Calculate sum and cumul by age group and by date (but people changes age group as time passes) - sql

Question

DBMS : postgreSQL
my problem :
In my database I have a person table with id and birth date, an events table that links a person, an event (id_event) and a date, an age table used for grouping ages. In the real database the person table is about 40 millions obs, and events 3 times bigger.
I need to produce a report (sum and cumul of X events) by age (age_group) and date (event_date). There isn't any problem to count the number of events by date. The problem lies with the cumul : contrary to other variables (sex for example), a person grow older and changes age group
as time passes, so for a given age group the cumul can increase then decrease. I want that the event's cumul, on every date in my report, uses the age of the persons on these dates.
Example of my inputs and desired output
The only way I found is to do a Cartesian product on the tables person and the dates v_dates, so it's easy to follow an event and make it change age_group. The code below uses this method.
BUT I can't use a cartesian product on my real data (makes a table way too big) and I need to use another method.
reproductible example
In this simplified example I want to produce a report by month from 2020-07-01 to 2022-07-01 (view v_dates). In reality I need to produce the same report by day but the logic remains the same.
My inputs
/* create table person*/
DROP TABLE IF EXISTS person;
CREATE TABLE person
(
person_id varchar(1),
person_birth_date date
);
INSERT INTO person
VALUES ('A', '2017-01-01'),
('B', '2016-07-01');
person_id
person_birth_date
A
2000-10-01
B
2010-02-01
/* create table events*/
DROP TABLE IF EXISTS events;
CREATE TABLE events
(
person_id varchar(1),
event_id integer,
event_date date
);
INSERT INTO events
VALUES ('A', 1, '2020-07-01'),
('A', 2, '2021-07-01'),
('B', 1, '2021-01-01'),
('B', 2, '2022-01-01');
person_id
event_id
event_date
A
1
2020-01-01
A
2
2021-01-01
B
1
2020-07-01
B
2
2021-01-01
/* create table age*/
DROP TABLE IF EXISTS age;
CREATE TABLE age
(
age integer,
age_group varchar(8)
);
INSERT INTO age
VALUES (0,'[0-4]'),
(1,'[0-4]'),
(2,'[0-4]'),
(3,'[0-4]'),
(4,'[0-4]'),
(5,'[5-9]'),
(6,'[5-9]'),
(7,'[5-9]'),
(8,'[5-9]'),
(9,'[5-9]');
/* create view dates : contains monthly dates from 2020-07-01 to 2022-07-01*/
CREATE or replace view v_dates AS
SELECT GENERATE_SERIES('2020-07-01'::date, '2022-07-01'::date, '6 month')::date as event_date;
age
age_group
0
[0-4]
1
[0-4]
5
[5-9]
My current method using a cartesian product
CROSS JOIN person * v_dates
with a LEFT JOIN to get info from table events
with a LEFT JOIN to get age_group from table age
CREATE or replace view v_person_event AS
SELECT
pdev.person_id,
pdev.event_date,
pdev.age,
ag.age_group,
pdev.event1,
pdev.event2
FROM
(
SELECT pd.person_id,
pd.event_date,
date_part('year', age(pd.event_date::TIMESTAMP, pd.person_birth_date::TIMESTAMP)) as age,
CASE WHEN ev.event_id = 1 THEN 1 else 0 END as event1,
CASE WHEN ev.event_id = 2 THEN 1 else 0 END as event2
FROM
(
SELECT *
FROM person
CROSS JOIN v_dates
) pd
LEFT JOIN events ev
on pd.person_id = ev.person_id
and pd.event_date = ev.event_date
) pdev
Left JOIN age as ag on pdev.age = ag.age
ORDER by pdev.person_id, pdev.event_date;
add columns event1_cum and event2_cum
CREATE or replace view v_person_event_cum AS
SELECT *,
SUM(event1) OVER (PARTITION BY person_id ORDER BY event_date) event1_cum,
SUM(event2) OVER (PARTITION BY person_id ORDER BY event_date) event2_cum
FROM v_person_event;
SELECT * FROM v_person_event_cum;
person_id
event_date
age
age_group
event1
event2
event1_cum
event2_cum
A
2020-07-01
3
[0-4]
1
0
1
0
A
2021-01-01
4
[0-4]
0
0
1
0
A
2021-07-01
4
[0-4]
0
1
1
1
A
2022-01-01
5
[5-9]
0
0
1
1
A
2022-07-01
5
[5-9]
0
0
1
1
B
2020-07-01
4
[0-4]
0
0
0
0
B
2021-01-01
4
[0-4]
1
0
1
0
B
2021-07-01
5
[5-9]
0
0
1
0
B
2022-01-01
5
[5-9]
0
1
1
1
B
2022-07-01
6
[5-9]
0
0
1
1
desired output : create a report grouped by variables age_group and event_date
SELECT
age_group,
event_date,
SUM(event1) as event1,
SUM(event2) as event2,
SUM(event1_cum) as event1_cum,
SUM(event2_cum) as event2_cum
FROM v_person_event_cum
GROUP BY age_group, event_date
ORDER BY age_group, event_date;
age_group
event_date
event1
event2
event1_cum
event2_cum
[0-4]
2020-07-01
1
0
1
0
[0-4]
2021-01-01
1
0
2
0
[0-4]
2021-07-01
0
1
1
1
[5-9]
2021-07-01
0
0
1
0
[5-9]
2022-01-01
0
1
2
2
This is why this is not an ordinary cumul : for the age_group [0-4], event1_cum goes from 2 at '2021-01-01' to 1 at '2021-07-01' because A was in [0-4] at the time of the event 1, still in [0-4] at '2021-01-01' but in [5-9] at 2021-07-01
When we read the report:
the 2021-01-01, there was 2 person between 0 and 4 (at that date) who had event1 and 0 person who had event2.
the 2021-07-01, there was 1 person between 0 and 4 who had event1 and 1 person who had event2.
I can't get a solution to this problem without using a cartesian Product...
Thanks in advance!