数据比对总结

1. 数据量比对;
2. 数据内容比对;
3. 不同维度上比对, 查看维度因素;

一、 数据量比对

-- 总数据量比对
select
    aa.cnt as pre_cnt
    ,bb.cnt as pro_cnt
    ,( aa.cnt - bb.cnt ) as err
    ,( aa.cnt - bb.cn t)*1.000/bb.cnt as err_ratio
from ( -- 测试数据
        select 
            '1' as id
            ,count(1) as cnt
        from
           ${tbl_name_1}
        where
            dt = '20230730'
    ) aa
join ( -- 生产数据
        select 
            '1' as id
            ,count(1) as cnt
        from
           ${tbl_name_2}
        where
            dt = '20230730'
    ) bb
    on aa.id = bb.id
;
 
 
-- 单维度比对数据量
select
    aa.cnt pro_cnt
    ,bb.cnt pre_cnt
    ,(aa.cnt - bb.cnt) as err_cnt
    ,(aa.cnt - bb.cnt)*1.00/aa.cnt as err_ratio
from (
        select
            app_id
            ,count(1) as cnt
        from
            tbl_pro
        where
            dt = '20231010'
        group by
            app_id
    ) aa
full join (
        select
            app_id
            ,count(1) as cnt
        from
            tbl_pro
        where
            dt = '20231010'
        group by
            app_id
    ) bb
    on aa.app_id = bb.app_id
;

二、数据内容比对

-- 内容比对
with tmp as (
    select
        aa
        ,bb
    from
        ${tbl_name_1}
    where
        dt = '20231017'
        
    except
 
    select
        aa
        ,bb
    from
        ${tbl_name_2}
    where
        dt = '20231017'
)
select
    *
from
    tmp
;

三、指标比对

-- 比对指标总和
select
    aa.cnt pro_cnt
    ,bb.cnt pre_cnt
    ,(aa.cnt - bb.cnt) as err_cnt
    ,(aa.cnt - bb.cnt)*1.00/aa.cnt as err_ratio
from (
        select
            'id' as id
            ,sum(active_num) as cnt
        from
            pro_tbl
        where
            dt = '20231010'
    ) aa
join (
        select
            'id' as id
            ,sum(active_num) as cnt
        from
            pre_tbl
        where
            dt = '20231010'
    ) bb
    on aa.id = bb.id
;
 
-- 某些维度上指标和比对
select
    aa.app_id
    ,aa.cnt as cnt_pro
    ,bb.cnt as cnt_pre
    ,(aa.cnt - bb.cnt) as err_cnt
    ,(aa.cnt - bb.cnt)*1.000/aa.cnt as error
from (
        select
            app_id
            ,sum(active_num) as cnt
        from
            pro_tbl
        where
            dt = '20231010'
        group by
            app_id
    ) aa
full join (
        select
            app_id
            ,sum(active_num) as cnt
        from
            pre_tbl
        where
            dt = '20231010'
        group by
            app_id
    ) bb
    on aa.app_id = bb.app_id
order by aa.cnt desc
;

二十、差异分析

1. 随机函数

first、last函数会随机取维度值, 造成这些维度指标不完全一样, 可验证其它维度上指标是否一致.

2. join异常

表的join操作-CSDN博客

3. 增加过滤后数据变多

过滤掉了无效数据, 保留下了更过的有效数据;

比如pkgs数组字段有大量空值，row_number方式排序去重(rnk=1)前没有过滤pkgs为空的数据，这样获取了许多pkgs为空的数据，导致展开数组时会丢失有效数据.