Oracle 10g大表批量更新优化,其实,700万的表不算大表,作为测试够了
一,t表信息
SQL> alter table t add is_del number(1);
SQL> alter table t modify is_del default 0;
SQL> desc t
Name Type Nullable Default Comments
------ --------- -------- ------- --------
ID NUMBER Y
CODE NUMBER Y
IS_DEL NUMBER(1) Y 0
SQL> select count(*) from t;
COUNT(*)
----------
7136976
二,为了比较基准的一致性,先缓存t数据
update t set t.is_del = 0;
三,这里共总结了4种方法
SQL> set timing on
--0
SQL> update t set t.is_del = 0;
7136976 rows updated.
Elapsed: 00:08:28.64
--1
SQL> declare
2 rnt pls_integer := 0;
3 begin
4 for idx in (select rowid rid from t) loop
5 update t set t.is_del = 0 where rowid = idx.rid;
6 rnt := rnt + 1;
7 if mod(rnt,2000) = 0 then
8 commit;
9 end if;
10 end loop;
11 commit;
12 end;
13 /
PL/SQL procedure successfully completed.
Elapsed: 00:09:41.32
SQL>
--2
SQL> declare
2 rnt pls_integer := 0;
3 begin
4 for idx in (select rowid rid from t) loop
5 update t set t.is_del = 0 where rowid = idx.rid;
6 rnt := rnt + 1;
7 if rnt = 2000 then
8 rnt := 0;
9 commit;
10 end if;
11 end loop;
12 commit;
13 end;
14 /
PL/SQL procedure successfully completed.
Elapsed: 00:09:35.67
--3
SQL> declare
2 cursor cur_t is select rowid rid from t;
3 type tab_t is table of urowid index by binary_integer;
4 l_rid tab_t;
5 begin
6 open cur_t;
7 loop
8 fetch cur_t bulk collect into l_rid limit 2000;
9 forall idx in 1 .. l_rid.count
10 update t set t.is_del = 0 where rowid = l_rid(idx);
11 commit;
12 exit when cur_t%notfound;
13 end loop;
14 close cur_t;
15 end;
16 /
PL/SQL procedure successfully completed.
Elapsed: 00:06:48.84
通过上面的测试结果可以看到,方法3最好,方法0不建议使用,这会使undo快速增长,出现ora-01555错误。方法1和方法2在一些书籍上看到过测试,说方法2优于方法1,但我这次测试效果不明显,以后再进行一些测试。
注:
测试的数据库配置了闪回特性,db_recovery_file_dest_size=2g,归档日志放在db_recovery_file_dest目录中。开始时的更新操作,redo增长很快,常常hang住了,alert log报空间不足,所有增加了db_recovery_file_dest_size=4g。还有要注意undo表空间的监控。