SQL优化一例：巧用hint
Oracle的hint为DBA必须掌握的一项技能，下文中记录了用hint来解决的一个sql优化案例。
问题描述：
从10gR2迁移到11gR2后，有条SQL运行变的很慢，大概要30s左右。
select prd.delevery_order_id,
prd.sku_desc,
prd.item_id,
prd.product_name,
prd.product_num,
prd.sal_price,
dlo.order_status,
to_char(dlo.order_create_time, 'yyyy-MM-dd hh24:mi:ss') as order_create_time,
dlo.order_substatus,
prd.storage_name,
prd.attributes,
dlo.trans_usr_name,
nvl(to_char(dlo.buyer_pay_time, 'yyyy-MM-dd hh24:mi:ss'), '') as buyer_pay_time,
nvl(to_char(dlo.seller_ship_time, 'yyyy-MM-dd hh24:mi:ss'), '') as seller_ship_time,
nvl(to_char(dlo.complete_time, 'yyyy-MM-dd hh24:mi:ss'), '') as complete_time,
dlo.order_type,
co.invoice_required,
co.invoice_content,
co.invoice_title,
co.invoice_address
from checkout_order_prd prd, delevery_order dlo, ctoc_order co
where prd.delevery_order_id = dlo.delevery_order_id
and dlo.esc_orderid = co.esc_orderid(+)
and (not (dlo.order_status = 9 and
dlo.cancel_type not in (904, 906, 910, 911, 912, 913)))
and prd.merchant_onlyid = 2
and dlo.order_status in (4,5,6,7,8)
order by dlo.buyer_pay_time desc nulls last,
prd.delevery_order_id desc,
prd.item_id desc;查找原因：
查看11g中的执行计划：
执行计划
----------------------------------------------------------
Plan hash value: 903664453
------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 41040 | 9137K| | 158K (3)| 00:31:39 |
| 1 | SORT ORDER BY | | 41040 | 9137K| 9960K| 158K (3)| 00:31:39 |
|* 2 | HASH JOIN OUTER | | 41040 | 9137K| 8584K| 156K (3)| 00:31:15 |
|* 3 | HASH JOIN | | 41040 | 8095K| 5416K| 100K (2)| 00:20:03 |
| 4 | TABLE ACCESS BY INDEX ROWID| CHECKOUT_ORDER_PRD | 41040 | 4929K| | 22363 (1)| 00:04:29 |
|* 5 | INDEX RANGE SCAN | IDX_COP_MERCHANT_ONLYID | 41040 | | | 97 (2)| 00:00:02 |
|* 6 | TABLE ACCESS FULL | DELEVERY_ORDER | 2327K| 175M| | 67511 (3)| 00:13:31 |
| 7 | TABLE ACCESS FULL | CTOC_ORDER | 3128K| 77M| | 49872 (4)| 00:09:59 |
------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("DLO"."ESC_ORDERID"="CO"."ESC_ORDERID"(+))
3 - access("PRD"."DELEVERY_ORDER_ID"="DLO"."DELEVERY_ORDER_ID")
5 - access("PRD"."MERCHANT_ONLYID"=2)
6 - filter(("DLO"."ORDER_STATUS"=4 OR "DLO"."ORDER_STATUS"=5 OR "DLO"."ORDER_STATUS"=6 OR
"DLO"."ORDER_STATUS"=7 OR "DLO"."ORDER_STATUS"=8) AND ("DLO"."ORDER_STATUS"<>9 OR
("DLO"."CANCEL_TYPE"=904 OR "DLO"."CANCEL_TYPE"=906 OR "DLO"."CANCEL_TYPE"=910 OR "DLO"."CANCEL_TYPE"=911
OR "DLO"."CANCEL_TYPE"=912 OR "DLO"."CANCEL_TYPE"=913)))解决问题：
尝试改用NESTED LOOPS来进行连接各张表。在sql中加入hint：/*+use_nl (prd,dlo,co)*/
查看新的执行计划：
执行计划
----------------------------------------------------------
Plan hash value: 1786358075
------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 41040 | 9137K| | 217K (1)| 00:43:32 |
| 1 | SORT ORDER BY | | 41040 | 9137K| 9960K| 217K (1)| 00:43:32 |
| 2 | NESTED LOOPS OUTER | | 41040 | 9137K| | 215K (1)| 00:43:08 |
| 3 | NESTED LOOPS | | 41040 | 8095K| | 104K (1)| 00:20:55 |
| 4 | TABLE ACCESS BY INDEX ROWID| CHECKOUT_ORDER_PRD | 41040 | 4929K| | 22363 (1)| 00:04:29 |
|* 5 | INDEX RANGE SCAN | IDX_COP_MERCHANT_ONLYID | 41040 | | | 97 (2)| 00:00:02 |
|* 6 | TABLE ACCESS BY INDEX ROWID| DELEVERY_ORDER | 1 | 79 | | 2 (0)| 00:00:01 |
|* 7 | INDEX UNIQUE SCAN | PK_DELEVERY_ORDER | 1 | | | 1 (0)| 00:00:01 |
| 8 | TABLE ACCESS BY INDEX ROWID | CTOC_ORDER | 1 | 26 | | 3 (0)| 00:00:01 |
|* 9 | INDEX RANGE SCAN | IDX_CO_ESC_ORDERID1 | 1 | | | 2 (0)| 00:00:01 |
------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
5 - access("PRD"."MERCHANT_ONLYID"=2)
6 - filter(("DLO"."ORDER_STATUS"=4 OR "DLO"."ORDER_STATUS"=5 OR "DLO"."ORDER_STATUS"=6 OR
"DLO"."ORDER_STATUS"=7 OR "DLO"."ORDER_STATUS"=8) AND ("DLO"."ORDER_STATUS"<>9 OR
("DLO"."CANCEL_TYPE"=904 OR "DLO"."CANCEL_TYPE"=906 OR "DLO"."CANCEL_TYPE"=910 OR "DLO"."CANCEL_TYPE"=911
OR "DLO"."CANCEL_TYPE"=912 OR "DLO"."CANCEL_TYPE"=913)))
7 - access("PRD"."DELEVERY_ORDER_ID"="DLO"."DELEVERY_ORDER_ID")
9 - access("DLO"."ESC_ORDERID"="CO"."ESC_ORDERID"(+))
已选择628行。
已用时间: 00: 00: 01.59小结：
1、从10g迁移到11g，可能会造成有些SQL运行变慢的情况，怀疑是统计信息不准所致，建议重新收集统计信息。
2、执行计划并不是最准确的，可能跟实际的运行情况相差很远。
3、当CBO不能正确选择Access Paths和Join Type等时，需要用hint来进行强制指定。