表格数据的存储对索引的影响
近期碰到一张包含日期字段的表格，用户反应即使在日期字段上建立了索引，查询报表时涉及的SQL仍然选择走全表扫描，而忽略索引。
对应表名为：T_GATE_CTN_FT
查询SQL如下：
select sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
注：2012由前台参数传入，语句不建议修改
执行计划如下：
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 4280972504
--------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
--------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 436K| 37M| 29487 (2)| 00:0
| 1 | SORT GROUP BY | | 436K| 37M| 29487 (2)| 00:0
|* 2 | HASH JOIN | | 436K| 37M| 29460 (2)| 00:0
|* 3 | TABLE ACCESS FULL | T_CTN_DM | 5549 | 94333 | 103 (1)| 00:0
|* 4 | HASH JOIN | | 436K| 30M| 29354 (2)| 00:0
|* 5 | TABLE ACCESS FULL| T_DATE_DM | 359 | 16514 | 42 (0)| 00:0
|* 6 | TABLE ACCESS FULL| T_GATE_CTN_FT | 4783K| 127M| 29277 (1)| 00:0
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T2867"."CTN_TYPE_UID"="T5253"."CTN_TYPE_UID")
3 - filter("T2867"."CTN_STATUS"='F')
4 - access("T2881"."DATE_UID"="T5253"."DATE_UID")
5 - filter("T2881"."OWC_YEAR_NO"=2012 OR "T2881"."OWC_YEAR_NO"+1=2012)
6 - filter("T5253"."MOVE_KIND"='RECV')
执行时间：
SQL>
ALTER SYSTEM FLUSH BUFFER_CACHE;
System altered
Executed in 0.156 seconds
ALTER SYSTEM FLUSH SHARED_POOL;
System altered
Executed in 0.468 seconds
select sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
Result:
……
……
105 rows selected
Executed in 18.299 seconds
检查表格和对应索引的统计信息：
select * from user_tables where table_name='T_GATE_CTN_FT';
select * from user_indexes where table_name='T_GATE_CTN_FT' and index_name='I_GATE_CTN_DATE_UID';
注意上图聚集因子达到了536655.
select * from user_ind_columns where index_name='I_GATE_CTN_DATE_UID';
如果强制使用索引：
语句改为：
select /*+index(T5253 I_GATE_CTN_DATE_UID)*/sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
执行计划：
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 1969617402
--------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | C
--------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 436K| 37M| 5
| 1 | SORT GROUP BY | | 436K| 37M| 5
|* 2 | HASH JOIN | | 436K| 37M| 5
|* 3 | TABLE ACCESS FULL | T_CTN_DM | 5549 | 94333 |
| 4 | NESTED LOOPS | | | |
| 5 | NESTED LOOPS | | 436K| 30M| 5
|* 6 | TABLE ACCESS FULL | T_DATE_DM | 359 | 16514 |
|* 7 | INDEX RANGE SCAN | I_GATE_CTN_DATE_UID | 1894 | |
|* 8 | TABLE ACCESS BY INDEX ROWID| T_GATE_CTN_FT | 1214 | 33992 |
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T2867"."CTN_TYPE_UID"="T5253"."CTN_TYPE_UID")
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
3 - filter("T2867"."CTN_STATUS"='F')
6 - filter("T2881"."OWC_YEAR_NO"=2012 OR "T2881"."OWC_YEAR_NO"+1=2012)
7 - access("T2881"."DATE_UID"="T5253"."DATE_UID")
8 - filter("T5253"."MOVE_KIND"='RECV')
执行时间：
SQL>
ALTER SYSTEM FLUSH BUFFER_CACHE;
System altered
Executed in 0.125 seconds
ALTER SYSTEM FLUSH SHARED_POOL;
System altered
Executed in 0.686 seconds
select /*+index(T5253 I_GATE_CTN_DATE_UID)*/sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
Result:
……
……
105 rows selected
Executed in 34.913 seconds
可见由于对于date_uid字段来说，数据存储到表格里的时候是无序的，导致索引的聚集因子过大，如果在查询中使用date_uid的索引，反而降低了查询的效率。
改进测试：
对于date_uid字段，因为是自增长类型，之后新填入的数据几乎都是顺序增长的。
那么建议在仓库比较空闲时重建此表格，按date_uid顺序重新插入数据，测试一下索引的可用情况。
建立测试表格：
CREATE TABLE T_GATE_CTN_FT2 AS SELECT * FROM T_GATE_CTN_FT ORDER BY DATE_UID;
CREATE INDEX I_GATE_CTN_FT2_DATE_UID ON T_GATE_CTN_FT2(DATE_UID);
检查统计信息：
SELECT * FROM USER_TABLES WHERE TABLE_NAME='T_GATE_CTN_FT2';
SELECT * FROM USER_INDEXES WHERE TABLE_NAME='T_GATE_CTN_FT2';
注意聚集因子为106590.
替换表格，检查执行计划：
select sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT2 T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 746580745
--------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes
--------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 325K| 28
| 1 | SORT GROUP BY | | 325K| 28
|* 2 | HASH JOIN | | 325K| 28
|* 3 | TABLE ACCESS FULL | T_CTN_DM | 5549 | 94333
| 4 | NESTED LOOPS | | |
| 5 | NESTED LOOPS | | 325K| 22
|* 6 | TABLE ACCESS FULL | T_DATE_DM | 359 | 16514
|* 7 | INDEX RANGE SCAN | I_GATE_CTN_FT2_DATE_UID | 1827 |
|* 8 | TABLE ACCESS BY INDEX ROWID| T_GATE_CTN_FT2 | 907 | 25396
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T2867"."CTN_TYPE_UID"="T5253"."CTN_TYPE_UID")
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
3 - filter("T2867"."CTN_STATUS"='F')
6 - filter("T2881"."OWC_YEAR_NO"=2012 OR "T2881"."OWC_YEAR_NO"+1=2012)
7 - access("T2881"."DATE_UID"="T5253"."DATE_UID")
8 - filter("T5253"."MOVE_KIND"='RECV')
执行情况：
SQL>
ALTER SYSTEM FLUSH BUFFER_CACHE;
System altered
Executed in 0.156 seconds
ALTER SYSTEM FLUSH SHARED_POOL;
System altered
Executed in 0.468 seconds
select sum(T5253.TEU) as c1,
T2881.OWC_WEEK_DESC as c2,
T2881.OWC_WEEK_NUMBER as c3,
T2881.OWC_YEAR_NO as c4
from T_CTN_DM T2867, T_DATE_DM T2881, T_GATE_CTN_FT2 T5253
where (T2867.CTN_STATUS = 'F' and T2867.CTN_TYPE_UID = T5253.CTN_TYPE_UID and
T2881.DATE_UID = T5253.DATE_UID and T5253.MOVE_KIND = 'RECV' and
(T2881.OWC_YEAR_NO + 1 in (2012) or T2881.OWC_YEAR_NO in (2012)))
group by T2881.OWC_WEEK_NUMBER, T2881.OWC_YEAR_NO, T2881.OWC_WEEK_DESC
order by c2, c3
;
Result:
……
……
105 rows selected
Executed in 8.391 seconds
可见当表格数据按照date_uid顺序存入时，索引的聚集因子比较低，相应的索引的可用性比较好，oracle的CBO会自动选择索引扫描。
来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/10009036/viewspace-1070401/，如需转载，请注明出处，否则将追究法律责任。
转载于:http://blog.itpub.net/10009036/viewspace-1070401/
392
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
