马虎的运维DBA
粗心的运维DBA
要下班的时候，网友发来QQ消息，说SQL跑得慢，要1-3秒出结果，希望优化一下：
SELECT /*+INDEX(TMS,IDX1_TB_EVT_DLV_W)*/
TMS.MAIL_NUM,
TMS.DLV_BUREAU_ORG_CODE AS DLVORGCODE,
RO.ORG_SNAME AS DLVORGNAME,
TMS.DLV_PSEG_CODE AS DLVSECTIONCODE,
TMS.DLV_PSEG_NAME AS DLVSECTIONNAME,
TO_CHAR(TMS.DLV_DATE,'YYYY-MM-DD HH24:MI:SS') AS RECTIME,
TMS.DLV_STAFF_CODE AS HANDOVERUSERCODE,
TU2.REALNAME AS HANDOVERUSERNAME,
DECODE( TMS.DLV_STS_CODE ,'I','妥投','H','未妥投', TMS.DLV_STS_CODE) AS DLV_STS_CODE,
CASE WHEN TMS.MAIL_NUM LIKE 'EC%' THEN '代收'
WHEN TMS.MAIL_NUM LIKE 'ED%CW' THEN '代收'
WHEN TMS.MAIL_NUM LIKE 'FJ%' THEN '代收'
WHEN TMS.MAIL_NUM LIKE 'GC%' THEN '代收'
ELSE
'非代收'
END MAIL_NUM_TYPE
FROM TB_EVT_DLV_W TMS
LEFT JOIN RES_ORG RO ON TMS.DLV_BUREAU_ORG_CODE=RO.ORG_CODE
LEFT JOIN TB_USER TU2 ON TU2.DELVORGCODE=TMS.DLV_BUREAU_ORG_CODE AND TU2.USERNAME=TMS.DLV_STAFF_CODE
WHERE NOT EXISTS (SELECT /*+INDEX(TDW,IDX1_TB_MAIL_SECTION_STORE)*/ MAIL_NUM FROM TB_MAIL_SECTION_STORE TDW WHERE
TDW.MAIL_NUM = TMS.MAIL_NUM
AND TDW.DLVORGCODE = TMS.DLV_BUREAU_ORG_CODE
and TDW.DLVORGCODE='35000133'
AND TDW.RECTIME >= TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS')
AND TO_DATE('2012-11-08 15:15','YYYY-MM-DD HH24:MI:SS') >= TDW.RECTIME and rownum=1 )
AND
TMS.DLV_BUREAU_ORG_CODE = '35000133'
AND TMS.DLV_DATE >= TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS')
AND TO_DATE('2012-11-08 15:15','YYYY-MM-DD HH24:MI:SS') >= TMS.DLV_DATE
AND ('' IS NULL OR TMS.DLV_STAFF_CODE ='')
AND ('' IS NULL OR TU2.REALNAME LIKE '%%')
AND TMS.REC_AVAIL_FLAG = '1'
Plan hash value: 1159587453
----------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
----------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 322K(100)| | | |
|* 1 | FILTER | | | | | | | |
| 2 | NESTED LOOPS OUTER | | 131 | 13493 | 928 (1)| 00:00:12 | | |
|* 3 | HASH JOIN RIGHT OUTER | | 129 | 10191 | 670 (1)| 00:00:09 | | |
|* 4 | TABLE ACCESS BY INDEX ROWID | EMS_USER | 6 | 120 | 8 (0)| 00:00:01 | | |
|* 5 | INDEX RANGE SCAN | EMS_USER_NEW_INX_ORG | 7 | | 3 (0)| 00:00:01 | | |
|* 6 | TABLE ACCESS BY GLOBAL INDEX ROWID | TB_EVT_DLV_W | 129 | 7611 | 661 (0)| 00:00:08 | ROWID | ROWID |
|* 7 | INDEX RANGE SCAN | IDX1_TB_EVT_DLV_W | 586 | | 86 (0)| 00:00:02 | | |
|* 8 | COUNT STOPKEY | | | | | | | |
