[转帖]date列统计信息陈旧导致sql没有选择最优执行计划_MySQL, Oracle及数据库讨论区_Weblogic技术|Tuxedo技术|中间件技术|Oracle论坛|JAVA论坛|Linux/Unix技术|hadoop论坛_联动北方技术论坛  
网站首页 | 关于我们 | 服务中心 | 经验交流 | 公司荣誉 | 成功案例 | 合作伙伴 | 联系我们 |
联动北方-国内领先的云技术服务提供商
»  游客             当前位置:  论坛首页 »  自由讨论区 »  MySQL, Oracle及数据库讨论区 »
总帖数
1
每页帖数
101/1页1
返回列表
0
发起投票  发起投票 发新帖子
查看: 4173 | 回复: 0   主题: [转帖]date列统计信息陈旧导致sql没有选择最优执行计划        下一篇 
derek
注册用户
等级:中校
经验:1550
发帖:209
精华:0
注册:2011-7-21
状态:离线
发送短消息息给derek 加好友    发送短消息息给derek 发消息
发表于: IP:您无权察看 2011-8-1 21:41:37 | [全部帖] [楼主帖] 楼主

开发人员报告一条sql运行的比较慢,请求优化,该sql结构如下

select count(s.id) as orderNum,sum(s.amount) as amount
from  kk s
inner join  ff af
on s.id = af.addrr_id
where s.create_time > sysdate -60
and af.t_num =10612711;


首先说一下这两个表大致情况,kk有1千万条记录,ff有两千万;
kk上的id为主键,create_time建有索引;ff上的addrr_id和t_num分别有索引
先查看一下执行计划
该sql 先访问kk上的create_time字段,然后跟 ff的结果集做nested loop;但是outer table的结果集候选记录太多(60天以内的所有记录),导致��价过高;
理想情况应该是先通过t_num找出ff的数据集,然后根据s.id = af.addrr_id条件访问kk的id字段(unique index scan),这样以来整个sql的执行代价应该会变小很多

--------------------------------------------------------------------------------------------------------------
| Id  | Operation                      | Name                        | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                                         |                             |     1 |    30 |    10   (0)| 00:00:01 |
|   1 |  KKRT AGGREGATE                                            |                                                          |     1 |    30 |            |          |
|*  2 |   TABLE ACCESS BY INDEX ROWID               | FF                                                    |     1 |    11 |     3   (0)| 00:00:01 |
|   3 |    NESTED LOOPS                                               |                                                         |     2 |    60 |    10   (0)| 00:00:01 |
|   4 |     TABLE ACCESS BY INDEX ROWID              | KK                                                  |     2 |    38 |     4   (0)| 00:00:01 |
|*  5 |      INDEX RANGE SCAN                                     | IDX_KK_CREATE_DATE          |     2 |       |     3   (0)| 00:00:01 |
|*  6 |     INDEX RANGE SCAN                                      | IDX_ADDRR_ID                          |     1 |       |     2   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("AF"."T_NUM"=10612711)
5 - access("S"."CREATE_TIME">SYSDATE@!-60)
6 - access("S"."ID"="AF"."ADDRR_ID")
Statistics
----------------------------------------------------------
493  recursive calls
0  db block gets
1535934  consistent gets
0  physical reads
1036  redo size
579  bytes sent via SQL*Net to client
487  bytes received via SQL*Net from client
2  SQL*Net roundtrips to/from client
12  kkrts (memory)
0  kkrts (disk)
1  rows processed


