Support for first(returnField,searchField) and last(returnField,searchField) function

Use case: In timeseries use cases one might often want to get the latest or first value for a specific device/machine/thing. This can be achieved with sub-queries, which might become rather cumbersome and/or slow.

Therefore it would be convenient to have a (fast) function that returns the value of another field/column when searching for the max/min value of another column.

Feature description:

first(returnField, searchField)
last(returnField, searchField)

first() returns the value of column returnField where searchField is the minimum. last() returns the value of column returnField where searchField is the maximum.

e.g. get the latest reported speed for every device:

SELECT device, last(speed, ts)
FROM data
GROUP BY device;

(slow) workaround:

SELECT device, speed
FROM data a,
     (
         SELECT max(ts) AS ts, device
         FROM data
         GROUP BY device
     ) b
WHERE (a.ts = b.ts)
  AND (a.device = b.device)

---

Some DBs call these max_by and min_by

[proddata]

CTE + JOIN

WITH cte AS (SELECT device_id, MAX(ts) ts FROM data GROUP BY 1)
SELECT r.*
FROM data r JOIN cte using(device_id,ts)
WHERE r.time >= (SELECT min(ts) FROM cte);

CTE/JOIN: 12 seconds on the same set 🚀 caveat: only works fast, if all devices report values regularly .

alternative

WITH cte AS (SELECT device_id, MAX(ts) ts FROM data GROUP BY 1)
SELECT r.*
FROM data r JOIN cte using(device_id,ts)
WHERE r.time IN (SELECT ts FROM cte)

---

CTE + concatenated predicates

WITH cte AS (SELECT device_id, MAX(ts) ts FROM data group by 1)
SELECT r.*
FROM data r
where
  r.ts in (SELECT ts FROM cte)
  AND
  CONCAT(ts,',',device_id) in (SELECT CONCAT(ts,',',device_id) FROM cte);
-- 6.189 seconds 🚀

CTE + object comparison

WITH CTE AS (SELECT device_id, MAX(ts) ts FROM data group by 1)
SELECT r.*
FROM data r
where
  r.ts in (SELECT ts FROM cte) AND
  {ts=ts, device_id=device_id} in (SELECT {ts=ts, device_id=device_id} FROM cte);
-- 8.818 seconds

[hlcianfagna]

There is another faster workaround

SELECT
    device,
    split_part(
        MAX(
            CONCAT (date_format(ts), ',', speed::STRING)
        )
        /* optionally FILTER(WHERE speed  IS NOT NULL)
         to mimic IGNORE NULLS OVER ... */
        ',',
        2)::DOUBLE AS last_speed_in_groupbybucket_over_orderby_ts
FROM
    data
GROUP BY
    device;