技术头条 - 一个快速在微博传播文章的方式     搜索本站
您现在的位置首页 --> Oracle --> (oracle)逻辑读异常(主键查询)

(oracle)逻辑读异常(主键查询)

浏览:1217次  出处信息

今天早上在查看各库情况时,发现以primary key查询,竟然消耗了5301个,按理应该是4个逻辑读或以上一点!

以下是代码片段:
select t_id,
       cat_id,
       t_top1,
       t_top2,
       t_top3,
       t_top4,
       t_top5,
       t_top6,
       t_top7,
       t_top8,
       t_top9,
       t_top10,
       t_top11,
       t_top12,
       t_top13,
       t_top14
  from t
 where t_id = :1
   and cat_id=:2      
------------------------------------------------------------------------------------------
|SELECT STATEMENT                |---------- 149755262 ----------|       |      |      4 |
|TABLE ACCESS BY INDEX ROWID     |T          |     1 |  108 |      4 |
| INDEX UNIQUE SCAN              |PK_T_ID    |     1 |      |      3 |
------------------------------------------------------------------------------------------
buffer_gets=5301.07

随后我随机取了几个值代入,奇怪的事,逻辑读正常了!这时没有删除T的sql在跑,但早上发生的那种情况,发现删除T的sql在跑。

怀疑和一致读有关,为了维护一致读,sql要不停得读取undo(oracle可能采用递归调用)。

下面测试来证实我的猜测:

以下是代码片段:
SQL> !tbsql index tmp_lx_t
                                 Column
CONSTRAINT_NAME                C COLUMN_NAME                      POSITION STATUS   VALIDATED
------------------------------ - ------------------------------ ---------- -------- -------------
PK_T                           P ID                                      1 ENABLED  VALIDATED

正常情况下:

以下是代码片段:
SQL> select * from tmp_lx_t  where id = 1178961568;
--------------------------------------------------------------------------------------------------------
| Id  | Operation                   | Name                     | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |                          |     1 |   204 |     1   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| TMP_LX_T |     1 |   204 |     1   (0)| 00:00:01 |
|*  2 |   INDEX UNIQUE SCAN         | PK_T |     1 |       |     1   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
          4  consistent gets
          0  physical reads
          0  redo size
       1580  bytes sent via SQL*Net to client
        492  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed

异常情况:

session 1:

以下是代码片段:
SQL> delete from tmp_lx_t;  表尽量大一点,延长delete操作时间,效果更明显!  

session 2:

以下是代码片段:
SQL> select * from tmp_lx_t  where id = 1178961568;   
--------------------------------------------------------------------------------------------------------   
| Id  | Operation                   | Name                     | Rows  | Bytes | Cost (%CPU)| Time     |   
--------------------------------------------------------------------------------------------------------   
|   0 | SELECT STATEMENT            |                          |     1 |   204 |     1   (0)| 00:00:01 |   
|   1 |  TABLE ACCESS BY INDEX ROWID| TMP_LX_t |     1 |   204 |     1   (0)| 00:00:01 |   
|*  2 |   INDEX UNIQUE SCAN         | PK_T |     1 |       |     1   (0)| 00:00:01 |   
--------------------------------------------------------------------------------------------------------   
Statistics  
----------------------------------------------------------   
          0  recursive calls   
          0  db block gets   
         17  consistent gets   
          0  physical reads   
         64  redo size  
       1580  bytes sent via SQL*Net to client   
        492  bytes received via SQL*Net from client   
          2  SQL*Net roundtrips to/from client   
          0  sorts (memory)   
          0  sorts (disk)   
          1  rows processed  

逻辑读发生明显的变化,go on

以下是代码片段:
SQL> select * from tmp_lx_t  where id = 1178961568;   
--------------------------------------------------------------------------------------------------------   
| Id  | Operation                   | Name                     | Rows  | Bytes | Cost (%CPU)| Time     |   
--------------------------------------------------------------------------------------------------------   
|   0 | SELECT STATEMENT            |                          |     1 |   204 |     1   (0)| 00:00:01 |   
|   1 |  TABLE ACCESS BY INDEX ROWID| TMP_LX_t |     1 |   204 |     1   (0)| 00:00:01 |   
|*  2 |   INDEX UNIQUE SCAN         | PK_T |     1 |       |     1   (0)| 00:00:01 |   
--------------------------------------------------------------------------------------------------------   
Statistics  
----------------------------------------------------------   
          0  recursive calls   
          0  db block gets   
         38  consistent gets   
          0  physical reads   
         64  redo size  
       1580  bytes sent via SQL*Net to client   
        492  bytes received via SQL*Net from client   
          2  SQL*Net roundtrips to/from client   
          0  sorts (memory)   
          0  sorts (disk)   
          1  rows processed  

逻辑读不停再发生变化!!!

小结:在删除表中记录时,删除的记录刚好是符合查询条件的记录,oracle为了构造一致性读,必须读取undo的信息,所以导致逻辑读异常偏高!!!

QQ技术交流群:445447336,欢迎加入!
扫一扫订阅我的微信号:IT技术博客大学习
© 2009 - 2024 by blogread.cn 微博:@IT技术博客大学习

京ICP备15002552号-1