尝试添加hint,

SQL> select count(s.id) as orderNum,sum(s.amount) as amount
2           from  kk s
3          inner join  ff af
4             on s.id = af.addrr_id
5          where s.create_time > sysdate -60
6            and af.t_num =10612711;
-----------------------------------------------------------------------------------------------------
| Id  | Operation                     | Name                | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT              |                     |     1 |    30 |   218   (0)| 00:00:03 |
|   1 |  KKRT AGGREGATE               |                     |     1 |    30 |            |          |
|   2 |   NESTED LOOPS                |                     |     2 |    60 |   218   (0)| 00:00:03 |
|   3 |    TABLE ACCESS BY INDEX ROWID| FF                  |   143 |  1573 |    57   (0)| 00:00:01 |
|*  4 |     INDEX RANGE SCAN          | IDX_T_NUM           |   148 |       |     3   (0)| 00:00:01 |
|*  5 |    TABLE ACCESS BY INDEX ROWID| KK                  |     1 |    19 |     2   (0)| 00:00:01 |
|*  6 |     INDEX UNIQUE SCAN         | PK7                 |     1 |       |     1   (0)| 00:00:01 |
-----------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
4 - access("AF"."T_NUM"=10612711)
5 - filter("S"."CREATE_TIME">SYSDATE@!-60)
6 - access("S"."ID"="AF"."ADDRR_ID")
Statistics
----------------------------------------------------------
0  recursive calls
0  db block gets
3  consistent gets
0  physical reads
0  redo size
579  bytes sent via SQL*Net to client
487  bytes received via SQL*Net from client
2  SQL*Net roundtrips to/from client
0  kkrts (memory)
0  kkrts (disk)
1  rows processed


仅从显示的信息来看,采用pk7索引扫描性能提升的简直不可思议,但是运行的时候两条语句的执行时间相差并不是太大,尽管后者更快一点。
这里有两点疑问
1、采用pk7之后的代价小的有点离谱,且运行时间并没有比不加hint之前提升多少
2、为什么数据库自己不能自动访问到pk7,逻辑备库上同样的sql可以自动采用pk7
查询表ff 上t_num=10612711的候选记录,结果为0,则第一个疑问可以得到解释,当添加hint后,sql首先查找FF里的候选记录,返回0行,以此做nest loop,逻辑读肯定非常之低

