Snowflake 的时差替代 Postgres INTERVAL

Question

我有以毫秒为单位的 UNIX 持续时间，需要将其转换为人类可读的格式。在 Postgres 中可以使用 INTERVAL 类型，但如何在 Snowflake 中完成？挑战在于其中一些值可能超过 24 小时。我已经在 Snowflake forum 上检查了类似问题的解决方案（对 ms 进行了调整）。但是不支持时间>24h

示例数据：

7200000 -- 2h
28800000 --8h
1800000 -- 30min
252000000 -- 70h
86400000 -- 24h
3593314 -- 59min 53sec

以便所需的输出看起来像：

02:00:00
08:00:00
00:30:00
70:00:00
24:00:00
00:59:53

可能是我搜索不好，但是至今没能找到解决这个问题的方法。

适用于 Postgres 的解决方案SELECT time_unix::bigint * INTERVAL '1 ms'

Answer 1

select *
    ,floor(ms/3600000) as hh
    ,floor((ms%3600000)/60000) as mm
    ,floor((ms%60000)/1000) as ss
    ,nullifzero(hh) as a1
    ,nullifzero(mm) as a2
    ,nullifzero(ss) as a3
    ,trim(nvl(a1||'h ','') || nvl(a2||'m ','') || nvl(a3||'s','')) as r 
from values
    (7200000, '2h'),
    (28800000 , '8h'),
    (1800000, '30min'),
    (252000000, '70h'),
    (86400000, '24h'),
    (3593314, '59min 53sec')
    t(ms, wanted)

给出：

MS	WANTED	HH	MM	SS	A1	A2	A3	R
7,200,000	2h	2	0	0	2			2h
28,800,000	8h	8	0	0	8			8h
1,800,000	30min	0	30	0		30		30m
252,000,000	70h	70	0	0	70			70h
86,400,000	24h	24	0	0	24			24h
3,593,314	59min 53sec	0	59	53		59	53	59m 53s

因此：

select *
    ,trim(nvl(nullifzero(floor(ms/3600000))||'h ','') || nvl(nullifzero(floor((ms%3600000)/60000))||'m ','') || nvl(nullifzero(floor((ms%60000)/1000))||'s','')) as r 
from values
    (7200000, '2h'),
    (28800000 , '8h'),
    (1800000, '30min'),
    (252000000, '70h'),
    (86400000, '24h'),
    (3593314, '59min 53sec')
    t(ms, wanted)

给出：

MS	WANTED	R
7,200,000	2h	2h
28,800,000	8h	8h
1,800,000	30min	30m
252,000,000	70h	70h
86,400,000	24h	24h
3,593,314	59min 53sec	59m 53s

或者我可以阅读：

select *
    ,trim(nvl(nullifzero(floor(ms/3600000))||'h ','') || nvl(nullifzero(floor((ms%3600000)/60000))||'m ','') || nvl(nullifzero(floor((ms%60000)/1000))||'s','')) as r 
    ,lpad(floor(ms/3600000), 2, 0) || ':' || lpad(floor((ms%3600000)/60000), 2, 0) || ':' || lpad(floor((ms%60000)/1000), 2, 0) as r2
from values
    (7200000, '2h'),
    (28800000 , '8h'),
    (1800000, '30min'),
    (252000000, '70h'),
    (86400000, '24h'),
    (3593314, '59min 53sec')
    t(ms, wanted)

MS	WANTED	R	R2
7,200,000	2h	2h	02:00:00
28,800,000	8h	8h	08:00:00
1,800,000	30min	30m	00:30:00
252,000,000	70h	70h	70:00:00
86,400,000	24h	24h	24:00:00
3,593,314	59min 53sec	59m 53s	00:59:53

Answer 2

想知道是否有人可以想出更糟糕的解决方案:)

create or replace function pginterval( TIME number )
returns varchar
language sql
as
$$
select  LPAD( (extract( day from to_timestamp( TIME, 3 )) - 1 ) * 24 + extract( hour from  to_timestamp( TIME, 3 )), 2, '0' )
|| ':' || right( to_timestamp( TIME, 3 )::time, 5 )
$$;

select pginterval( x )
from values 
(7200000),
(28800000),
(1800000),
(252000000),
(86400000),
(3593314) tmp (x);

PGINTERVAL( X )
02:00:00
08:00:00
00:30:00
70:00:00
24:00:00
00:59:53

Snowflake 的时差替代 Postgres INTERVAL

Snowflake's time difference alternative to Postgres INTERVAL

snowflake-cloud-data-platform

或者我可以阅读：