union all语句优化改写经验分享

1、问题

前面几个案例是因为提前生成HEAP TABLE，条件没有提前过滤导致性能问题，今天想跟大家分享一个案例是适合做HEAP TABLE优化的。原语句如下：

bash 复制代码

SELECT * 
  FROM(SELECT INNER_TABLE.*, 
               ROWNUM OUTER_TABLE_ROWNUM 
          FROM (SELECT T1.CODE,
              case when T1.CODE='-9999' then '合计'
              else t2.cname end cname,
              nvl(firstnum,0)firstnum,
              nvl(secondnum,0)secondnum,
              nvl(thirdnum,0)thirdnum,
              nvl(fourthnum,0)fourthnum,
              nvl(fifthnum,0)fifthnum
              from ( select '-9999' code,
                 sum(case when status=1 then 1 else 0 end) firstnum,
                 sum(case when status=2 then 1 else 0 end) secondnum,
                 sum(case when status=3 then 1 else 0 end) thirdnum,
                 sum(case when status=4 then 1 else 0 end) fourthnum,
                 sum(case when status=5 then 1 else 0 end) fifthnum
                 from t1
                 where c1 like 'A1%'
                 union all
           select code,
                 sum(case when status=1 then 1 else 0 end) firstnum,
                 sum(case when status=2 then 1 else 0 end) secondnum,
                 sum(case when status=3 then 1 else 0 end) thirdnum,
                 sum(case when status=4 then 1 else 0 end) fourthnum,
                 sum(case when status=5 then 1 else 0 end) fifthnum
                 from t1
                 where c1 like 'A1%'
                 group by code) t1 left join t2 on t1.code=t2.code
               ORDER BY T1.CODE)INNER_TABLE ) OUTER_TABLE 
 WHERE OUTER_TABLE_ROWNUM<=200AND OUTER_TABLE_ROWNUM >0

计划：

从计划上对t1表过滤重复两次，从语句中可以看到

bash 复制代码

select '-9999' code,
                 sum(case when status=1 then 1 else 0 end) firstnum,
                 sum(case when status=2 then 1 else 0 end) secondnum,
                 sum(case when status=3 then 1 else 0 end) thirdnum,
                 sum(case when status=4 then 1 else 0 end) fourthnum,
                 sum(case when status=5 then 1 else 0 end) fifthnum
                 from t1
                 where c1 like 'A1%'
                 union all
           select code,
                 sum(case when status=1 then 1 else 0 end) firstnum,
                 sum(case when status=2 then 1 else 0 end) secondnum,
                 sum(case when status=3 then 1 else 0 end) thirdnum,
                 sum(case when status=4 then 1 else 0 end) fourthnum,
                 sum(case when status=5 then 1 else 0 end) fifthnum
                 from t1
                 where c1 like 'A1%'
                 group by code

是union all两部分，一部分是合计，另一部分是分组统计，此时我想到可以先分组统计，再把分组统计结果进行合计，分组统计的结果可以放到临时表中即heap table,从数学角度上讲就是提取公因子。

2、改写

bash 复制代码

SELECT * 
  FROM(SELECT INNER_TABLE.*, 
               ROWNUM OUTER_TABLE_ROWNUM 
          FROM (SELECT T1.CODE,
              case when T1.CODE='-9999' then '合计'
              else t2.cname end cname,
              nvl(firstnum,0)firstnum,
              nvl(secondnum,0)secondnum,
              nvl(thirdnum,0)thirdnum,
              nvl(fourthnum,0)fourthnum,
              nvl(fifthnum,0)fifthnum
              from (with temp as ( 
           select code,
                 sum(case when status=1 then 1 else 0 end) firstnum,
                 sum(case when status=2 then 1 else 0 end) secondnum,
                 sum(case when status=3 then 1 else 0 end) thirdnum,
                 sum(case when status=4 then 1 else 0 end) fourthnum,
                 sum(case when status=5 then 1 else 0 end) fifthnum
                 from t1
                 where c1 like 'A1%'
                 group by code)
                 select '-9999' code, sum(firstnum) firstnum,
                 sum(secondnum) secondnum, sum(thirdnum)thirdnum,
                 sum(fourthnum) fourthnum,sum(fifthnum) fifthnum
                 from temp t1
                 union all
                 select * from temp
                 ) t1 
                  left join t2 on t1.code=t2.code
               ORDER BY T1.CODE)INNER_TABLE ) OUTER_TABLE 
 WHERE OUTER_TABLE_ROWNUM<=200AND OUTER_TABLE_ROWNUM >0

计划：

结果符合预期，性能提升一倍以上。

3、小结

使用union all和group by rollup（前面案例中有提过-单表查询-group by rollup优化）进行数据统计的sql优化思路就是提取公因子，利用临时表（HEAP TABLE）减少扫描次数，从而提升性能。

前面讲过优化不是单向，前面是讲HEAP TABLE如何拆解优化，而这次是如何用HEAP TABLE减少扫描次数优化，优化做的事不过是对症下药。

4、附加测试数据

bash 复制代码

create table t1(id int primary key,c1 varchar2(20),status int,code varchar2(20));
insert into t1 select level,'A'||to_char(round(dbms_random.value(1,10000),0)),round(dbms_random.value(1,5),0),
'0'||to_char(round(dbms_random.value(300,322),0)) from dual connect by level<=300000;
commit;
dbms_stats.gather_table_stats(USER,'T1',null,100);
create table t2(id int primary key,cname varchar2(20),code varchar2(20));
insert into t2 select level,'A'||to_char(round(dbms_random.value(300,322),0)),
'0'||to_char(round(dbms_random.value(300,322),0)) from dual connect by level<=10000;
commit;
dbms_stats.gather_table_stats(USER,'T2',null,100);
create index IDX_DM_T1_C1 on T1(c1);
create index IDX_DM_t2_CODE on T2(CODE);