
          1          1 ala               
          1          2 ma                
          1          2 ma                
          1          2 ma                
          1          2 ma                
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          1          3 kota              
          2          1 as                
          2          2 to                
          2          2 to                
          2          2 to                
          2          2 to                
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          2          3 pies              
          3          1 baba              
          3          2 jaga              
          3          2 jaga              
          3          2 jaga              
          3          2 jaga              
          4          1 zupa              
          4          2 obiad             
          4          2 obiad             
          4          2 obiad             
          4          2 obiad             

52 rows selected. 

此查询仅为4个输入行生成 52个输出记录,其中包含10个值。你可以猜到4万会有多少  该查询生成数百个thausands甚至数百万行,然后DISTINCT对这个巨大的结果集进行排序,以消除重复。

以下查询应该执行得更好,因为它只生成10条记录,不多或少,只需执行此任务所需的记录:

SELECT  a.VARIABLE_ID, b.lev_el,
       trim( regexp_substr( a.MATERIALS, '[^;]+', 1, b.lev_el )) as MATERIAL_NAME
FROM SRC_VARS_OCEAN_ALL a
JOIN (
  SELECT level as lev_el
  FROM dual CONNECT BY level <= 100
) b
ON b.lev_el <= regexp_count( a.MATERIALS, ';' ) + 1

VARIABLE_ID     LEV_EL MATERIAL_NAME 
----------- ---------- --------------
          1          1 ala           
          2          1 as            
          3          1 baba          
          4          1 zupa          
          1          2 ma            
          2          2 to            
          3          2 jaga          
          4          2 obiad         
          1          3 kota          
          2          3 pies          

10 rows selected. 

我假设每个列表中的值不超过100个(每个单独的行都有一个不超过100个值的列表),因此有FROM dual CONNECT BY level <= 100子句。