Select all projects that have matching tags

Question

I\'m trying to find the most efficient way of dealing with this but I must tell you front-head I\'ve made a mess of it. Looked around SO and found nothing of relevance so he

Answer 1

How about... (example for project 1)

SELECT p.num, p.title
FROM projects_to_tags pt1, projects_to_tags pt2, projects p
where pt1.project_id = 1 and 
      pt2.project_id != 1 and 
      pt1.tag_id = pt2.tag_id and 
      p.num = pt2.project_id 
group by pt2.project_id

And maybe add a separate index for tag_id in projects_to_tags so you can use it alone, instead of the composite. No more type ALL. (Table Scan) Replacing both 1 with 4 give also the desired results.

Answer 2

In any of the following cases, if you don't know the PROJECT.num/PROJECT_TO_TAGS.project_id, you'll have to join to the PROJECTS table to get the id value for finding out what tags it has associated.

Using IN

SELECT p.*
  FROM PROJECTS p
  JOIN PROJECTS_TO_TAGS pt ON pt.project_id = p.num
 WHERE pt.tag_id IN (SELECT x.tag_id
                       FROM PROJECTS_TO_TAGS x
                      WHERE x.project_id = 4)

Using EXISTS

SELECT p.*
  FROM PROJECTS p
  JOIN PROJECTS_TO_TAGS pt ON pt.project_id = p.num
 WHERE EXISTS (SELECT NULL
                 FROM PROJECTS_TO_TAGS x
                WHERE x.project_id = 4
                  AND x.tag_id = pt.tag_id)

Using JOINS (this the most efficient one!)

The DISTINCT is necessary because JOINs risk duplicated data turning up in the resultset...

SELECT DISTINCT p.*
  FROM PROJECTS p
  JOIN PROJECTS_TO_TAGS pt ON pt.project_id = p.num
  JOIN PROJECTS_TO_TAGS x ON x.tag_id = pt.tag_id
                         AND x.project_id = 4

Answer 3

Something like this... ?

SELECT *
FROM projects AS L
WHERE
   EXISTS (
      SELECT 1
      FROM
         projects_to_tags PT
         INNER JOIN projects_to_tags PT2 ON PT.tag_id = PT2.tag_id
      WHERE
         L.num = PT.project_id
         AND PT2.project_id = 4
         AND PT2.project_id <> L.num
   )

That's 2 seeks and a scan.

UPDATE

Taking a page from jdelard's book, one tiny modification switches my query to outperform his (of course I'm doing this on SQL Server meaning I took out his GROUP BY and put in a DISTINCT, so YMMV on MySQL):

SELECT *
FROM projects AS L
WHERE
   L.num != 4 -- instead of <> PT2.project_id inside
   AND EXISTS (
      SELECT 1
      FROM
         projects_to_tags PT
         INNER JOIN projects_to_tags PT2 ON PT.tag_id = PT2.tag_id
      WHERE
         L.num = PT.project_id
         AND PT2.project_id = 4
   )

The improvement over his query comes from not doing a DISTINCT or aggregate, and using a semi join instead of a complete join so not every row has to be joined. Otherwise, semantically they are largely the same.

I will have to remember jdelard's trick as it is a very useful tool. For some reason the query engine was not smart enough to compute that given {a = 4, a != b} then {b != 4}.