SQL> select count(*) from ff where t_num =10612711;
COUNT(*)
----------
0


对于第二个疑问,可以通过10053事件找出答案
以下是摘录

***************************************
SINGLE TABLE ACCESS PATH
-----------------------------------------
BEGIN Single Table Cardinality Estimation
-----------------------------------------
Column (#24): CREATE_TIME(DATE)
AvgLen: 8.00 NDV: 98285 Nulls: 0 Density: 1.0174e-05 Min: 2454786 Max: 2455582
Using prorated density: 9.5809e-06 of col #24 as selectivity of out-of-range value pred
Table: KK  Alias: S
Card: Original: 241910  Rounded: 2  Computed: 2.32  Non Adjusted: 2.32
-----------------------------------------
END   Single Table Cardinality Estimation
-----------------------------------------
Access Path: TableScan
Cost:  6475.05  Resp: 6475.05  Degree: 0
Cost_io: 6424.00  Cost_cpu: 437390807
Resp_io: 6424.00  Resp_cpu: 437390807
Using prorated density: 9.5809e-06 of col #24 as selectivity of out-of-range value pred
Access Path: index (RangeScan)
Index: IDX_KK_CREATE_DATE
resc_io: 4.00  resc_cpu: 30916
ix_sel: 9.5809e-06  ix_sel_with_filters: 9.5809e-06
Cost: 4.00  Resp: 4.00  Degree: 1
Best:: AccessPath: IndexRange  Index: IDX_KK_CREATE_DATE
Cost: 4.00  Degree: 1  Resp: 4.00  Card: 2.32  Bytes: 0  --kk表的最佳访问路径为索引IDX_KK_CREATE_DATE,且cost为4,有点太低了,但是 根据公式cost=blevel +ceiling(leaf_blocks * effective index selectivity) --+ceiling(clustering_factor * effective table selectivity) = 2 + (2406*9.5809e-06) +(95454*9.5809e-06) =4,确实结果为4
***************************************
SINGLE TABLE ACCESS PATH
-----------------------------------------
BEGIN Single Table Cardinality Estimation
-----------------------------------------
Column (#3): T_NUM(NUMBER)
AvgLen: 5.00 NDV: 105 Nulls: 0 Density: 9.6246e-05 Min: 714 Max: 8215831
Histogram: Freq  #Bkts: 104  UncompBkts: 5195  EndPtVals: 104
Using prorated density: 6.8165e-05 of col #3 as selectivity of out-of-range value pred
Table: FF  Alias: AF
Card: Original: 2029216  Rounded: 138  Computed: 138.32  Non Adjusted: 138.32
-----------------------------------------
END   Single Table Cardinality Estimation
-----------------------------------------
Access Path: TableScan
Cost:  31755.82  Resp: 31755.82  Degree: 0
Cost_io: 31602.00  Cost_cpu: 1317927216
Resp_io: 31602.00  Resp_cpu: 1317927216
Using prorated density: 6.8165e-05 of col #3 as selectivity of out-of-range value pred
Access Path: index (AllEqRange)
Index: IDX_AOI_T_NUM
resc_io: 53.00  resc_cpu: 431646
ix_sel: 6.8165e-05  ix_sel_with_filters: 6.8165e-05
Cost: 53.05  Resp: 53.05  Degree: 1
Best:: AccessPath: IndexRange  Index: IDX_AOI_T_NUM
Cost: 53.05  Degree: 1  Resp: 53.05  Card: 138.32  Bytes: 0--ff表的最佳访问路径为索引IDX_AOI_T_NUM,cost为53.05
***************************************
OPTIMIZER STATISTICS AND COMPUTATIONS
***************************************
GENERAL PLANS
***************************************
Considering cardinality-based initial join order.
Permutations for Starting Table :0
***********************
Join order[1]:  KK[S]#0  FF[AF]#1--oracle在对两表的候选数据做join的时候,选择了kk表作为driving table,且kk表的访问路径为索引IDX_KK_CREATE_DATE,这个就能很好的解释为什么不会用到pk7
***************
Now joining: FF[AF]#1
***************
NL Join
Outer table: Card: 2.32  Cost: 4.00  Resp: 4.00  Degree: 1  Bytes: 19
Inner table: FF  Alias: AF
Access Path: TableScan
NL Join:  Cost: 63513.65  Resp: 63513.65  Degree: 1
Cost_io: 63206.00  Cost_cpu: 2635885348
Resp_io: 63206.00  Resp_cpu: 2635885348
Access Path: index (AllEqJoinGuess)
Index: IDX_ADDRR_ID
resc_io: 3.00  resc_cpu: 23045
ix_sel: 4.9348e-07  ix_sel_with_filters: 4.9348e-07
NL Join: Cost: 10.01  Resp: 10.01  Degree: 1
Cost_io: 10.00  Cost_cpu: 77005
Resp_io: 10.00  Resp_cpu: 77005
Using prorated density: 6.8165e-05 of col #3 as selectivity of out-of-range value pred
Access Path: index (AllEqJoin)
Index: IDX_AOI_T_NUM
resc_io: 52.00  resc_cpu: 425375
ix_sel: 6.8165e-05  ix_sel_with_filters: 6.8165e-05
NL Join: Cost: 108.10  Resp: 108.10  Degree: 1
Cost_io: 108.00  Cost_cpu: 881666
Resp_io: 108.00  Resp_cpu: 881666
Best NL cost: 10.01
resc: 10.01 resc_io: 10.00 resc_cpu: 77005
resp: 10.01 resp_io: 10.00 resp_cpu: 77005


sql选择非最优执行计划的原因为:CBO错误估计了kk表上IDX_KK_CREATE_DATE的访问代价;
其中Using prorated density: 9.5809e-06 of col #24 as selectivity of out-of-range value pred是关键,
当where条件中的值超出了统计信息里的该列的上下限范围后,oracle将采用一个估计的selectivity,查看表kk的统计信息

