Count grouping by specific date intervals - sql

Question

I want to count the amount of "services" a customer have in 30 days period groups, since the contract start day. So I have to count the services in a monthly based period since his start date. Simplifying the table is something like this:
services
------------------
id serial
id_customer bigint
service_date date
Lets imaging there is only one type of service. I solve it like this:
SELECT
DATE_PART('year',service_date)||'-'|| CASE WHEN DATE_PART('day',service_date) >= 15 THEN
DATE_PART('month',service_date)
ELSE
CASE WHEN DATE_PART('month',service_date) = 1 THEN
12
ELSE
DATE_PART('month',service_date)-1
END
END bill, count(id)
FROM services
WHERE id_customer = 1
GROUP BY bill
results would be
bill | count
-------------------
2019-02 | 2455333
In the example the start date for id_customer 1 is 2019-02-15 but for 2019-02 period I will be counting the services until 2019-03-14.
What I want to know is, there is a better/more efficient solution?
I saw the solution here but implies an INNER JOIN with a GROUP BY with the same table which I think it would be slower, because my table has a lot of records.

Answer 1

You don't need to worry about the actual number of days in a month nor the month, year or day-of-month.
Just use the start date for a customer and let PostgreSQL generate the correct billing cycle periods for you.
To run a single query over all customers, I have used a separate table with the customer id as well as a billing_start date configured, for which we can then run a query such as the following:
WITH
periods (id, period_start, period_end) AS (
SELECT
id,
generate_series(billing_start, current_date, '1 month'::interval)::date,
(generate_series(billing_start, current_date, '1 month'::interval) + '1 month'::interval)::date
FROM test_customers
),
data AS (
SELECT
periods.id AS customer,
period_start,
count(test_services.*) AS service_calls
FROM periods INNER JOIN test_services ON (test_services.id_customer = periods.id)
WHERE test_services.service_date >= periods.period_start AND test_services.service_date < periods.period_end
GROUP BY 1, 2
)
SELECT customer, to_char(period_start, 'YYYY-MM') AS bill, service_calls
FROM data
ORDER BY 1, 2
;
...resulting in an output such as the following:
customer | bill | service_calls
----------+---------+---------------
1 | 2018-12 | 382736
1 | 2019-01 | 382735
1 | 2019-02 | 345696
2 | 2018-12 | 382736
2 | 2019-01 | 382734
2 | 2019-02 | 234580
3 | 2018-12 | 382734
3 | 2019-01 | 382736
3 | 2019-02 | 123463
4 | 2018-12 | 382734
4 | 2019-01 | 382736
4 | 2019-02 | 12346
5 | 2019-01 | 382735
5 | 2019-02 | 283965
6 | 2019-01 | 382735
6 | 2019-02 | 172848
7 | 2019-01 | 382734
7 | 2019-02 | 61732
8 | 2019-02 | 333351
9 | 2019-02 | 222234
10 | 2019-02 | 111117
(21 rows)
Complete online example: https://rextester.com/IHLJ95398
An important thing to note for this to be fast is a multi-column index on id_customer and service_date because that's where the counting takes place, which then can be done without sorting:
CREATE INDEX idx_svc_customer_date ON test_services (id_customer, service_date);
(otherwise, the sorting will most likely be done on disk, rather than in memory for large data sets)
If you just want the cycles for a single customer, use it like this:
WITH
periods (id, period_start, period_end) AS (
SELECT
id,
generate_series(billing_start, current_date, '1 month'::interval)::date,
(generate_series(billing_start, current_date, '1 month'::interval) + '1 month'::interval)::date
FROM test_customers WHERE id = 4
),
data AS (
SELECT
periods.id AS customer,
period_start,
count(test_services.*) AS service_calls
FROM periods INNER JOIN test_services ON (test_services.id_customer = periods.id)
WHERE test_services.service_date >= periods.period_start AND test_services.service_date < periods.period_end
GROUP BY 1, 2
)
SELECT customer, to_char(period_start, 'YYYY-MM') AS bill, service_calls
FROM data
ORDER BY 1, 2
;
...giving:
bill | service_calls
---------+---------------
2018-12 | 382734
2019-01 | 382736
2019-02 | 12346
(3 rows)