SQL query: Simulating an “AND” over several rows instead of sub-querying

Question

Suppose I have a \"tags\" table with two columns: tagid and contentid. Each row represents a tag assigned to a piece of content. I want a

Answer 1

Here's a solution that has worked much faster than the for me on a very large database of objects and tags. This is an example for a three-tag intersection. It just chains many joins on the object-tag table (objtags) to indicate the same object and stipulates the tag IDs in the WHERE clause:

SELECT w0.objid

FROM       objtags t0
INNER JOIN objtags t1 ON t1.objid=t0.objid
INNER JOIN objtags t2 ON t2.objid=t1.objid

WHERE t0.tagid=512
  AND t1.tagid=256
  AND t2.tagid=128

I have no idea why this runs faster. It was inspired by the search code in the MusicBrainz server. Doing this in Postgres, I usually get a ~8-10x speedup over the HAVING COUNT(...) solution.