SQL> select to_char(last_analyzed,'yyyy-mm-dd hh24:mi:ss') from user_tab_columns where table_name ='KK' and column_name ='CREATE_TIME';
TO_CHAR(LAST_ANALYZED,'YYYY-MM
------------------------------
2011-01-19 15:43:15
 最后1月19号收集的,而现在是11年5月11号,中间差了100多天,当使用create_time > sysdate -60时,无怪乎会报告 as selectivity of out-of-range value pred。

修改一下查询条件,将create_time > sysdate-60 改为sysdate -160,重新用10053跟踪一把,以下是摘录

SINGLE TABLE ACCESS PATH
  -----------------------------------------
  BEGIN Single Table Cardinality Estimation
  -----------------------------------------
  Column (#24): CREATE_TIME(DATE)
AvgLen: 8.00 NDV: 98285 Nulls: 0 Density: 1.0174e-05 Min: 2454786 Max: 2455582
Table: KK  Alias: S
Card: Original: 241910  Rounded: 14586  Computed: 14585.52  Non Adjusted: 14585.52
-----------------------------------------
END   Single Table Cardinality Estimation
-----------------------------------------
Access Path: TableScan
Cost:  6475.05  Resp: 6475.05  Degree: 0
Cost_io: 6424.00  Cost_cpu: 437390807
Resp_io: 6424.00  Resp_cpu: 437390807
Access Path: index (RangeScan)
Index: IDX_KK_CREATE_DATE
resc_io: 6033.00  resc_cpu: 54886848
ix_sel: 0.060293  ix_sel_with_filters: 0.060293
Cost: 6039.41  Resp: 6039.41  Degree: 1
Best:: AccessPath: IndexRange  Index: IDX_KK_CREATE_DATE


         Cost: 6039.41  Degree: 1  Resp: 6039.41  Card: 14585.52  Bytes: 0—最优访问路径依旧是IDX_KK_CREATE_DATE,但cost为6039,不再是4,也没有了那句Using prorated density:

