Question

Given the following table:

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

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

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

I can’t find a way (without creating additional tables) to write a query that fills the time gaps (for example, 15:29:00 for itemId 101, and 15:31:00 for both items) by returning NULL in value.

The expected resultset would be:

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

I’ve found solutions having a separate time table with the full serie of timestamps, but I would much prefer to solve this in the query alone. Is this possible?

Answer 1

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
        ;

Output:

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)