Postgresql - 24 hour rolling window - postgresql

Question

I am currently using Metabase to put together a live dashboard of some internal company metrics and one of the things I am trying to query is a 24 hour rolling window for transaction on our mobile app. Metabase has a useful visualization tool called "Smart Number" which allows you to compare changes in values over a defined time period. Like this.
I am having trouble writing a query that outputs data in 24 hour intervals so I can compare the past 24 hours to the 24 hours before that. I have tried using the date_trunc function to divide the transactions by hour and then possibly limit the results to the last 24 but it doesn't print out hours that don't have transactions. I also tried using the filter function as seen in the code below but the data needs to to be transposed for "Smart Numbers" to work. Does anyone have any suggestions as to how I should approach this problem?
Example of one of my approaches:
SELECT
(DATE_TRUNC('hour', (reservations.created_at::timestamptz))) as hour,
SUM(reservations.covers) as total_covers
FROM reservations
JOIN restaurants on restaurants.id = reservations.restaurant_id
WHERE reservations.origin = 'mobile'
and restaurants.relationship_type in ('listing_only', 'difficult', 'ipad')
GROUP BY hour
ORDER BY hour desc
Which outputs something like this:
hour total_covers
"2019-02-19 15:00:00+00" 4
"2019-02-19 13:00:00+00" 15
"2019-02-19 12:00:00+00" 4
"2019-02-19 11:00:00+00" 4
"2019-02-19 10:00:00+00" 26
"2019-02-19 09:00:00+00" 5
"2019-02-19 08:00:00+00" 8
"2019-02-19 07:00:00+00" 12
"2019-02-19 03:00:00+00" 2
I would like to get something like this:
Time_Interval Total_Covers
24 Hours 389
48 Hours 254
72 hours 459
96 Hours 239

Answer 1

This query uses the date function to group results.
SELECT
DATE(reservations.created_at) AS day,
SUM(reservations.covers) AS total_covers
FROM reservations
WHERE
reservations.origin = 'mobile' AND
restaurants.relationship_type IN ('listing_only', 'difficult', 'ipad')
GROUP BY day
ORDER BY day DESC
This query calculates the total covers from the time of the query, and groups them by the number of days ago.
SELECT
CEIL(EXTRACT(EPOCH FROM NOW() - reservations.created_at) / 86400) as days_ago
SUM(reservations.covers) as total_covers,
FROM reservations
WHERE
reservations.origin = 'mobile' AND
restaurants.relationship_type IN ('listing_only', 'difficult', 'ipad')
GROUP BY days_ago
ORDER BY days_ago;
See it in action on rextester.

Answer 2

Here is an example of using a generated series for the from, and then left-joining the data you want to count. Idea being you want every hour to appear in the result, and every piece of data to be counted, but it is optional.
Note I just checked the query compiled; I don't have a mock schema on hand to test this.
SELECT
hours.x as hour,
coalesce(SUM(reservations.covers), 0) as total_covers
FROM (
select * from generate_series(date_trunc('hour', (now() - INTERVAL '24 hours')), now(), '1 hour')
) hours(x)
LEFT JOIN reservations on hours.x = (DATE_TRUNC('hour', (reservations.created_at::timestamptz)))
LEFT JOIN restaurants on restaurants.id = reservations.restaurant_id
WHERE reservations.origin = 'mobile'
and restaurants.relationship_type in ('listing_only', 'difficult', 'ipad')
GROUP BY hours.x
ORDER BY hours.x desc