MySQL: troubles with HAVING COUNT 'exact', not 'at least' - mysql

Question

I have a table that holds relations of users participating in conversations like follows:
CREATE TABLE `so` (
`id` int(11) NOT NULL AUTO_INCREMENT PRIMARY KEY,
`user_id` int(11) NOT NULL,
`conversation_id` int(11) NOT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8;
ALTER TABLE `so`
ADD UNIQUE KEY `uc` (`user_id`,`conversation_id`) USING BTREE;
INSERT INTO `so` (`id`, `user_id`, `conversation_id`) VALUES
(1, 1, 1),
(3, 1, 2),
(2, 2, 1),
(4, 2, 2),
(5, 3, 2);
According to sample data, users 1 and 2 have conversation with ID of 1 and users 1, 2, 3 - conversation with ID of 2.
I need to get unique conversation_id for the list of user ids.
My current query is:
SELECT conversation_id, COUNT(user_id) as usersCount
FROM so
WHERE user_id IN (1,2)
GROUP BY conversation_id
HAVING usersCount = 2
ORDER BY NULL
But it returns 2 rows for both conversations and I expect the row with conversation_id of 1.
How can I select the row that belongs exactly to users 1 and 2, and not to 1, 2, 3? Thanks.
UPDATE:
I can't use subqueries on joins for performance reasons because users list in the query may be up to 30 ids and I'm afraid 30 subqueries is not the case.

Answer 1

You can use group_concat
select conversation_id
from so
group by conversation_id
having group_concat(user_id order by user_id) = '1,2';
To avoid full index scan, you can put your original query in a subquery:
SELECT a.conversation_id
FROM (
SELECT conversation_id
FROM so
WHERE user_id IN (1,2)
GROUP BY conversation_id
HAVING COUNT(conversation_id) = 2) a
JOIN so b ON a.conversation_id = b.conversation_id
GROUP BY a.conversation_id
HAVING COUNT(a.conversation_id) = 2;

Answer 2

Instead of checking the user_id in the WHERE clause, compare the number of rows that satisfy that condition with the total rows for each conversation.
SELECT conversation_id, COUNT(*) AS allCount, SUM(user_id IN (1, 2)) AS userCount
FROM so
GROUP BY conversation_id
HAVING allCount = 2 AND allCount = userCount

Answer 3

This answer is an alternative to the already given, and will provide better efficiency through not using sub-selects.
HAVING COUNT(user_id IN ('1','2') OR NULL) > 0 specifies that you want conversations with userid 1 and 2.
COUNT(user_id) = 2 says that there can only be 2 users in the conversation.
You could even remove COUNT(user_id) as usersCount from the result set if you don't actually use it as part of your exercise.
SELECT conversation_id, COUNT(user_id) as usersCount
FROM so
GROUP BY conversation_id
HAVING COUNT(user_id IN ('1','2') OR NULL) > 0 AND
COUNT(user_id) = 2;
To avoid a full index scan you would have to use a where clause as #Fabricator has shown in his answer. When you apply conditions to groups of rows, it has to group them first, and then do the aggregations and conditions, and a where clause only applies conditions to single rows. How big is your table out of interest?