We have a table where total presence time on a given (period, volunteer) pair is saved and another in which actions are logged: we know which presence the action was taken in. But we do not track time spent on a given action. It is not a great data structure but I cannot change it.
CREATE TABLE volunteer_presence (id integer, volunteer_id integer, minutes integer);
CREATE TABLE logged_actions (id integer, presence_id integer, section varchar);
What we would like to do is to approximate time spent on a given action by assuming time was split equally among all actions in a given presence. It's not great but we are happy to at least have that. For example let us say we have the following data:
INSERT INTO volunteer_presence (id, volunteer_id, minutes) values (1, 333, 400);
INSERT INTO volunteer_presence (id, volunteer_id, minutes) values (2, 444, 90);
INSERT INTO volunteer_presence (id, volunteer_id, minutes) values (3, 555, 80);
INSERT INTO logged_actions (id, presence_id, section) values (10, 1, 'Cats');
INSERT INTO logged_actions (id, presence_id, section) values (11, 1, 'Dogs');
INSERT INTO logged_actions (id, presence_id, section) values (13, 1, 'Cats');
INSERT INTO logged_actions (id, presence_id, section) values (14, 1, 'Cats');
INSERT INTO logged_actions (id, presence_id, section) values (15, 1, 'Front');
INSERT INTO logged_actions (id, presence_id, section) values (16, 2, 'Dogs');
We would expect the following result:
| section | presence_id | estimated_minutes |
|---|---|---|
| Cats | 1 | 240.0 |
| Dogs | 1 | 80.0 |
| Front | 1 | 80.0 |
| Dogs | 2 | 90.0 |
For example (3 out of 5 actions)*400 is 240.
The solution I came up with feels wrong with two subqueries in the join clauses. It feels it would not scale well.
SELECT action_count_table.section,
action_count_table.presence_id,
((action_count_table.total_action_count / presence_total_actions.action_count) * time_table.minutes) AS estimated_minutes
FROM (
SELECT event.section AS section,
volunteer_presence.id AS presence_id,
SUM(event.action_count) AS total_action_count
FROM (
SELECT logged_actions.presence_id,
logged_actions.section,
COUNT(logged_actions.id) AS action_count
FROM logged_actions
GROUP BY logged_actions.presence_id, logged_actions.section
ORDER BY logged_actions.presence_id, logged_actions.section
) AS event
INNER JOIN volunteer_presence
ON volunteer_presence.id = event.presence_id
GROUP BY event.section, volunteer_presence.id
) AS action_count_table
INNER JOIN (
SELECT DISTINCT logged_actions.presence_id AS presence_id,
volunteer_presence.minutes AS minutes
FROM logged_actions
INNER JOIN volunteer_presence
ON volunteer_presence.id = logged_actions.presence_id
) AS time_table
ON action_count_table.presence_id = time_table.presence_id
INNER JOIN (
SELECT logged_actions.presence_id,
COUNT(logged_actions.id) AS action_count
FROM logged_actions
GROUP BY logged_actions.presence_id
ORDER BY logged_actions.presence_id
) AS presence_total_actions
ON presence_total_actions.presence_id = time_table.presence_id
Is there a better solution, that would scale better or at least be simpler ?
SELECT
la.presence_id,
la.section,
vp.minutes * COUNT(*) / total -- 4
FROM (
SELECT
*,
COUNT(*) OVER (PARTITION BY presence_id) as total -- 1
FROM logged_actions
) la
JOIN volunteer_presence vp ON vp.id = la.presence_id -- 2
GROUP BY la.presence_id, la.section, la.total, vp.minutes -- 3
ORDER BY la.presence_id, la.section
COUNT() window function adds the total count per presence_id to each recordvolunteer_presencesection per presence_id, divide by the previously calculated total count per presence_id and multiply with the minutes.