下面有一个用于搜索功能的oracle SQL查询,它包含在存储过程中。
它运行得很慢。
OPEN p_cursor FOR 选择 TableA。*,TB_S.someColumn,TB_S.SomeColumn ..
FROM customer_site TableA LEFT JOIN TableB TB on TB.some_id = TableA.some_id LEFT JOIN TableC TB_S on TB_S.Bla_ID = TB.Bla_ID LEFT JOIN TableC TB_CS on TB_CS.Bla2_ID = TB.Bla2_ID LEFT JOIN TableC TB_1 on TB_1.Bla3_ID = TB.Bla3_ID LEFT JOIN TableC TB_2 on TB_DC.Bla4_ID = TB.Bla4_ID LEFT JOIN TableD SP on SP.SP_ID = TableA.SP_ID LEFT JOIN TableC TB_3 on TB_3.Bla5_ID = TB.Bla5_ID LEFT JOIN TableC TB_4 on TB_4.Bla6_ID = TB.Bla6_ID LEFT JOIN TableC TB_5 on TB_5.Bla7_ID = TB.Bla7_ID WHERE (p_nmi IS NULL OR TableA.someid LIKE p_nmi) AND (p_last_name IS NULL OR TableA.last_name LIKE p_last_name) AND (p_full_address IS NULL OR UPPER(assemble_address(flat_number, street_number, street_name, street_suffix, apartment_number, building_name, suburb, state, postcode)) LIKE p_full_address) AND (p_param1 IS NULL OR TB.owner = p_param1) AND (p_param2 IS NULL OR TB.status = p_param2) AND (p_param3 IS NULL OR TB.contact_stage = p_param3) AND (p_param4 IS NULL OR TB.no_access_code = p_param4) AND (p_param5 IS NULL OR TB.defect_code = p_param5) AND (p_param6 IS NULL OR TB.REFUSAL_RESOLUTION = p_param6) AND (p_param7 IS NULL OR TB.DEFECT_LEVEL = p_param7) AND (p_param8 IS NULL OR TB.AMI_CTR_STATUS = p_param8) AND (p_param9 IS NULL OR TableA.meter_route LIKE p_param9) AND (p_param10 IS NULL OR TableA.sp_id = p_param10) AND (p_inTBdent_date_from IS NULL OR TB.inTBdent_date >= p_inTBdent_date_from) AND (p_inTBdent_date_to IS NULL OR TB.inTBdent_date <= p_inTBdent_date_to) AND rownum < 1001 ORDER BY TB.inTBdent_date;
有些数据库专家可以帮助我如何调整上述查询吗?
谢谢。
答案 0 :(得分:1)
正如@Annjawn建议的那样,将(p_param IS NULL or col = p_param)
的所有实例更改为col = nvl(p_param, col)
。 Oracle可以优化这些条件,如this Jonathan Lewis article所述。这可能会启用INDEX RANGE SCAN
而不是TABLE ACCESS FULL
。但是,当列为NOT NULL
时,该更改在逻辑上是等效的,因为null = null
不会返回true。 (我个人更喜欢你编写条件的方式,但Oracle似乎讨厌OR。)
如果这没有帮助,请发布解释计划以确定真正的问题。首先,修改过程并添加提示SELECT /*+ gather_plan_statistics */ TableA.*, ...
。运行该过程,然后使用select * from v$sql where lower(sql_fulltext) like '%gather_plan_statistics%';
之类的查询查找相关的SQL_ID。最后,发布结果
select * from table(dbms_xplan.display_cursor(sql_id => '<sql_id from previous step>', format => 'allstats last'));
。这将告诉我们执行计划以及优化程序可能做出的糟糕决策。