Oracle update性能优化

  当执行批量刷新数据时，以前我写过最好是写成merge into，当然还有一种方式，如下：

SQL> create table test1 as select * from dba_objects where rownum <100;

SQL> create table test2 as select * from dba_objects where rownum <1000;
SQL> create unique index ind_t1_object_id on test1(object_id);
SQL> create unique index ind_t2_object_id on test2(object_id);
SQL> exec dbms_stats.gather_table_stats(user,'test1');
SQL> exec dbms_stats.gather_table_stats(user,'test2');
SQL> set autotrace traceonly
SQL> update test1 t1
       set t1.object_name = (select t2.object_name
                               from test2 t2
                              where t1.object_id = t2.object_id)
     where exists (select 1 from test2 t3 where t3.object_id = t1.object_id);--千万不能丢掉exists，否则很多匹配不上的记录object_name就为null。
已更新99行。
执行计划
----------------------------------------------------------
Plan hash value: 1549919212
-------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name             | Rows  | Bytes | Cost (%CPU)| Time     |
-------------------------------------------------------------------------------------------------
|   0 | UPDATE STATEMENT             |                  |    99 |  1584 |   304  (33)| 00:00:04 |
|   1 |  UPDATE                      | TEST1            |       |       |            |          |
|*  2 |   HASH JOIN SEMI             |                  |    99 |  1584 |     7  (15)| 00:00:01 |
|   3 |    TABLE ACCESS FULL         | TEST1            |    99 |  1188 |     3   (0)| 00:00:01 |
|   4 |    INDEX FAST FULL SCAN      | IND_T2_OBJECT_ID |   999 |  3996 |     3   (0)| 00:00:01 |
|   5 |   TABLE ACCESS BY INDEX ROWID| TEST2            |     1 |    20 |     2   (0)| 00:00:01 |
|*  6 |    INDEX RANGE SCAN          | IND_T2_OBJECT_ID |     1 |       |     1   (0)| 00:00:01 |
-------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   2 - access("T3"."OBJECT_ID"="T1"."OBJECT_ID")
   6 - access("T2"."OBJECT_ID"=:B1)
统计信息
----------------------------------------------------------
          3  recursive calls
        103  db block gets
      217  consistent gets
          0  physical reads
      23656  redo size
        559  bytes sent via SQL*Net to client
        922  bytes received via SQL*Net from client
          3  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
         99  rows processed


SQL> commit;
SQL> update (select t1.object_name, t2.object_name new_object_name
              from test1 t1, test2 t2
             where t1.object_id = t2.object_id)
    set object_name = new_object_name;
已更新99行。
执行计划
----------------------------------------------------------
Plan hash value: 1124869545
-----------------------------------------------------------------------------
| Id  | Operation           | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------
|   0 | UPDATE STATEMENT    |       |    99 |  3168 |    10  (10)| 00:00:01 |
|   1 |  UPDATE             | TEST1 |       |       |            |          |
|*  2 |   HASH JOIN         |       |    99 |  3168 |    10  (10)| 00:00:01 |
|   3 |    TABLE ACCESS FULL| TEST1 |    99 |  1188 |     3   (0)| 00:00:01 |
|   4 |    TABLE ACCESS FULL| TEST2 |   999 | 19980 |     6   (0)| 00:00:01 |
-----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   2 - access("T1"."OBJECT_ID"="T2"."OBJECT_ID")
统计信息
----------------------------------------------------------
          3  recursive calls
        103  db block gets
       25  consistent gets
          0  physical reads
      23736  redo size
        561  bytes sent via SQL*Net to client
        858  bytes received via SQL*Net from client
          3  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
         99  rows processed

SQL> commit;

  总结：可以看到，第二种写法比第一种写法少扫描一张表，逻辑读也少了很多。