|* 9 | FILTER | | | | | | | |
| 10 | PARTITION RANGE ITERATOR | | 1 | 31 | 246 (0)| 00:00:03 | KEY | KEY |
|* 11 | TABLE ACCESS BY LOCAL INDEX ROWID| TB_MAIL_SECTION_STORE | 1 | 31 | 246 (0)| 00:00:03 | KEY | KEY |
|* 12 | INDEX RANGE SCAN | IDX1_TB_MAIL_SECTION_STORE | 1 | | 245 (0)| 00:00:03 | KEY | KEY |
| 13 | TABLE ACCESS BY INDEX ROWID | RES_ORG | 1 | 24 | 2 (0)| 00:00:01 | | |
|* 14 | INDEX RANGE SCAN | IDX_RES_ORG | 1 | | 1 (0)| 00:00:01 | | |
----------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter(TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS')<=TO_DATE('2012-11-08 15:15','YYYY-MM-DD HH24:MI:SS'))
3 - access("EU"."USERNAME"="TMS"."DLV_STAFF_CODE" AND "EU"."DELVORGCODE"="TMS"."DLV_BUREAU_ORG_CODE")
4 - filter("EU"."POSTMANKIND"<>5)
5 - access("EU"."DELVORGCODE"='35000133')
6 - filter(("TMS"."DLV_DATE">=TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS') AND "TMS"."REC_AVAIL_FLAG"='1' AND
"TMS"."DLV_DATE"<=TO_DATE('2012-11-08 15:15','YYYY-MM-DD HH24:MI:SS')))
7 - access("TMS"."DLV_BUREAU_ORG_CODE"='35000133')
filter( IS NULL)
8 - filter(ROWNUM=1)
9 - filter((TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS')<=TO_DATE('2012-11-08 15:15','YYYY-MM-DD HH24:MI:SS') AND
:B1='35000133'))
11 - filter(("TDW"."RECTIME">=TO_DATE('2012-11-01 00:00','YYYY-MM-DD HH24:MI:SS') AND "TDW"."RECTIME"<=TO_DATE('2012-11-08
15:15','YYYY-MM-DD HH24:MI:SS')))
12 - access("TDW"."DLVORGCODE"=:B1 AND "TDW"."MAIL_NUM"=:B2)
14 - access("TMS"."DLV_BUREAU_ORG_CODE"="RO"."ORG_CODE")
拿到这个执行计划之后，大概看了一下，花了1分钟就搞定问题(我帮别人优化SQL都不连接他们数据库，直接用眼睛看，或者问几句就OK了)
落落 17:55:35
你那个应该是分区表吧？ ---我指的是这个表 TB_EVT_DLV_W
落落 17:55:50
根据 DLV_DATE 分区的？？
落落 17:55:53
是不是?
网友XXX 17:56:01
SHI
网友XXX 17:56:03
是的
落落 17:56:17
create index idx on TB_EVT_DLV_W
(DLV_BUREAU_ORG_CODE) local;
落落 17:56:29
你那个索引不是local的
--------------------------
6点下班回家过后，晚上网友发来QQ消息
--------------------------
网友XXX 20:41:33
比原来快了不少
网友XXX 20:41:38
给力落落兄
这个SQL性能问题，是由于他们运维DBA 在创建索引的时候，搞忘了 local关键字，
从而导致 SQL 在有 where条件(分区键包含在where条件中) 能进行 分区裁剪的时候，不能有效的踢出大量数据。
最终导致ID=7 这里这个分区扫描大量的leaf block 并且到ID=6 这里还去回表，又进行大量的 过滤操作。
各位读者，你们能1分钟内解决这个SQL性能问题吗？ 如果不能，来参加我的SQL优化培训吧
联系方式： QQ 692162374
- 1楼sky13and234天前 09:50
- 分区还是本地索引好维护，全局索引一旦分区有变化索引会失效
- Re: robinson19883天前 10:05
- 回复sky13and23nYES