Clever Geek Handbook

In postgres, what's the best way to AND do with a join table?

I have different user categories and a connection table that allows users to be in multiple categories. My join table is named category_users and consists of user_id and category_id.

I want to filter for users who are in category 1 and category2. For example, I want to find everyone who is interested in both baseball and football.

What's the best way to do this in PostgreSQL? I have the following workers:

select * from users 
  where users.id IN 
    ( Select categories_users.user_id from categories_users 
      JOIN categories ON categories.id = categories_users.category_id 
      where categories.id = 1 OR categories.parent_id = 1) 
  AND users.id IN 
    (Select categories_users.user_id from categories_users 
    JOIN categories ON categories.id = categories_users.category_id 
    where categories.id = 2 OR categories.parent_id = 2)

However, this seems clunky and I'm wondering if there is a better way to do this. I've tried various joins but always end up looking for rows in the category_users table that have a category_id of 1 and 2, which is not possible.

Edit, I actually need to search in the parent category as well, so I modified the above query to include parent_id

+3


source to share


3 answers


Just join the same table twice (using aliases):



SELECT u.*
    FROM users u
    JOIN categories_users cu1 ON cu1.user_id = u.id
    JOIN categories_users cu2 ON cu2.user_id = u.id
    WHERE cu1.category_id = 1 AND cu2.category_id = 2
+2


source 


select u.*
from
    users u
    inner join (
        select user_id
        from categories_users
        group by user_id
        having
            bool_or(1 in (category_id, parent_id)) and
            bool_or(2 in (category_id, parent_id))
    ) s on s.user_id = u.id


+1


source


You can also use COUNT (*) above a section to see how many categories a user has in the set of searched categories.

I created the following sample to see how this can be defined and parameterized. I created a function test.find_users_in_categories(BIGINT[])

that accepts an array of categories for which we need a list of users. Thus, the function will return all users in all given categories.

Solution for - search for users in all specified categories

CREATE SCHEMA test;

CREATE TABLE test.categories_users (
  category_id BIGINT NOT NULL,
  user_id BIGINT NOT NULL
);

INSERT INTO test.categories_users
  (user_id, category_id)
  VALUES
  (33, 103),
  (34, 104),
  (35, 105),
  (37, 105),
  (35, 106),
  (37, 106);

CREATE OR REPLACE FUNCTION test.find_users_in_categories(BIGINT[])
  RETURNS TABLE (
    user_id BIGINT
  )
AS
$$
DECLARE
  categories ALIAS FOR $1;
BEGIN
  RETURN QUERY
  SELECT t.user_id
    FROM
      (
        SELECT
          cu.user_id,
          cu.category_id,
          COUNT(*) OVER (PARTITION BY cu.user_id ) AS cnt
        FROM test.categories_users AS cu
        WHERE cu.category_id = ANY(categories)
      ) AS t
      WHERE t.cnt = array_length(categories, 1)
      GROUP BY t.user_id;
END;
$$
LANGUAGE plpgsql;

SELECT * FROM test.find_users_in_categories(ARRAY[105, 106]);

DROP SCHEMA test CASCADE;

EDIT - [recursive solution]

Solution for - search for users in all specified categories and subcategories

See the following code for implementing a solution using JOIN + recursive CTE. I used JOIN instead of COUNT () because it looks better for this case.

CREATE SCHEMA test;

CREATE TABLE test.categories (
  category_id BIGINT PRIMARY KEY,
  parent_id BIGINT REFERENCES test.categories(category_id)
);

CREATE TABLE test.categories_users (
  category_id BIGINT NOT NULL REFERENCES test.categories(category_id),
  user_id BIGINT NOT NULL
);

INSERT INTO test.categories
  (category_id, parent_id)
  VALUES
  (100, NULL),
  (101, 100),
  (102, 100),
  (103, 101),
  (104, 101),
  (105, 101),
  (106, NULL);


INSERT INTO test.categories_users
  (user_id, category_id)
  VALUES
  (33, 103),
  (34, 104),
  (35, 105),
  (37, 105),
  (35, 106),
  (37, 106);


CREATE OR REPLACE FUNCTION test.find_users_in_categories(BIGINT[])
  RETURNS TABLE (
    user_id BIGINT
  )
AS
$$
DECLARE
  main_categories ALIAS FOR $1;
BEGIN
  RETURN QUERY
  WITH
    -- get all main categories and subcategories
    RECURSIVE cte_categories (category_id, main_category_id) AS
    (
      SELECT cat.category_id, cat.category_id AS main_category_id
        FROM test.categories AS cat
        WHERE cat.category_id = ANY(main_categories)
      UNION ALL
      SELECT cat.category_id, cte.main_category_id
        FROM cte_categories AS cte
        INNER JOIN test.categories AS cat
          ON cte.category_id = cat.parent_id
    ),
    -- filter main categories that are found as children of other categories
    cte_categories_unique AS
    (
      SELECT cte.*
        FROM cte_categories AS cte
        LEFT JOIN
        (
          SELECT category_id
            FROM cte_categories
            WHERE category_id <> main_category_id
            GROUP BY category_id
        ) AS to_exclude
          ON cte.main_category_id = to_exclude.category_id
        WHERE to_exclude.category_id IS NULL
    ),
    -- compute the count of main categories
    cte_main_categories_count AS
    (
      SELECT COUNT(DISTINCT main_category_id) AS cnt
        FROM cte_categories_unique
    )
  SELECT t.user_id
    FROM
      (
        -- get the users which are found in each category/sub-category then group them under the main category
        SELECT
          cu.user_id,
          cte.main_category_id
        FROM test.categories_users AS cu
        INNER JOIN cte_categories_unique AS cte
          ON cu.category_id = cte.category_id
        GROUP BY cu.user_id, cte.main_category_id
      ) AS t
      GROUP BY t.user_id
      -- filter users that do not have a match on all main categories or their sub-categories
      HAVING COUNT(*) = (SELECT cnt FROM cte_main_categories_count);
END;
$$
LANGUAGE plpgsql;


SELECT * FROM test.find_users_in_categories(ARRAY[101, 106]);

DROP SCHEMA test CASCADE;
+1


source






More articles:


All Articles