sql - 填补时间线空白的窗口函数

Question

数据

session                             time_interval       activity
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:40 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:41 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:42 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:43 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:44 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:45 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:46 running
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:47 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:48 (null)
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:49 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:50 walking
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:51 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:52 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:53 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:54 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:55 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:56 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:57 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:58 resting
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:59 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:00 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:01 (null)
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:02 walking

SQL

SELECT session, 
       time_interval, 
       activity, 
       FIRST_VALUE(activity) 
         OVER ( 
           PARTITION BY session 
           ORDER BY time_interval 
               RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) AS activity_b
FROM   my_table; 

但这只取 session 的第一个值。如何获取每秒的前值？

想要的结果

session                             time_interval       activity
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:40 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:41 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:42 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:43 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:44 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:45 walking
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:46 running
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:47 running
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:48 running
c889ddb532e76c961c2944dd90b10142    2017-05-25 20:16:49 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:50 walking
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:51 walking
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:52 walking
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:53 walking
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:54 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:55 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:56 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:57 running
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:58 resting
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:16:59 resting
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:00 resting
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:01 resting
dddjg894hlog8sdlf2090288fmma201c    2017-05-25 20:17:02 walking

SQL Fiddle 已满负荷，所以这里有一些 DDL

CREATE TABLE public.my_table (
  session varchar(32),
  time_interval timestamp,
  activity varchar(10));

INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:40','walking');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:41','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:42','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:43','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:44','walking');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:45','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:46','running');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:47','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:48','');
INSERT INTO public.my_table VALUES ('c889ddb532e76c961c2944dd90b10142','2017-05-25 20:16:49','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:50','walking');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:51','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:52','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:53','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:54','running');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:55','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:56','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:57','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:58','resting');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:16:59','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:17:00','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:17:01','');
INSERT INTO public.my_table VALUES ('dddjg894hlog8sdlf2090288fmma201c','2017-05-25 20:17:02','resting');

Answer 1

这正是您需要ignore null选项的地方。但那是不可用的。因此，一种方法使用最大扫描和 join:

select t.session, t.time_interval, tt.activity
from (select t.*,
              max(case when t.activity is not null then t.time_interval end) over (partition by t.session order by t.time_interval) as value_ti
      from t
     ) t left join
     t tt
     on t.value_ti = tt.time_interval and t.session = tt.session;

当值不为 NULL 时，这会计算每行的最近时间间隔。然后它会重新加入以获取当时的事件。

如果你知道连续的 NULL 永远不会超过 3 个，你也可以使用 lag():

select t.session, t.time_interval,
       coalesce(t.activity,
                lag(t.activity, 1) over (partition by t.session order by t.time_interval),
                lag(t.activity, 2) over (partition by t.session order by t.time_interval),
                lag(t.activity, 3) over (partition by t.session order by t.time_interval)
               ) as acctivity
from t;