merge into 和 update 的效率对比

以前只考虑 merge into 只是在特定场合下方便才使用的，今天才发现，merge into 竟然会比 update 在更新数据时有这么大的改进。
其实呢，merge into部分的update和update也没啥不同的，不同的地方在于使用merge into后执行计划变了。
赶紧测试看看，但是看到下面的结果，我和我的小伙伴惊呆了~

测试数据：
create table test1 as select * from dba_objects where rownum<=10000;--10000条记录
create table test2 as select * from dba_objects;--13438条记录

更新相同的数据，看看下面merge into和update相比性能上有何改进。
测试1：update
SQL>  alter system flush shared_pool;

系统已更改。
SQL>  alter system flush buffer_cache;

系统已更改。
SQL> set linesize 400 pagesize 400
SQL> set autot trace
SQL> set timing on
SQL> update test1 t1
  2     set t1.object_name =
  3         (select t2.object_name
  4            from test2 t2
  5           where t2.object_id = t1.object_id);

已更新10000行。

已用时间:  00: 00: 25.24

执行计划
----------------------------------------------------------
Plan hash value: 3883393169

----------------------------------------------------------------------------
| Id  | Operation          | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | UPDATE STATEMENT   |       |  9606 |   741K|   518K  (2)| 01:43:46 |
|   1 |  UPDATE            | TEST1 |       |       |            |          |
|   2 |   TABLE ACCESS FULL| TEST1 |  9606 |   741K|    40   (0)| 00:00:01 |
|*  3 |   TABLE ACCESS FULL| TEST2 |   167 | 13193 |    53   (0)| 00:00:01 |
----------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - filter("T2"."OBJECT_ID"=:B1)

Note
-----
   - dynamic sampling used for this statement (level=4)

统计信息
----------------------------------------------------------
        234  recursive calls
      10665  db block gets
   
        335  physical reads
    1631056  redo size
        685  bytes sent via SQL*Net to client
        705  bytes received via SQL*Net from client
          3  SQL*Net roundtrips to/from client
         17  sorts (memory)
          0  sorts (disk)
      10000  rows processed
测试2：merge into
SQL>  alter system flush shared_pool;

系统已更改。

已用时间:  00: 00: 00.33
SQL>  alter system flush buffer_cache;

系统已更改。

已用时间:  00: 00: 00.11
SQL> merge into test1 t1
  2  using test2 t2
  3  on (t1.object_id = t2.object_id)
  4  when matched then
  5    update set t1.object_name = t2.object_name;

10000 行已合并。

已用时间:  00: 00: 01.14

执行计划
----------------------------------------------------------
Plan hash value: 818823782

--------------------------------------------------------------------------------------
| Id  | Operation            | Name  | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------
|   0 | MERGE STATEMENT      |       |  9607 |  1238K|       |   373   (1)| 00:00:05 |
|   1 |  MERGE               | TEST1 |       |       |       |            |          |
|   2 |   VIEW               |       |       |       |       |            |          |
|*  3 |    HASH JOIN         |       |  9607 |  3996K|  2168K|   373   (1)| 00:00:05 |
|   4 |     TABLE ACCESS FULL| TEST1 |  9606 |  2054K|       |    40   (0)| 00:00:01 |
|   5 |     TABLE ACCESS FULL| TEST2 | 16669 |  3369K|       |    53   (0)| 00:00:01 |
--------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - access("T1"."OBJECT_ID"="T2"."OBJECT_ID")

Note
-----
   - dynamic sampling used for this statement (level=4)

统计信息
----------------------------------------------------------
        359  recursive calls
      10265  db block gets
       
        343  physical reads
    2725336  redo size
        685  bytes sent via SQL*Net to client
        698  bytes received via SQL*Net from client
          3  SQL*Net roundtrips to/from client
         23  sorts (memory)
          0  sorts (disk)
      10000  rows processed

具体看下面的结果：
SQL> set autot off
SQL> update /*+gather_plan_statistics*/test1 t1
  2     set t1.object_name =
  3         (select t2.object_name
  4            from test2 t2
  5           where t2.object_id = t1.object_id);

已更新10000行。

已用时间:  00: 00: 27.26
SQL> select * from table(dbms_xplan.display_cursor(null,null,'iostats'));

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------
SQL_ID  c0pc2fq4pj4zq, child number 0
-------------------------------------
update /*+gather_plan_statistics*/test1 t1    set t1.object_name =
  (select t2.object_name           from test2 t2          where
t2.object_id = t1.object_id)

Plan hash value: 3883393169

--------------------------------------------------------------------------------------
| Id  | Operation          | Name  | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
--------------------------------------------------------------------------------------
|   0 | UPDATE STATEMENT   |       |      1 |        |      0 |00:00:27.26 |    1800K|
|   1 |  UPDATE            | TEST1 |      1 |        |      0 |00:00:27.26 |    1800K|
|   2 |   TABLE ACCESS FULL| TEST1 |      1 |   9606 |  10000 |00:00:00.04 |     134 |
|*  3 |   TABLE ACCESS FULL| TEST2 |     167 |  10000 |00:00:27.03 |    1800K|
--------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - filter("T2"."OBJECT_ID"=:B1)

Note
-----
   - dynamic sampling used for this statement (level=4)

已选择26行。

SQL> merge /*+gather_plan_statistics*/into test1 t1
  2  using test2 t2
  3  on (t1.object_id = t2.object_id)
  4  when matched then
  5    update set t1.object_name = t2.object_name;

10000 行已合并。

已用时间:  00: 00: 00.25
SQL> select * from table(dbms_xplan.display_cursor(null,null,'iostats'));

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------
SQL_ID  cg8wb3hrjx2bd, child number 0
-------------------------------------
merge /*+gather_plan_statistics*/into test1 t1 using test2 t2 on
(t1.object_id = t2.object_id) when matched then   update set
t1.object_name = t2.object_name

Plan hash value: 818823782

-------------------------------------------------------------------------------------------------
| Id  | Operation            | Name  | Starts | E-Rows | A-Rows |   A-Time   | Buffers | Reads  |
-------------------------------------------------------------------------------------------------
|   0 | MERGE STATEMENT      |       |      1 |        |      0 |00:00:00.22 |   10568 |      1 |
|   1 |  MERGE               | TEST1 |      1 |        |      0 |00:00:00.22 |   10568 |      1 |
|   2 |   VIEW               |       |      1 |        |  10000 |00:00:00.05 |     314 |      0 |
|*  3 |    HASH JOIN         |       |      1 |   9607 |  10000 |00:00:00.05 |     314 |      0 |
|   4 |     TABLE ACCESS FULL| TEST1 |      1 |   9606 |  10000 |00:00:00.01 |     134 |      0 |
|   5 |     TABLE ACCESS FULL| TEST2 |      1 |  16669 |  13438 |00:00:00.01 |     180 |      0 |
-------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - access("T1"."OBJECT_ID"="T2"."OBJECT_ID")

Note
-----
   - dynamic sampling used for this statement (level=4)

已选择28行。