Mysql总和基于包含多个LEFT JOIN的其他列

Question

Ive got 5 tables that I'd like to LEFT JOIN together. Tables are : visitors, offers, contracts1, contracts2 and contracts3.

QUERY:

SELECT 
        count(DISTINCT visitors.ID) as visitors, 
        sum(
        CASE
        WHEN offers.ACTIVE = 1 THEN 1
        ELSE 0
        END) as offers, 
        count(contracts1.ID) as contracts1, sum(contracts1.PRICE) as sum_contracts1, 
        count(contracts2.ID) contracts2, 
        sum(
        CASE
        WHEN contracts2.PAYMENT = 'YEARLY' THEN contracts2.PRICE
        WHEN contracts2.PAYMENT = 'TWICE' THEN contracts2.PRICE*2
        ELSE contracts2.PRICE*4
        END) as sum_contracts2,
        count(contracts3.ID) as contracts3, sum(contracts3.PRICE) as sum_contracts3
        FROM visitors 
        LEFT JOIN offersON offers.VISITOR_ID = visitors.ID AND (offers.IP > 100 OR offers.IP < 0)
        LEFT JOIN contracts1 ON 
        (offers.ID = contracts1.ID_OFFER)
        LEFT JOIN contracts2 ON 
        (offers.ID = contracts2.ID_OFFER)
        LEFT JOIN contracts3 ON 
        (offers.ID = contracts3.ID_OFFER)
        WHERE  visitors.TIME >= '2017-01-01 00:00:00' AND visitors.TIME <= '2017-05-25 23:59:59'

Problem here is, that contracts1, contracts2 and contracts3 have no common column in order to be joined together. So instead of 20 rows for contracts1, 30 for contracs2 and 50 for contracts3 i get all the combination for all of them. Because they are joined based on visitors and offers tables. Simple GROUP BY in the end of the query would normally solve the problem, but if I use GROUP BY in the END for one of those tables (or all of them), it will create MULTIPLE ROWS instead of 1 that I want. And also it would erase all the other result for the part where i count visitors by ID and also offers by ID ... I can use DISTINCT on count() parts of the SELECT but not one the sum() because PRICE of the contracts may be same even though IDs are not (you know like for example 2 chocolates are 2 rows with different IDs but same PRICE for 10 dollars each).

So my question is:

Is there any way to SUM only those PRICES of contracts1, contracts2 and contracts3, that have DISTINCT ID and though get rid of adding up the duplicates? And is it possible without creating VIEW?

I also tried GROUP BY inside of the LEFT JOIN but again when i LEFT JOINED all 3 contracts tables together, even though i GROUPED them before I ended up with duplicates.

Example of expected result:

In that time horizon which I stated above I would expect: 80 visitors that have 35 offers and 5 contracts1 with sum of 1000 euros, 12 contracts2 with sum of 686 euros and 3 contracts3 with sum of 12 euros. It is ONE ROW with 8 columns of data.

Instead of expected result I got: 80 visitors, 35 offers, 180 contracts1 (sum is also bad), 180 contracts2 (sum is also bad), 180 contracts3 (sum is also bad).

Answer 1

With CTEs ( Supported by MariaDB 10.2.1 ) I would write something like this:

WITH v AS (
    SELECT ID as VISITOR_ID
    FROM visitors 
    WHERE visitors.TIME >= '2017-01-01 00:00:00'
      AND visitors.TIME <= '2017-05-25 23:59:59'
), o AS (
    SELECT offers.ID as ID_OFFER
    FROM v
    JOIN offers USING(VISITOR_ID)
    WHERE offers.ACTIVE = 1
      AND (offers.IP > 100 OR offers.IP < 0)
), c1 AS (
    SELECT count(*) as contracts1, sum(contracts1.PRICE) as sum_contracts1
    FROM o JOIN contracts1 USING(ID_OFFER)
), c2 AS (
    SELECT
        count(*) contracts2, 
        sum(CASE contracts2.PAYMENT
            WHEN 'YEARLY' THEN contracts2.PRICE
            WHEN 'TWICE'  THEN contracts2.PRICE*2
            ELSE contracts2.PRICE*4
        END) as sum_contracts2
    FROM o JOIN contracts2 USING(ID_OFFER)
), c3 AS (
    SELECT count(*) as contracts3, sum(contracts3.PRICE) as sum_contracts3
    FROM o JOIN contracts3 USING(ID_OFFER)
)
    SELECT c1.*, c2.*, c3.*,
        (SELECT count(*) FROM v) as visitors,
        (SELECT count(*) FROM o) as offers,
    FROM c1, c2, c3;