***************************************
SINGLE TABLE ACCESS PATH
-----------------------------------------
BEGIN Single Table Cardinality Estimation
-----------------------------------------
Column (#3): T_NUM(NUMBER)
AvgLen: 5.00 NDV: 103 Nulls: 0 Density: 9.3440e-05 Min: 1787 Max: 8215831
Histogram: Freq  #Bkts: 102  UncompBkts: 5351  EndPtVals: 102
Using prorated density: 6.6174e-05 of col #3 as selectivity of out-of-range value pred
Table: FF  Alias: AF
Card: Original: 2118837  Rounded: 140  Computed: 140.21  Non Adjusted: 140.21
-----------------------------------------
END   Single Table Cardinality Estimation
-----------------------------------------
Access Path: TableScan
Cost:  32187.65  Resp: 32187.65  Degree: 0
Cost_io: 32030.00  Cost_cpu: 1350695253
Resp_io: 32030.00  Resp_cpu: 1350695253
Using prorated density: 6.6174e-05 of col #3 as selectivity of out-of-range value pred
Access Path: index (AllEqRange)
Index: IDX_AOI_T_NUM
resc_io: 52.00  resc_cpu: 425305
ix_sel: 6.7504e-05  ix_sel_with_filters: 6.7504e-05
Cost: 52.05  Resp: 52.05  Degree: 1
Best:: AccessPath: IndexRange  Index: IDX_AOI_T_NUM
Cost: 52.05  Degree: 1  Resp: 52.05  Card: 140.21  Bytes: 0
***************************************
OPTIMIZER STATISTICS AND COMPUTATIONS
***************************************
GENERAL PLANS
***************************************
Considering cardinality-based initial join order.
Permutations for Starting Table :0
***********************
Join order[1]:  FF[AF]#0  KK[S]#1—CBO选择表FF作为outer table
***************
Now joining: KK[S]#1
***************
NL Join
Outer table: Card: 140.21  Cost: 52.05  Resp: 52.05  Degree: 1  Bytes: 11
Inner table: KK  Alias: S
Access Path: TableScan
NL Join:  Cost: 906319.17  Resp: 906319.17  Degree: 1
Cost_io: 899172.00  Cost_cpu: 61235138246
Resp_io: 899172.00  Resp_cpu: 61235138246
Access Path: index (UniqueScan)
Index: PK7
resc_io: 2.00  resc_cpu: 17253
ix_sel: 4.1338e-06  ix_sel_with_filters: 4.1338e-06
NL Join: Cost: 208.26  Resp: 208.26  Degree: 1
Cost_io: 208.04  Cost_cpu: 1820368
Resp_io: 208.04  Resp_cpu: 1820368
Access Path: index (RangeScan)
Index: IDX_KK_CREATE_DATE
resc_io: 6032.00  resc_cpu: 54880576
ix_sel: 0.060293  ix_sel_with_filters: 0.060293
NL Join: Cost: 845428.82  Resp: 845428.82  Degree: 1
Cost_io: 844532.00  Cost_cpu: 7683705956
Resp_io: 844532.00  Resp_cpu: 7683705956
Access Path: index (AllEqUnique)
Index: PK7
resc_io: 2.00  resc_cpu: 17253
ix_sel: 4.1338e-06  ix_sel_with_filters: 4.1338e-06
NL Join: Cost: 208.26  Resp: 208.26  Degree: 1
Cost_io: 208.04  Cost_cpu: 1820368
Resp_io: 208.04  Resp_cpu: 1820368
Best NL cost: 208.26
resc: 208.26 resc_io: 208.04 resc_cpu: 1820368
resp: 208.26 resp_io: 208.04 resp_cpu: 1820368
Join Card:  140.21 = outer (140.21) * inner (14585.52) * sel (6.8559e-05)
Join Card - Rounded: 140 Computed: 140.21
………………….
Current SQL statement for this session:
select  count(s.id) as orderNum,sum(s.amount) as amount
from  kk s
inner join  ff af
on s.id = af.addrr_id
where s.create_time > sysdate -160
and af.t_num =10612711
============
Plan Table
============
------------------------------------------------------------+-----------------------------------+
| Id  | Operation                      | Name               | Rows  | Bytes | Cost  | Time      |
------------------------------------------------------------+-----------------------------------+
| 0   | SELECT STATEMENT               |                    |       |       |   208 |           |
| 1   |  KKRT AGGREGATE                              |                    |     1 |    30 |       |           |
| 2   |   NESTED LOOPS                                  |                    |   140 |  4200 |   208 |  00:00:03 |
| 3   |    TABLE ACCESS BY INDEX ROWID | FF                 |   140 |  1540 |    52 |  00:00:01 |
| 4   |     INDEX RANGE SCAN                        | IDX_T_NUM          |   140 |       |     3 |  00:00:01 |
| 5   |    TABLE ACCESS BY INDEX ROWID | KK                 |     1 |    19 |     2 |  00:00:01 |
| 6   |     INDEX UNIQUE SCAN                       | PK7                |     1 |       |     1 |  00:00:01 |
------------------------------------------------------------+-----------------------------------+


最后sql选择了正确的执行计划。

总结: sql选择错误执行计划的根本原因是统计信息过于陈旧,该表的统计信息又被锁定

SQL> select stattype_locked from user_tab_statistics where table_name ='KK';
STATTYPE_LOCKED
---------------
ALL


导致定时收集统计信息的job无法定期更新此表的统计信息,重新收集一下表kk的统计信息,问题得到解决。
为防止类似问题的发生,创建一个job, 定期更新kk表上的IDX_KK_CREATE_DATE索引统计信息




赞(0)    操作        顶端 
总帖数
1
每页帖数
101/1页1
返回列表
发新帖子
请输入验证码: 点击刷新验证码
您需要登录后才可以回帖 登录 | 注册
技术讨论