Using a column in sql join without adding it to group by clause

问题

My actual table structures are much more complex but following are two simplified table definitions:

Table invoice

CREATE TABLE invoice (
  id integer NOT NULL,
  create_datetime timestamp with time zone NOT NULL,
  total numeric(22,10) NOT NULL
);

id   create_datetime   total    
----------------------------
100  2014-05-08        1000

Table payment_invoice

CREATE TABLE payment_invoice (
  invoice_id integer,
  amount numeric(22,10)
);

invoice_id  amount
-------------------
100         100
100         200
100         150

I want to select the data by joining above 2 tables and selected data should look like:-

month      total_invoice_count  outstanding_balance
05/2014    1                    550

The query I am using:

select
to_char(date_trunc('month', i.create_datetime), 'MM/YYYY') as month,
count(i.id) as total_invoice_count,
(sum(i.total) - sum(pi.amount)) as outstanding_balance
from invoice i
join payment_invoice pi on i.id=pi.invoice_id
group by date_trunc('month', i.create_datetime)
order by date_trunc('month', i.create_datetime);

Above query is giving me incorrect results as sum(i.total) - sum(pi.amount) returns (1000 + 1000 + 1000) - (100 + 200 + 150) = 2550.
I want it to return (1000) - (100 + 200 + 150) = 550

And I cannot change it to i.total - sum(pi.amount), because then I am forced to add i.total column to group by clause and that I don't want to do.

回答1:

You need a single row per invoice, so aggregate payment_invoice first - best before you join.
When the whole table is selected, it's typically fastest to aggregate first and join later:

SELECT to_char(date_trunc('month', i.create_datetime), 'MM/YYYY') AS month
     , count(*)                                   AS total_invoice_count
     , (sum(i.total) - COALESCE(sum(pi.paid), 0)) AS outstanding_balance
FROM   invoice i
LEFT   JOIN  (
    SELECT invoice_id AS id, sum(amount) AS paid
    FROM   payment_invoice pi
    GROUP  BY 1
    ) pi USING (id)
GROUP  BY date_trunc('month', i.create_datetime)
ORDER  BY date_trunc('month', i.create_datetime);

LEFT JOIN is essential here. You do not want to loose invoices that have no corresponding rows in payment_invoice (yet), which would happen with a plain JOIN.

Accordingly, use COALESCE() for the sum of payments, which might be NULL.

SQL Fiddle with improved test case.

回答2:

Do the aggregation in two steps. First aggregate to a single line per invoice, then to a single line per month:

select
  to_char(date_trunc('month', t.create_datetime), 'MM/YYYY') as month,
  count(*) as total_invoice_count,
  (sum(t.total) - sum(t.amount)) as outstanding_balance
from (
    select i.create_datetime, i.total, sum(pi.amount) amount
    from invoice i
    join payment_invoice pi on i.id=pi.invoice_id
    group by i.id, i.total
) t
group by date_trunc('month', t.create_datetime)
order by date_trunc('month', t.create_datetime);

回答3:

See sqlFiddle

SELECT TO_CHAR(invoice.create_datetime, 'MM/YYYY') as month,
       COUNT(invoice.create_datetime) as total_invoice_count,
       invoice.total - payments.sum_amount as outstanding_balance
FROM invoice
JOIN 
(
    SELECT invoice_id, SUM(amount) AS sum_amount
    FROM payment_invoice
    GROUP BY invoice_id
) payments
ON invoice.id = payments.invoice_id
GROUP BY TO_CHAR(invoice.create_datetime, 'MM/YYYY'), 
         invoice.total - payments.sum_amount

来源：https://stackoverflow.com/questions/23572238/using-a-column-in-sql-join-without-adding-it-to-group-by-clause

标签

sql

postgresql

aggregate-functions