CODEWITHSUNDEEP
sqlpostgresqldatetimegroup-byleft-join
ittus
ittus

Reputation: 22403

Return 0 for all missing combinations in the joined table

I have 2 tables, timeseries and orders

timeseries

+------------+
|  datetime  |
+------------+
| 2018-11-01 |
| 2018-11-02 |
| 2018-11-03 |
+------------+

orders

+------------+-------------+----------+
|  datetime  | customer_id | order_id |
+------------+-------------+----------+
| 2018-11-01 |           1 |        1 |
| 2018-11-02 |           1 |        2 |
| 2018-11-03 |           2 |        3 |
+------------+-------------+----------+

I would like to get number of orders for each customer in each day.

Expected results:

+------------+-------------+--------------+
|  datetime  | customer_id | number_order |
+------------+-------------+--------------+
| 2018-11-01 |           1 |            1 |
| 2018-11-02 |           1 |            1 |
| 2018-11-03 |           1 |            0 |
| 2018-11-01 |           2 |            0 |
| 2018-11-02 |           2 |            0 |
| 2018-11-03 |           2 |            1 |
+------------+-------------+--------------+

I tried LEFT JOIN, but it doesn't return all time series for all customer

SELECT datetime, customer_id, COUNT(order_id) as number_order 
FROM timeseries
LEFT JOIN orders
ON timeseries.datetime = orders.datetime
GROUP BY datetime, customer_id
ORDER BY datetime, customer_id

>> Result

+------------+-------------+--------------+
|  datetime  | customer_id | number_order |
+------------+-------------+--------------+
| 2018-11-01 |           1 |            1 |
| 2018-11-02 |           1 |            1 |
| 2018-11-03 |           2 |            1 |
+------------+-------------+--------------+

I understand left join only ensures returning all rows in table timeseries, but what I need is all rows in table timeseries with each customer_id.

Thank for your help!

Upvotes: 2

Views: 64

Answers (2)

Salman Arshad
Salman Arshad

Reputation: 272106

You need to cross join all dates and all customers to get all possibile combinations of date and customer ids. Then left join with orders:

SELECT timeseries.datetime, customers.customer_id, COUNT(orders.order_id) as number_order
FROM timeseries
CROSS JOIN (SELECT DISTINCT customer_id FROM orders) AS customers
LEFT JOIN orders ON orders.datetime = timeseries.datetime AND orders.customer_id = customers.customer_id
GROUP BY timeseries.datetime, customers.customer_id
ORDER BY timeseries.datetime, customers.customer_id

Upvotes: 5

Zaynul Abadin Tuhin
Zaynul Abadin Tuhin

Reputation: 31993

use union all then left join

select t1.datetime,customer_id,COUNT(order_id) as number_order from 

(select datetime from  timeseries
union all
select datetime from orders
) t1 left join 
    orders  on t1.datetime=orders.datetime
group by  t1.datetime,customer_id

Upvotes: 0

Related Questions