Teradata SQL：对同一查询中跨年份的日期范围的结果进行分组

Question

所以我有这个查询，它可以让我计算在给定时间范围内每个月至少进行一次某种交易的帐户。在这种情况下，我正在检查 3 个月内的交易。请参阅下面的查询：

select  
  Yr
, count(distinct acct_nbr) as AcctCt

from  (
        select          
          Yr
        , acct_nbr
        , count(distinct Mth) as Mths        

        from  (
                select                  
                  extract(year from tran_dt) as Yr
                , acct_nbr        
                , extract(month from tran_dt) as Mth
                , tran_id

                from br.bdft

                where prod_type = 44
                and (tran_dt between '2017-10-01' and '2017-12-31'
                or tran_dt between '2016-10-01' and '2016-12-31'
                or tran_dt between '2015-10-01' and '2015-12-31'
                or tran_dt between '2014-10-01' and '2014-12-31'
                or tran_dt between '2013-10-01' and '2013-12-31')
                and tran_cd = 'DD'
              ) dt

        group by 1,2                
      ) dt2

where Mths = 3

group by 1

我得到的结果如下：

Yr:    AcctCt:
2017   258527
2016   231304
2015   188550
2014   210339
2013   170781

我目前通过从交易日期中提取 YEAR 来按 YEAR 对结果进行分组。当日期范围不跨越年份时（例如我在上面的查询中使用的），这可以正常工作。但是一旦我跨年，例如2017-11-01 到 2018-01-31，这种方法不再有效。

有没有更好的方法来对结果进行分组，无论我使用的日期范围如何，它始终有效？

Answer 1

创建第二个表来保存时间段（大量伪代码：创建表 time_periods 作为 ID、period_name、start_date、end_date）。加入此表

from br.bdft 
inner join time_periods on bdft.tran_dt between start_date and end_date

按 period_name 列分组。

抱歉，时间紧迫，或者我会提供更多信息……但是这里有这个概念（在表格中定义时间段并使用 between 连接到该表格）。试一试，如果您仍然需要，我将在今天/明天重新访问完整的解决方案

Answer 2

当他们跨年时，只需减去一定的时间。对于您的示例，请将此表达式用于 Yr：

extract(year from add_months(tran_dt, -1) ) as Yr

Answer 3

在提取年份之前，只需使用 ADD_MONTHS 调整日期。此外，不需要两个 COUNT(DISTINCT)。

这是您的简化查询：

select          
  Yr
, count(*) as AcctCt      

from  (
        select                  
          Extract(YEAR From Add_Months(tran_dt,-1)) as Yr
        , acct_nbr        

        from br.bdft

        where prod_type = 44
        and (tran_dt between '2017-11-01' and '2018-01-31'
          or tran_dt between '2016-11-01' and '2017-01-31'
          or tran_dt between '2015-11-01' and '2016-01-31'
          or tran_dt between '2014-11-01' and '2015-01-31'
          or tran_dt between '2013-11-01' and '2014-01-31')
        and tran_cd = 'DD'
        group by acct_nbr
        , yr
        having count(distinct extract(month from tran_dt)) = 3
      ) dt
 group by 1

编辑：

要获得基于今天的三个月范围，您可以在 CURRENT_DATE 应用 TRUNC/LAST_DAY/ADD_MONTHS 计算。

select          
  Yr
, count(*) as AcctCt      

from  (
        select 
          -- shift back the date between 0 and 2 months based on the month of the current quarter
          -- to get the same year for consecutive months
          Extract(YEAR From Add_Months(tran_dt, -Extract(MONTH From currdt) MOD 3)) as Yr
        , acct_nbr        

        from br.bdft

        where prod_type = 44
        -- get the current months plus the two previous for the current and the previous four years
        -- will be calculated once = Explain shows hard-coded dates
        AND (   tran_dt BETWEEN Trunc(Add_Months(Current_Date,-2      ), 'mon') AND Last_Day(           Current_Date       ) 
             OR tran_dt BETWEEN Trunc(Add_Months(Current_Date,-2 -1*12), 'mon') AND Last_Day(Add_Months(Current_Date,-1*12))
             OR tran_dt BETWEEN Trunc(Add_Months(Current_Date,-2 -2*12), 'mon') AND Last_Day(Add_Months(Current_Date,-2*12))
             OR tran_dt BETWEEN Trunc(Add_Months(Current_Date,-2 -3*12), 'mon') AND Last_Day(Add_Months(Current_Date,-3*12))
             OR tran_dt BETWEEN Trunc(Add_Months(Current_Date,-2 -4*12), 'mon') AND Last_Day(Add_Months(Current_Date,-4*12))
            )
        and tran_cd = 'DD'
        group by acct_nbr
        , yr
        having count(distinct extract(month from tran_dt)) = 3
      ) dt
 group by 1 as Yr