Question

需要在5分钟内插入大量记录。这是我尝试过的pl / SQL，

procedure insert_student(name_      in varchar2,
                         address_   in varchar2,
                         phone_     in varchar2,
                         class_     in varchar2) is                                  

        begin

          insert into student.student_scholarship(name, address,
          phone, class, date)
          values (name_, address_, phone_, class_, sysdate)
          );
        commit;                
end insert_student;

Answer 1

对于以下模拟，我们{A}生成了一个STUDENT表，其中包含1,000,000行，并填充有随机字符串。由于您没有告诉我们要从何处加载数据，因此我们已经{B}将数据导出/卸载到了CSV文件，{C}通过EXTERNAL表使用了数据，然后利用各种INSERT技术。（全部使用Oracle 12c“ Developer Days” VM完成）

{A}“源表”

create table student ( name, address, phone, sclass )
as
select 
  dbms_random.string( 'x', 25 )
, dbms_random.string( 'x', 40 )
, dbms_random.string( 'x', 20 )
, dbms_random.string( 'x', 5 )
from dual
connect by level <= 1000000 ;
-- Elapsed: 00:03:25.032

-- quick check
select count(*) from student ;

  COUNT(*)
----------
   1000000

{B}将1,000,000行写入CSV文件

set term off
set feed off
set sqlformat csv
spool /home/oracle/data_out/out.csv
select /*+ parallel */* from student ;
spool off

{C}外部表

create table external_ ( 
  name    varchar2( 4000 )
, address varchar2( 4000 )
, phone   varchar2( 4000 )
, sclass  varchar2( 4000 )
) 
organization external (
  type oracle_loader 
  default directory external_tables
  access parameters 
  ( 
     records field names all files
     fields CSV with embedded record terminators
  ) 
  location 
  (
    'out.csv'
  ) 
)
/
-- quick check
SQL> select count(*) from external_ ;

  COUNT(*)
----------
   1000000

“目的地”表

create table scholarship (
  name    varchar2( 25 )
, address varchar2( 40 )
, phone   varchar2( 20 )
, sclass   varchar2( 5 )
, sdate   date default sysdate
);

使用纯SQL插入1,000,000行时，我们得到了以下次数（测试运行3次，两次测试之间都删除了SCHOLARSHIP表）。

-- {1}  SQL: INSERT ... SELECT ...
insert into scholarship ( name, address, phone, sclass ) 
select name, address, phone, sclass from external_ ;

-- 1,000,000 rows inserted.
-- Elapsed: 00:00:02.607
-- Elapsed: 00:00:02.300
-- Elapsed: 00:00:02.473

最糟糕的选择可能是：使用PL / SQL和CURSOR FOR LOOP（测试运行3次，两次测试之间都放弃了SCHOLARSHIP）。

--{2}  PL/SQL: use a cursor for loop ("slow by slow")
begin
  for rec_ in ( select * from external_ )
  loop
    insert into scholarship ( name, address, phone, sclass )
    values ( rec_.name, rec_.address, rec_.phone, rec_.sclass ) ;
  end loop ;
  commit ;
end ;
/

-- PL/SQL procedure successfully completed.
-- Elapsed: 00:00:24.777
-- Elapsed: 00:00:22.700
-- Elapsed: 00:00:24.291

好一点：使用PL / SQL的批量操作（再次进行3次测试的“经过时间”）。

--{3}  PL/SQL: use BULK COLLECT and FORALL (no need to re-compile in between tests)
create or replace procedure insert_students is
  type student_t is table of external_%rowtype index by pls_integer ;
  lstudents student_t ;
begin
  select * bulk collect into lstudents from external_  ;
  forall i in 1.. lstudents.count
    insert into scholarship ( name, address, phone, sclass )
    values ( lstudents( i ).name, lstudents( i ).address, lstudents( i ).phone, lstudents( i ).sclass  );
end ;
/

begin
  insert_students ;
  commit ;
end ;
/

-- PL/SQL procedure successfully completed.
-- Elapsed: 00:00:08.706
-- Elapsed: 00:00:06.762
-- Elapsed: 00:00:04.989

许多人会告诉您：尽可能使用SQL（仅）。现在，您可能会发现，最初的方法-使用带有参数的过程，并且一次仅执行一次INSERT-可能不是解决问题的最佳方法。

Answer 2

使用FORALL语句。它比使用FOR循环一个或一个地保存记录要快得多，因为它不会在每次PL / SQL处理器通过SQL语句时都更改PL / SQL和SQL之间的上下文。

CREATE TABLE students (
    id NUMBER(19,0),
    address VARCHAR2(300)
);
/

CREATE OR REPLACE PACKAGE pack AS
TYPE t_students IS TABLE OF students%ROWTYPE INDEX BY BINARY_INTEGER;

PROCEDURE insert_students( l_students IN t_students);

END pack;
/

CREATE OR REPLACE PACKAGE BODY pack AS

PROCEDURE insert_students( l_students IN t_students) AS
    BEGIN
      FORALL i IN 1..l_students.COUNT
      INSERT INTO students VALUES (l_students(i).id, l_students(i).address);
    END;
END pack;
/

如何在5分钟内在pl / sql oracle中少插入1百万个数据？

2 个答案: