with tmp1 as ( select t1.day ,t1.value ,row_number() over(order by t1.day) as rn from ( select '2015-01-01' as day, 1 as value union all select '2015-02-01' as day, 1 as value union all select '2015-05-01' as day, 1 as value union all select '2015-08-01' as day, 2 as value union all select '2015-11-21' as day, 2 as value union all select '2016-01-21' as day, 3 as value union all select '2016-05-21' as day, 3 as value union all select '2016-08-21' as day, 5 as value union all select '2016-11-21' as day, 6 as value ) t1 ) -- 将记录按日期排序生成次序列 ,tmp2 as ( select case when t2.day is null then '0001-01-01' else t1.day end as day ,t1.value ,row_number() over(order by case when t2.day is null then '0001-01-01' else t1.day end) as rn from tmp1 t1 left join tmp1 t2 on t1.rn = (t2.rn + 1) where nvl(t1.value,'') <> nvl(t2.value,'') order by day ) -- 去重后并再次排序的数据,逻辑:按序列+1进行自关联,并把关联后两个值相等的记录剔除,然后再次按时间排序,生成新的次序列 select t1.day as start_day ,nvl(date_add(t2.day,-1),'2999-12-31') as end_day ,t1.value as change_value from tmp2 t1 left join tmp2 t2 on t1.rn = (t2.rn - 1) ;
+-------------+-------------+---------------+--+ | start_day | end_day | change_value | +-------------+-------------+---------------+--+ | 0001-01-01 | 2015-07-31 | 1 | | 2015-08-01 | 2016-01-20 | 2 | | 2016-01-21 | 2016-08-20 | 3 | | 2016-08-21 | 2016-11-20 | 5 | | 2016-11-21 | 2999-12-31 | 6 | +-------------+-------------+---------------+--+