分页查询的优化

今天总结一下分页查询，下面几个链接是yangtingkun blog上的测试，写的很详细!

Oracle分页查询语句（1）

[url]http://yangtingkun.itpub.net/post/468/100278[/url]

Oracle分页查询语句（2）

[url]http://yangtingkun.itpub.net/post/468/101703[/url]

Oracle分页查询语句（3）

[url]http://yangtingkun.itpub.net/post/468/104595[/url]

Oracle分页查询语句（4）

[url]http://yangtingkun.itpub.net/post/468/104867[/url]

 

下面是对优化的进一步总结和补充：

首先，创建测试表

create table t2 as select * from dba_objects;

create idx idx-t2 on t2(object_id);

最早使用分页方法：

SQL> set autotrace on
SQL> /

        RN  OBJECT_ID OBJECT_NAME
---------- ---------- ----------------------------------------------------------
        96         97 ACCESS$
        97         98 AUD$
        98         99 LINK$
        99        100 TRUSTED_LIST$
       100        101 PROPS$

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE
   1    0   VIEW
   2    1     COUNT
   3    2       VIEW
   4    3         TABLE ACCESS (BY INDEX ROWID) OF 'T2'
   5    4           INDEX (RANGE SCAN) OF 'IDX_T2' (NON-UNIQUE)

 

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
        346  consistent gets
          0  physical reads
          0  redo size
        604  bytes sent via SQL*Net to client
        504  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          5  rows processed

 

我们发现，逻辑读为346 ，做第一次改进，将rn<=100移到第二层循环：

SQL> select rn,object_id,object_name from
  2  (
  3  select rownum rn,a.* from
  4   (
  5     select * from t2   where object_id<500 order by object_id
  6   ) a where rownum<100
  7   )
  8   where rn>95;

        RN  OBJECT_ID OBJECT_NAME
---------- ---------- ----------------------------------------------------------
        96         97 ACCESS$
        97         98 AUD$
        98         99 LINK$
        99        100 TRUSTED_LIST$

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE
   1    0   VIEW
   2    1     COUNT (STOPKEY)
   3    2       VIEW
   4    3         TABLE ACCESS (BY INDEX ROWID) OF 'T2'
   5    4           INDEX (RANGE SCAN) OF 'IDX_T2' (NON-UNIQUE)

 

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
         63  consistent gets
          0  physical reads
          0  redo size
        585  bytes sent via SQL*Net to client
        504  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          4  rows processed

我们发现，逻辑读降到了63，执行计划中用到了stopkey，也就是排序的时候只排到前100条就停止。

上面的改进中我们发现，排序虽然只排到了前100个，但是仍然有95条记录是做的无用功（因为只输出5条记录），这样，翻的页数越大，做的无用功越多，效率自然也越低，这时，我们可以做进一步的改进：

SQL> select rn,t2.object_id,object_name from t2,
  2  (
  3  select * from
  4  (
  5  select rownum rn,a.* from
  6  (
  7   select rowid rd,object_id from t2  where object_id<500 order by object_id
  8   ) a where rownum<100
  9   ) where rn>95
 10   )c
 11   where t2.rowid=c.rd;

        RN  OBJECT_ID OBJECT_NAME
---------- ---------- ----------------------------------------------------------
        96         97 ACCESS$
        97         98 AUD$
        98         99 LINK$
        99        100 TRUSTED_LIST$

Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE
   1    0   NESTED LOOPS
   2    1     VIEW
   3    2       COUNT (STOPKEY)
   4    3         VIEW
   5    4           INDEX (RANGE SCAN) OF 'IDX_T2' (NON-UNIQUE)
   6    1     TABLE ACCESS (BY USER ROWID) OF 'T2'

 

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
          7  consistent gets
          0  physical reads
          0  redo size
        585  bytes sent via SQL*Net to client
        504  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          4  rows processed

 

此时，我们直接对索引进行排序后的提取，然后用rowid和原表进行表关联，因为索引是拍好序的，因此对索引的rowid的提取非常快，提取出的 rowid和原表做nested loop也非常快（rowid查询是最快的方法），因此，我们发现，逻辑读居然降到了只有7，大大提高了分页的查询速度。