7

给定下表：

CREATE TABLE channel1m (
  ts TIMESTAMP WITHOUT TIME ZONE NOT NULL,
  itemId BIGINT,
  value BIGINT
)

每个 itemId 每分钟可以插入一行，如下所示：

ts                    itemId         value
2012-12-03 15:29:00   100            1
2012-12-03 15:30:00   100            2
2012-12-03 15:30:00   101            0
2012-12-03 15:32:00   100            1
2012-12-03 15:32:00   101            1

我找不到一种方法（不创建额外的表）来编写一个通过返回NULL值来填充时间间隔的查询（例如，对于 itemId 101 的 15:29:00 和两个项目的 15:31:00）。

预期的结果集将是：

ts                    itemId         value
2012-12-03 15:29:00   100            1
2012-12-03 15:29:00   101            NULL
2012-12-03 15:30:00   100            2
2012-12-03 15:30:00   101            0
2012-12-03 15:31:00   100            NULL
2012-12-03 15:31:00   101            NULL
2012-12-03 15:32:00   100            1
2012-12-03 15:32:00   101            1

我找到了具有完整系列时间戳的单独时间表的解决方案，但我更愿意单独在查询中解决这个问题。这可能吗？

4

4 回答 4

11 
DROP SCHEMA tmp CASCADE;
CREATE SCHEMA tmp ;
SET search_path = tmp;

DROP TABLE IF EXISTS channel1m CASCADE;
CREATE TABLE channel1m (
  zts TIMESTAMP WITHOUT TIME ZONE NOT NULL,
  zitemid BIGINT,
  zvalue BIGINT
);

-- in which a row may be inserted each minute, per zitemid, as follows:

INSERT INTO channel1m(zts, zitemid, zvalue) VALUES
 ('2012-12-03 15:29:00',   100,            1)
,('2012-12-03 15:30:00',   100,            2)
,('2012-12-03 15:30:00',   101,            0)
,('2012-12-03 15:32:00',   100,            1)
,('2012-12-03 15:32:00',   101,            1)
        ;

        -- CTE to the rescue!!!
WITH cal AS (
        WITH mm AS (
                SELECT MIN(xx.zts) AS minmin, MAX(xx.zts) AS maxmax
                 FROM channel1m xx)
        SELECT generate_series(mm.minmin , mm.maxmax , '1 min'::interval) AS stamp
        FROM mm
        )
, ite AS (
        SELECT DISTINCT zitemid AS zitemid
        FROM channel1m
        )
SELECT cal.stamp
        , ite.zitemid
        , tab.zvalue
FROM cal
JOIN ite ON 1=1 -- Note: this is a cartesian product of the {time,id} -domains
LEFT JOIN channel1m tab ON tab.zts = cal.stamp AND tab.zitemid = ite.zitemid
ORDER BY stamp ASC
        ;

输出：

NOTICE:  drop cascades to table tmp.channel1m
DROP SCHEMA
CREATE SCHEMA
SET
NOTICE:  table "channel1m" does not exist, skipping
DROP TABLE
CREATE TABLE
INSERT 0 5
        stamp        | zitemid | zvalue 
---------------------+---------+--------
 2012-12-03 15:29:00 |     101 |       
 2012-12-03 15:29:00 |     100 |      1
 2012-12-03 15:30:00 |     100 |      2
 2012-12-03 15:30:00 |     101 |      0
 2012-12-03 15:31:00 |     100 |       
 2012-12-03 15:31:00 |     101 |       
 2012-12-03 15:32:00 |     100 |      1
 2012-12-03 15:32:00 |     101 |      1
(8 rows)
于 2012-12-05T17:22:38.040 回答
7

您将需要：带有 all 的表格itemId，以及带有所有必需日期的（伪）表格。

您可能拥有所有 distinct 的表itemId。让我们称之为item_table。

带有日期的伪表，您可以使用generate_series('start_date','end_date', interval '1 minute'). 详情在这里。

查询：

SELECT gs.ts, it.itemId, ch1m.value
FROM item_table it
CROSS JOIN generate_series('start_date','end_date', interval '1 minute') gs(ts)
LEFT JOIN channel1m ch1m ON it.itemId = ch1m.itemId 
                         AND gs.ts = ch1m.ts

替换'start_date','end_date'为所需的值或从子查询中获取它们。

这个查询：

1) 建立所有对 item-time 通过CROSS JOIN

2) 获取value通过LEFT JOIN

于 2012-12-05T17:20:25.887 回答
1

我认为最易读的方法是构建一系列表表达式。分钟和项目 ID 号之间的交叉连接将为您提供每种组合。

with all_minutes as (
  select ('2012-12-03 15:29'::timestamp + 
           (n || ' minute')::interval)::timestamp as ts
  from generate_series(0,10) n
),
item_ids as (
  select distinct itemid from channel1m
),
all_items_and_minutes as (
  select all_minutes.ts, item_ids.itemid from all_minutes cross join item_ids
)
select all_items_and_minutes.ts, all_items_and_minutes.itemId, channel1m.value
from all_items_and_minutes 
left join channel1m 
       on all_items_and_minutes.ts = channel1m.ts
      and all_items_and_minutes.itemid = channel1m.itemid
order by all_items_and_minutes.ts, all_items_and_minutes.itemid

您可以用 SELECT 语句替换时间戳文字以获得您需要的实际范围。如果您有一个包含所有唯一项目 ID 编号的不同表，则最好从该表中进行选择，而不是从 channel1m 表中选择不同的值。

于 2012-12-05T17:28:32.637 回答
0
  1. 使用time_bucket或date_trunc创建存储桶边界。
  2. 用于generate_series为您的时间范围生成空桶。
  3. 用于UNION将空时间桶数据集与您的数据合并。
  4. 用于DISTINCT ON为每个时间桶挑选唯一的行，优先选择那些有数据的行。

例子：

WITH
  timeseries_data AS (
    SELECT
      time_bucket(interval '5 minutes', started_at) time_bucket_start,
      count(distinct v1.value) unique_row_count
    FROM
      probe_execution pe1
    CROSS JOIN LATERAL (
      SELECT value
      FROM jsonb_array_elements(pe1.result)
    ) v1
    WHERE
      pe1.probe_id = 8 AND
      pe1.result_count > 0 AND
      pe1.started_at > now() - interval '1 day' AND
      pe1.ended_at < now()
    GROUP BY time_bucket_start
    UNION
    SELECT
      gs1 time_bucket_start,
      0 unique_row_count
    FROM
      generate_series(
        time_bucket(interval '5 minutes', now() - interval '1 day'),
        time_bucket(interval '5 minutes', now()),
        interval '1 minute'
      ) as gs1
  )
SELECT DISTINCT ON (td1.time_bucket_start)
  td1.time_bucket_start,
  td1.unique_row_count
FROM timeseries_data td1
ORDER BY td1.time_bucket_start, td1.unique_row_count DESC
于 2019-03-03T18:43:08.353 回答