Question

General Case

How do you perform a left join across a many-to-many relationship when you want to add a condition to the foreign side of the relation?

Specific Case

We’re dealing with two tables: team and pool. There is also a team_pool table serving as a many-to-many junction table between them. Additionally, a pool has a stage_id column.

I want to retrieve all teams with that team’s pool for a specific stage. If the pool doesn’t exist, I want the pool to be NULL. Sample, idealized results:

+--------+----------+------+
| team   | stage_id | pool |
+--------+----------+------+
| Team 1 |        2 | C    |
| Team 2 |     NULL | NULL | //this team doesn't belong to a pool for this stage (but it could belong to a pool for a different stage)
| Team 3 |        2 | G    |
| Team 3 |        2 | H    | //if a team belongs to a 2 pools for the same stage
| Team 4 |        2 | D    |

If it’s relevant, I’m using MySQL.

SQL

Here’s the (simplified) SQL used to create the tables:

CREATE TABLE team (id BIGINT AUTO_INCREMENT, name VARCHAR(50), PRIMARY KEY(id));
CREATE TABLE team_pool (team_id BIGINT, pool_id BIGINT, PRIMARY KEY(team_id, pool_id));
CREATE TABLE pool (id BIGINT AUTO_INCREMENT, stage_id BIGINT, name VARCHAR(40) NOT NULL, PRIMARY KEY(id));

Ideal Solution

The ideal solution would:

• Not require a change to my schema (ORM really wants it this way)
• Require a single, non UNION query.

Attempted solutions

• Use an INNER JOIN rather than a LEFT JOIN from team to team_pool and team_pool to pool. Issue: we lose teams that don’t belong to a pool
• LEFT JOIN from team to team_pool and team_pool to pool using a WITH condition that the stage_id on pool matches what we’re looking for. Issue: when a team belongs to many pools, we get multiple results. Adding a GROUP BY doesn’t help.

EDIT: Chosen Solution

There are a lot of good solutions here.

Given that my ideal solution is not possible, I’d rather add stage_id to team_pool than use UNION or subqueries. This has the added benefit of letting me enforce that a team can only belong to one pool per stage. It also makes the queries simple:

SELECT t.name, p.name, tp.stage_id FROM team t LEFT JOIN team_pool tp ON t.id = tp.team_id AND tp.stage_id = 2 LEFT JOIN pool p ON p.id = tp.pool_id

Answer 1

If I understand the concepts behind your schema, then I would think stage_id should be a column in team_pool rather than pool. The stage is not an attribute of the pool, it is a factor in the mapping of teams to pools, right?

Regardless, this is how I would write your query in Oracle. I’m not sure if this exact syntax is right for MySQL. Presumably you would want to parameterize the literal value for stage_id.

SELECT t.name, p.name
  FROM (SELECT team.name, pool_id
          FROM team LEFT JOIN team_pool
            ON team_pool.team_id = team.team_id ) t
       LEFT JOIN (SELECT pool_id, name FROM pool WHERE stage_id = 2) p
            ON p.pool_id = t.pool_id