Without CTEs you can rewrite it to use temporary tables:

CREATE TEMPORARY TABLE v AS
    SELECT ID as VISITOR_ID
    FROM visitors 
    WHERE visitors.TIME >= '2017-01-01 00:00:00'
      AND visitors.TIME <= '2017-05-25 23:59:59';

CREATE TEMPORARY TABLE o AS
    SELECT offers.ID as ID_OFFER
    FROM v
    JOIN offers USING(VISITOR_ID)
    WHERE offers.ACTIVE = 1
      AND (offers.IP > 100 OR offers.IP < 0);

CREATE TEMPORARY TABLE c1 AS
    SELECT count(*) as contracts1, sum(contracts1.PRICE) as sum_contracts1
    FROM o JOIN contracts1 USING(ID_OFFER);

CREATE TEMPORARY TABLE c2 AS
    SELECT
        count(*) contracts2, 
        sum(CASE contracts2.PAYMENT
            WHEN 'YEARLY' THEN contracts2.PRICE
            WHEN 'TWICE'  THEN contracts2.PRICE*2
            ELSE contracts2.PRICE*4
        END) as sum_contracts2
    FROM o JOIN contracts2 USING(ID_OFFER);

CREATE TEMPORARY TABLE c3 AS
    SELECT count(*) as contracts3, sum(contracts3.PRICE) as sum_contracts3
    FROM o JOIN contracts3 USING(ID_OFFER);

SELECT c1.*, c2.*, c3.*,
    (SELECT count(*) FROM v) as visitors,
    (SELECT count(*) FROM o) as offers,
FROM c1, c2, c3;

Answer 2

Just a proof of concept where I don't account for the time and activity constraints as well as the payment type, but couldn't it be something along those lines?

SELECT
   VISITOR_ID,
   SUM(CASE WHEN TYPE="contract1" THEN 1 else 0 END) as c1_count,
   SUM(CASE WHEN TYPE="contract1" THEN PRICE else 0 END) as c1_total_price,
   SUM(CASE WHEN TYPE="contract2" THEN 1 else 0 END) as c2_count,
   SUM(CASE WHEN TYPE="contract2" THEN PRICE else 0 END) as c2_total_price,
   SUM(CASE WHEN TYPE="contract3" THEN 1 else 0 END) as c3_count,
   SUM(CASE WHEN TYPE="contract3" THEN PRICE else 0 END) as c3_total_price 
FROM (
    (SELECT "contract1" as TYPE, ID, PRICE, ID_OFFER, PAYMENT FROM contracts1) 
    UNION
    (SELECT "contract2" as TYPE, ID, PRICE, ID_OFFER, PAYMENT FROM contracts2)
    UNION
    (SELECT "contract3" as TYPE, ID, PRICE, ID_OFFER, PAYMENT FROM contracts3)
 ) as all_contracts 
 JOIN offers on offers.id = all_contracts.ID_OFFER
 JOIN visitors on visitors.ID = offers.VISITOR_ID
 GROUP BY visitors.ID

The idea is that first you merge the different contracts into one result where you store their type in a column called "TYPE" (that's the purpose of the UNION queries) and once you have such a nice table where each contract is exactly once, you can get your desired result quite straightforward. I just outlined how you get the sum and count for each type of contract. Of course, the final query would be a bit more complicated but the core idea should be the same.

But despite your statement that you don't want to use (temporary) views, I would encourage you to try it - I have a feeling that putting those "all_contracts" joined with offers and visitors into a temporary view would improve the performance, if that's your concern, without making the query too ugly, mainly in the case when you would want to see the stats just for one visitor or to filter them further (by time, activity and so on), because unnecessary rows won't be materialized. But that's just an impression since I haven't tried the query on a bigger data set - you can play with it.