MySql：三个级别的相关表上的 SUM/COUNT

Question

I'm trying to do some stats on my tables and I need your help. I have three tables : "sessions", "subscriptions" and "invoices"

So a "session" can have multiple "subscribers", and each subscriber can have multiple invoices. The goal here is to show a summarize in one query :

+-----------+--------------------+--------------------+---------------------+
|  Session  | Subscription count | Estimated cost sum | Invoices amount sum |
+-----------+--------------------+--------------------+---------------------+
| Session 1 |                  2 |                 15 |                  16 |
| Session 2 |                  1 |                 20 |                  20 |
| Session 3 |                  2 |                 50 |                  40 |
+-----------+--------------------+--------------------+---------------------+

You can find corresponding SQL structure and data :

SET NAMES utf8;
SET foreign_key_checks = 0;
SET sql_mode = 'NO_AUTO_VALUE_ON_ZERO';

DROP TABLE IF EXISTS `stack_invoices`;
CREATE TABLE `stack_invoices` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `inscription_id` int(10) unsigned NOT NULL,
  `amount` double NOT NULL,
  PRIMARY KEY (`id`),
  KEY `inscription_id` (`inscription_id`),
  CONSTRAINT `stack_invoices_ibfk_1` FOREIGN KEY (`inscription_id`) REFERENCES `stack_subscriptions` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

INSERT INTO `stack_invoices` (`id`, `inscription_id`, `amount`) VALUES
(1, 1,  5),
(2, 1,  6),
(3, 2,  5),
(4, 3,  10),
(5, 3,  5),
(6, 3,  5),
(7, 4,  40);

DROP TABLE IF EXISTS `stack_sessions`;
CREATE TABLE `stack_sessions` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `title` varchar(10) NOT NULL,
  PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

INSERT INTO `stack_sessions` (`id`, `title`) VALUES
(1, 'Session 1'),
(2, 'Session 2'),
(3, 'Session 3');

DROP TABLE IF EXISTS `stack_subscriptions`;
CREATE TABLE `stack_subscriptions` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `name` varchar(15) NOT NULL,
  `estimated_cost` double DEFAULT NULL,
  `session_id` int(10) unsigned NOT NULL,
  PRIMARY KEY (`id`),
  KEY `session_id` (`session_id`),
  CONSTRAINT `stack_subscriptions_ibfk_1` FOREIGN KEY (`session_id`) REFERENCES `stack_sessions` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

INSERT INTO `stack_subscriptions` (`id`, `name`, `estimated_cost`, `session_id`) VALUES
(1, 'Mister A', 10, 1),
(2, 'Mister B', 5,  1),
(3, 'Mister C', 20, 2),
(4, 'Mister D', 50, 3),
(5, 'Mister E', NULL,   3);

Here are my attempts : Showing estimated cost sum :

SELECT
  stack_sessions.title as session,
  SUM(stack_subscriptions.estimated_cost) as estimated_cost_sum
FROM stack_sessions
LEFT JOIN 
  stack_subscriptions 
ON
  stack_subscriptions.session_id = stack_sessions.id
GROUP BY
  stack_sessions.id

It is ok.

Adding subscription count :

SELECT
  stack_sessions.title as session,
  (
    SELECT 
      COUNT(subscriptions1.id) 
    FROM 
      stack_subscriptions subscriptions1 
    WHERE 
      subscriptions1.session_id = stack_sessions.id
  ) as subscriptions_count,
  SUM(stack_subscriptions.estimated_cost) as estimated_cost_sum
FROM stack_sessions
LEFT JOIN 
  stack_subscriptions 
ON
  stack_subscriptions.session_id = stack_sessions.id
GROUP BY
  stack_sessions.id

It is ok.

But when I want to show the sum of all invoice group by the session... It no longer work :

SELECT
  stack_sessions.title as session,
  (
    SELECT 
      COUNT(subscriptions1.id) 
    FROM 
      stack_subscriptions subscriptions1 
    WHERE 
      subscriptions1.session_id = stack_sessions.id
  ) as subscriptions_count,
  SUM(stack_subscriptions.estimated_cost) as estimated_cost_sum,
  SUM(stack_invoices.amount) as invoices_amount_sum
FROM stack_sessions
LEFT JOIN 
  stack_subscriptions 
ON
  stack_subscriptions.session_id = stack_sessions.id
LEFT JOIN
  stack_invoices
ON
  stack_invoices.inscription_id = stack_subscriptions.id
GROUP BY
  stack_sessions.id

With this attempt, data are correct for subscriptions count and invoices amount sum but not for estimated cost sum .

I tried another way :

SELECT
  stack_sessions.title as session,
  (
    SELECT 
      COUNT(subscriptions1.id) 
    FROM 
      stack_subscriptions subscriptions1 
    WHERE 
      subscriptions1.session_id = stack_sessions.id
  ) as subscriptions_count,
  SUM(stack_subscriptions.estimated_cost) as estimated_cost_sum,
  (
    SELECT
      SUM(stack_invoices.amount) as invoices_amount_sum
    FROM 
      stack_invoices
    WHERE
      stack_invoices.inscription_id = stack_subscriptions.id
  )
FROM stack_sessions
LEFT JOIN 
  stack_subscriptions 
ON
  stack_subscriptions.session_id = stack_sessions.id
GROUP BY
  stack_sessions.id

But MySql ask me to group by correctly and I can't find the good columns to group by.

I don't know if it is possible in one query ? Perhaps there is some SQL notion that I miss...

Thanks for your help,

Answer 1

I think one of the solutions might be this:

SELECT
  stack_sessions.title AS session,
  ANY_VALUE(stack_subscription_summary.subscriptions_count) AS subscriptions_count,
  ANY_VALUE(stack_subscription_summary.estimated_cost_sum) AS estimated_cost_sum,
  SUM(stack_invoices.amount) AS invoices_amount_sum
FROM
    stack_sessions
    LEFT JOIN (
        SELECT
            session_id AS session_id,
            SUM(estimated_price) AS estimated_cost_sum,
            COUNT(id) AS subscriptions_count
        FROM
            stack_subscriptions
        GROUP BY
            session_id
    ) AS stack_subscription_summary ON
        stack_sessions.id = stack_subscription_summary.session_id
    LEFT JOIN stack_subscriptions ON
        stack_subscriptions.session_id = stack_sessions.id
    LEFT JOIN stack_invoices ON
        stack_invoices.inscription_id = stack_subscriptions.id
GROUP BY
  stack_sessions.id

How does it solve your issue?

Your code stopped working because you were doing SUM of nondistinct rows - meaning you added a value of subscription more then once if it had more invoices.

That means you need to group some of your data in a subquery. I'm showing this on grouping subscription data (therefore getting their summaries), but the probably more elegant way (and probably faster) would be to group invoice data like this:

SELECT
  stack_sessions.title AS SESSION,
  COUNT(stack_subscriptions.id) AS subscriptions_count,
  SUM(stack_subscriptions.estimated_cost) AS estimated_cost_sum,
  SUM(stack_invoices_summary.amount) AS invoices_amount_sum
FROM
  stack_sessions
  LEFT JOIN stack_subscriptions ON
    stack_subscriptions.session_id = stack_sessions.id
  LEFT JOIN (
    SELECT  
      inscription_id AS inscription_id,
      SUM(amount) AS amount
    FROM
      stack_invoices
    GROUP BY
      inscription_id
  ) AS stack_invoices_summary ON
    stack_invoices_summary.inscription_id = stack_subscriptions.id
GROUP BY
  stack_sessions.id

I haven't tested this on your data, so let me know if it helps!