如何使用日期时间列按分钟获取记录数

Question

我有一个表格，下面有列：

Customer	Time_Start
A	01/20/2020 01:25:00
A	01/22/2020 14:15:00
A	01/20/2020 03:23:00
A	01/21/2020 20:37:00

我正在尝试获取一个表格，该表格在给定日期按分钟（包括零）输出表格。

即

Customer	Time_Start	Count
A	01/20/2020 00:01:00	5
A	01/20/2020 00:02:00	2
A	01/20/2020 00:03:00	0
A	01/20/2020 00:04:00	12

我希望它一次只为 1 位客户显示 1 天。

这是我目前所拥有的：

select 
customer,
cast(time_start as time) + ((cast(time_start as time) - cast('00:00:00' as time)) hour(2)) as TimeStampHour,
count(*) as count
from    Table_1
where   customer in ('A')
group by    customer, TimeStampHour
order by    TimeStampHour

Answer 1

在 Teradata 16.20 中，这将是一项使用新的时间序列聚合

的简单任务

SELECT 
   customer
  ,Cast(Begin($Td_TimeCode_Range) AS VARCHAR(16))
  ,Count(*)
FROM table_1
WHERE customer = 'A'
  AND time_start BETWEEN TIMESTAMP'2020-01-20 00:00:00' 
                     AND Prior(TIMESTAMP'2020-01-20 00:00:00' + INTERVAL '1' DAY)
GROUP BY TIME(Minutes(1) AND customer) 
         USING timecode(time_start)
         FILL (0)
;

在您必须像 Ramin Faracov answer 一样实现它之前，先创建一个所有分钟的列表，然后再加入它。但我更喜欢在加入之前进行计数：

WITH all_minutes AS
 ( -- create a list of all minutes
   SELECT 
      Begin(pd) AS bucket
   FROM 
    ( -- EXPAND ON requires FROM and TRUNC materializes the FROM avoiding error
      -- "9303. EXPAND ON clause must not be specified in a query expression with no table references."
      SELECT Cast(Trunc(DATE '2020-01-20') AS TIMESTAMP(0)) AS start_date
    ) AS dt
   EXPAND ON PERIOD(start_date, start_date + INTERVAL '1' DAY) AS pd
          BY INTERVAL '1' MINUTE  
 )
SELECT customer
  ,bucket
  ,Coalesce(Cnt, 0)
FROM all_minutes
LEFT JOIN
 (
   SELECT customer
     ,time_start
      - (Extract (SECOND From time_start) * INTERVAL '1' SECOND) AS time_minute
      ,Count(*) AS Cnt
   FROM table_1
   WHERE customer = 'A'
     AND time_start BETWEEN TIMESTAMP'2020-01-20 00:00:00' 
                        AND Prior(TIMESTAMP'2020-01-20 00:00:00' + INTERVAL '1' DAY)
   GROUP BY customer, time_minute
 ) AS counts
ON counts.time_minute = bucket 
ORDER BY bucket
;

Answer 2

首先，我们创建一个函数，该函数返回一个表，该表有一个字段和记录，从 time_start 值开始并增加一分钟。

CREATE OR REPLACE FUNCTION get_dates(time_start timestamp without time zone)
 RETURNS TABLE(list_dates timestamp without time zone)
 LANGUAGE plpgsql
AS $function$
declare
    time_end timestamp;
begin
    time_end = time_start + interval '1 day';
    
    return query 
    SELECT t1.dates
    FROM   generate_series(time_start, time_end, interval '1 min') t1(dates);
        
END;
$function$
;

考虑一下，这个函数返回的时间戳列表有一个小时和分钟，但秒和毫秒总是等于零。我们必须使用timestamp 字段将这个函数结果与我们的customer 表连接起来。但是我们customer 表的时间戳字段是完整的DateTime 格式，秒和毫秒不为空。因此，可以理解我们的连接条件将不正确。我们需要这样，在customer 表中，时间戳字段的所有数据总是有一个空的秒和空的毫秒来正确连接。为此，我创建了一个不可变函数，当您向该函数发送完全格式化的 DateTime 2021-02-03 18:24:51.203 时，该函数将返回空秒和空毫秒的时间戳，格式为：2021-02-03 18:24:00.000

CREATE OR REPLACE FUNCTION clear_seconds_from_datetime(dt timestamp without time zone)
 RETURNS timestamp without time zone
 LANGUAGE plpgsql
 IMMUTABLE
AS $function$
DECLARE
    str_date text;
    str_hour text; 
    str_min text;
    v_ret timestamp; 
begin

    str_date = (dt::date)::text; 
    str_hour = (extract(hour from dt))::text;
    str_min = (extract(min from dt))::text;

    str_date = str_date || ' ' || str_hour || ':' || str_min || ':' || '00';
    v_ret = str_date::timestamp; 
    return v_ret; 

END;
$function$
;

为什么我们的函数是不可变的？为了获得高性能，我想在 PostgreSQL 上使用基于函数的索引。但是，PostgreSQL 只需要在索引上使用不可变函数。我们在加入过程的条件下使用这个函数。现在我们来看看创建基于函数的索引的过程：

CREATE INDEX customer_date_function_idx ON customer USING btree ((clear_seconds_from_datetime(datestamp)));

让我们编写我们需要的主要查询：

select 
    t1.list_dates, 
    cc.userid, 
    count(cc.id) as count_as
from 
    get_dates('2021-02-03 00:01:00'::timestamp) t1 (list_dates)
left join 
    customer cc on clear_seconds_from_datetime(cc.datestamp) = t1.list_dates
group by t1.list_dates, cc.userid

我将 3000 万个样本数据插入到 customer 表中进行测试。但是，这个查询运行了 33 毫秒。查看生成的explain analyze 命令的查询计划：

HashAggregate  (cost=309264.30..309432.30 rows=16800 width=20) (actual time=26.076..26.958 rows=2022 loops=1)
  Group Key: t1.list_dates, lg.userid
  ->  Nested Loop Left Join  (cost=0.68..253482.75 rows=7437540 width=16) (actual time=18.870..24.882 rows=2023 loops=1)
        ->  Function Scan on get_dates t1  (cost=0.25..10.25 rows=1000 width=8) (actual time=18.699..18.906 rows=1441 loops=1)
        ->  Index Scan using log_date_cast_idx on log lg  (cost=0.43..179.09 rows=7438 width=16) (actual time=0.003..0.003 rows=1 loops=1441)
              Index Cond: (clear_seconds_from_datetime(datestamp) = t1.list_dates)
Planning Time: 0.398 ms
JIT:
  Functions: 12
  Options: Inlining false, Optimization false, Expressions true, Deforming true
  Timing: Generation 3.544 ms, Inlining 0.000 ms, Optimization 0.816 ms, Emission 16.709 ms, Total 21.069 ms
Execution Time: 31.429 ms

结果：

list_dates	group_count
2021-02-03 09:41:00.000	1
2021-02-03 09:42:00.000	3
2021-02-03 09:43:00.000	1
2021-02-03 09:44:00.000	3
2021-02-03 09:45:00.000	1
2021-02-03 09:46:00.000	5
2021-02-03 09:47:00.000	2
2021-02-03 09:48:00.000	1
2021-02-03 09:49:00.000	1
2021-02-03 09:50:00.000	1
2021-02-03 09:51:00.000	4
2021-02-03 09:52:00.000	0
2021-02-03 09:53:00.000	0
2021-02-03 09:54:00.000	2
2021-02-03 09:55:00.000	1