昨晚更新了一批数据,用update的老办法耗时20多分,而用 merge into 不到2秒结束,效率真是天壤之别。具体见下:
用T_TMP_SCHOOL(135868行)的BIRTH 字段更新T_TMP_NT_CUSTOMERDETAIL( 763119行) 的BIRTHDATE 字段,连接条件 T_TMP_SCHOOL.ID = t_tmp_nt_customerdetail.SCHOOLID
--表结构
create table T_TMP_NT_CUSTOMERDETAIL
(
CUSTOMERID VARCHAR2(15) not null,
DOCCATEGORY VARCHAR2(2) not null,
DOCNUMBER VARCHAR2(20) not null,
BIRTHDATE VARCHAR2(8),
...........
SCHOOLID VARCHAR2(60)
);
create table T_TMP_SCHOOL
(
ID VARCHAR2(20),
COMPANY VARCHAR2(100),
NAME VARCHAR2(20),
BIRTH VARCHAR2(20)
);
--两个表的数据见下:
select count(1) from t_tmp_nt_customerdetail t; --763119
select count(1) from t_tmp_school; --135868
--为了验证结果,测试前先清空birthdate的值,共更改 135879 行
update t_tmp_nt_customerdetail t
set t.birthdate = null
where t.schoolid is not null;
---实现的过程:
create or replace procedure p_tmp_update_customerdetail
is
v_BeginTran INT := 0; -- 事务标志,初始值为0,表示没有事务
v_ErrCode INT;
v_ErrMsg VARCHAR2(200); -- 处理异常变量
begin
-- 设置事务标志为1,表示开始事务
v_BeginTran := 1;
merge into t_tmp_nt_customerdetail t
using (select b.id, b.birth from t_tmp_school b where b.birth is not null) a
on (t.schoolid = a.id)
when matched then
update set t.birthdate = a.birth where t.schoolid is not null;
COMMIT;
-- 提交事务并且置事务标志为0。
v_BeginTran := 0;
EXCEPTION
WHEN OTHERS THEN
-- 如果异常,回滚事务。
IF v_BeginTran = 1 THEN
ROLLBACK;
END IF;
v_ErrCode := SQLCODE;
v_ErrMsg := SUBSTR(SQLERRM, 1, 150);
dbms_output.put_line(v_ErrCode);
dbms_output.put_line(v_ErrMsg);
end;
--执行过程,用时1.11秒
SQL> exec p_tmp_update_customerdetail;
--再次验证结果,先前清空birthdate的值已经有了,返回 135879 行
select count(1) from t_tmp_nt_customerdetail t
where t.schoolid is not null
and t.birthdate is not null;
--而用下面类似的语句,这些数据执行了24分钟多:
update t_tmp_nt_customerdetail t
set t.birthdate = (select b.birth
from t_tmp_school b
where t.schoolid = b.id)
where t.schoolid =
(select c.id from t_tmp_school c where t.schoolid = c.id)
and t.schoolid is not null; ---注:为什么要写这个罗嗦的条件呢?因为没有这个条件就把整个表的数据全部更新了,因此必须写,所以大家应该多实践,不要被一些表面现象所蒙蔽。
转载于:https://blog.51cto.com/srsunbing/1130337